This patch is a modification of the patch originally proposed
by Xiaotian Feng <[email protected]>: https://lkml.org/lkml/2012/11/5/413
This new version disables DMA channel interrupts and ensures that the tasklet
wil not be scheduled again before calling tasklet_kill().
Unfortunately the updated patch was not released at that time due to planned
rework of Tsi721 mport driver to use threaded interrupts (which has yet to
happen).
Recently the issue was reported again: https://lkml.org/lkml/2014/2/19/762.
Description from the original Xiaotian's patch:
"Some drivers use tasklet_disable in device remove/release process,
tasklet_disable will inc tasklet->count and return. If the tasklet is
not handled yet under some softirq pressure, the tasklet will be
placed on the tasklet_vec, never have a chance to be excuted. This
might lead to a heavy loaded ksoftirqd, wakeup with pending_softirq,
but tasklet is disabled. tasklet_kill should be used in this case."
This patch is applicable to kernel versions starting from v3.5.
Signed-off-by: Alexandre Bounine <[email protected]>
Cc: Matt Porter <[email protected]>
Cc: Xiaotian Feng <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: <[email protected]>
---
drivers/rapidio/devices/tsi721.h | 1 +
drivers/rapidio/devices/tsi721_dma.c | 25 ++++++++++++++++++-------
2 files changed, 19 insertions(+), 7 deletions(-)
diff --git a/drivers/rapidio/devices/tsi721.h b/drivers/rapidio/devices/tsi721.h
index b4b0d83..7061ac0 100644
--- a/drivers/rapidio/devices/tsi721.h
+++ b/drivers/rapidio/devices/tsi721.h
@@ -678,6 +678,7 @@ struct tsi721_bdma_chan {
struct list_head free_list;
dma_cookie_t completed_cookie;
struct tasklet_struct tasklet;
+ bool active;
};
#endif /* CONFIG_RAPIDIO_DMA_ENGINE */
diff --git a/drivers/rapidio/devices/tsi721_dma.c b/drivers/rapidio/devices/tsi721_dma.c
index 502663f..b57f650 100644
--- a/drivers/rapidio/devices/tsi721_dma.c
+++ b/drivers/rapidio/devices/tsi721_dma.c
@@ -206,8 +206,8 @@ void tsi721_bdma_handler(struct tsi721_bdma_chan *bdma_chan)
{
/* Disable BDMA channel interrupts */
iowrite32(0, bdma_chan->regs + TSI721_DMAC_INTE);
-
- tasklet_schedule(&bdma_chan->tasklet);
+ if (bdma_chan->active)
+ tasklet_schedule(&bdma_chan->tasklet);
}
#ifdef CONFIG_PCI_MSI
@@ -562,7 +562,7 @@ static int tsi721_alloc_chan_resources(struct dma_chan *dchan)
}
#endif /* CONFIG_PCI_MSI */
- tasklet_enable(&bdma_chan->tasklet);
+ bdma_chan->active = true;
tsi721_bdma_interrupt_enable(bdma_chan, 1);
return bdma_chan->bd_num - 1;
@@ -589,14 +589,25 @@ static void tsi721_free_chan_resources(struct dma_chan *dchan)
BUG_ON(!list_empty(&bdma_chan->active_list));
BUG_ON(!list_empty(&bdma_chan->queue));
- tasklet_disable(&bdma_chan->tasklet);
+ tsi721_bdma_interrupt_enable(bdma_chan, 0);
+ bdma_chan->active = false;
+
+#ifdef CONFIG_PCI_MSI
+ if (priv->flags & TSI721_USING_MSIX) {
+ synchronize_irq(priv->msix[TSI721_VECT_DMA0_DONE +
+ bdma_chan->id].vector);
+ synchronize_irq(priv->msix[TSI721_VECT_DMA0_INT +
+ bdma_chan->id].vector);
+ } else
+#endif
+ synchronize_irq(priv->pdev->irq);
+
+ tasklet_kill(&bdma_chan->tasklet);
spin_lock_bh(&bdma_chan->lock);
list_splice_init(&bdma_chan->free_list, &list);
spin_unlock_bh(&bdma_chan->lock);
- tsi721_bdma_interrupt_enable(bdma_chan, 0);
-
#ifdef CONFIG_PCI_MSI
if (priv->flags & TSI721_USING_MSIX) {
free_irq(priv->msix[TSI721_VECT_DMA0_DONE +
@@ -790,6 +801,7 @@ int tsi721_register_dma(struct tsi721_device *priv)
bdma_chan->dchan.cookie = 1;
bdma_chan->dchan.chan_id = i;
bdma_chan->id = i;
+ bdma_chan->active = false;
spin_lock_init(&bdma_chan->lock);
@@ -799,7 +811,6 @@ int tsi721_register_dma(struct tsi721_device *priv)
tasklet_init(&bdma_chan->tasklet, tsi721_dma_tasklet,
(unsigned long)bdma_chan);
- tasklet_disable(&bdma_chan->tasklet);
list_add_tail(&bdma_chan->dchan.device_node,
&mport->dma.channels);
}
--
1.7.8.4
On Fri, 21 Feb 2014, Alexandre Bounine wrote:
> This patch is a modification of the patch originally proposed
> by Xiaotian Feng <[email protected]>: https://lkml.org/lkml/2012/11/5/413
> This new version disables DMA channel interrupts and ensures that the tasklet
> wil not be scheduled again before calling tasklet_kill().
>
> Unfortunately the updated patch was not released at that time due to planned
> rework of Tsi721 mport driver to use threaded interrupts (which has yet to
> happen).
> Recently the issue was reported again: https://lkml.org/lkml/2014/2/19/762.
>
> Description from the original Xiaotian's patch:
>
> "Some drivers use tasklet_disable in device remove/release process,
> tasklet_disable will inc tasklet->count and return. If the tasklet is
> not handled yet under some softirq pressure, the tasklet will be
> placed on the tasklet_vec, never have a chance to be excuted. This
> might lead to a heavy loaded ksoftirqd, wakeup with pending_softirq,
> but tasklet is disabled. tasklet_kill should be used in this case."
>
> This patch is applicable to kernel versions starting from v3.5.
>
> Signed-off-by: Alexandre Bounine <[email protected]>
> Cc: Matt Porter <[email protected]>
> Cc: Xiaotian Feng <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Mike Galbraith <[email protected]>
> Cc: <[email protected]>
Reviewed-by: Thomas Gleixner <[email protected]>