Hi Linus,
Just the single fix below. Bug has been present for a while, but is
triggering with regularity since commit 77873803363c "net_dma: mark
broken" in 3.13-rc5. Full patch included below.
Please apply, thank you.
The following changes since commit 6d0abeca3242a88cab8232e4acd7e2bf088f3bc2:
Linux 3.14-rc3 (2014-02-16 13:30:25 -0800)
are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/djbw/dmaengine tags/dmaengine-fixes-3.14-rc4
for you to fetch changes up to da87ca4d4ca101f177fffd84f1f0a5e4c0343557:
ioat: fix tasklet tear down (2014-02-25 09:44:20 -0800)
----------------------------------------------------------------
dmaengine-fixes-3.14-rc4
Fix tasklet lifetime management in the ioat driver causing ksoftirqd to
spin indefinitely.
References:
https://lkml.org/lkml/2014/1/27/282
https://lkml.org/lkml/2014/2/19/672
----------------------------------------------------------------
Dan Williams (1):
ioat: fix tasklet tear down
drivers/dma/ioat/dma.c | 52 ++++++++++++++++++++++++++++++++++++++++-------
drivers/dma/ioat/dma.h | 1 +
drivers/dma/ioat/dma_v2.c | 11 +++++-----
drivers/dma/ioat/dma_v3.c | 3 +++
4 files changed, 54 insertions(+), 13 deletions(-)
8<---------
>From da87ca4d4ca101f177fffd84f1f0a5e4c0343557 Mon Sep 17 00:00:00 2001
From: Dan Williams <dan.j.williams@xxxxxxxxx>
Date: Wed, 19 Feb 2014 16:19:35 -0800
Subject: [PATCH] ioat: fix tasklet tear down
Since commit 77873803363c "net_dma: mark broken" we no longer pin dma
engines active for the network-receive-offload use case. As a result
the ->free_chan_resources() that occurs after the driver self test no
longer has a NET_DMA induced ->alloc_chan_resources() to back it up. A
late firing irq can lead to ksoftirqd spinning indefinitely due to the
tasklet_disable() performed by ->free_chan_resources(). Only
->alloc_chan_resources() can clear this condition in affected kernels.
This problem has been present since commit 3e037454bcfa "I/OAT: Add
support for MSI and MSI-X" in 2.6.24, but is now exposed. Given the
NET_DMA use case is deprecated we can revisit moving the driver to use
threaded irqs. For now, just tear down the irq and tasklet properly by:
1/ Disable the irq from triggering the tasklet
2/ Disable the irq from re-arming
3/ Flush inflight interrupts
4/ Flush the timer
5/ Flush inflight tasklets
References:
https://lkml.org/lkml/2014/1/27/282
https://lkml.org/lkml/2014/2/19/672
Cc: Ingo Molnar <mingo@xxxxxxx>
Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx>
Reported-by: Mike Galbraith <bitbucket@xxxxxxxxx>
Reported-by: Stanislav Fomichev <stfomichev@xxxxxxxxxxxxxx>
Tested-by: Mike Galbraith <bitbucket@xxxxxxxxx>
Tested-by: Stanislav Fomichev <stfomichev@xxxxxxxxxxxxxx>
Reviewed-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
drivers/dma/ioat/dma.c | 52 ++++++++++++++++++++++++++++++++++++++++-------
drivers/dma/ioat/dma.h | 1 +
drivers/dma/ioat/dma_v2.c | 11 +++++-----
drivers/dma/ioat/dma_v3.c | 3 +++
4 files changed, 54 insertions(+), 13 deletions(-)
diff --git a/drivers/dma/ioat/dma.c b/drivers/dma/ioat/dma.c
index 8752918..4e3549a 100644
--- a/drivers/dma/ioat/dma.c
+++ b/drivers/dma/ioat/dma.c
@@ -77,7 +77,8 @@ static irqreturn_t ioat_dma_do_interrupt(int irq, void *data)
attnstatus = readl(instance->reg_base + IOAT_ATTNSTATUS_OFFSET);
for_each_set_bit(bit, &attnstatus, BITS_PER_LONG) {
chan = ioat_chan_by_index(instance, bit);
- tasklet_schedule(&chan->cleanup_task);
+ if (test_bit(IOAT_RUN, &chan->state))
+ tasklet_schedule(&chan->cleanup_task);
}
writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
@@ -93,7 +94,8 @@ static irqreturn_t ioat_dma_do_interrupt_msix(int irq, void *data)
{
struct ioat_chan_common *chan = data;
- tasklet_schedule(&chan->cleanup_task);
+ if (test_bit(IOAT_RUN, &chan->state))
+ tasklet_schedule(&chan->cleanup_task);
return IRQ_HANDLED;
}
@@ -116,7 +118,6 @@ void ioat_init_channel(struct ioatdma_device *device, struct ioat_chan_common *c
chan->timer.function = device->timer_fn;
chan->timer.data = data;
tasklet_init(&chan->cleanup_task, device->cleanup_fn, data);
- tasklet_disable(&chan->cleanup_task);
}
/**
@@ -354,13 +355,49 @@ static int ioat1_dma_alloc_chan_resources(struct dma_chan *c)
writel(((u64) chan->completion_dma) >> 32,
chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH);
- tasklet_enable(&chan->cleanup_task);
+ set_bit(IOAT_RUN, &chan->state);
ioat1_dma_start_null_desc(ioat); /* give chain to dma device */
dev_dbg(to_dev(chan), "%s: allocated %d descriptors\n",
__func__, ioat->desccount);
return ioat->desccount;
}
+void ioat_stop(struct ioat_chan_common *chan)
+{
+ struct ioatdma_device *device = chan->device;
+ struct pci_dev *pdev = device->pdev;
+ int chan_id = chan_num(chan);
+ struct msix_entry *msix;
+
+ /* 1/ stop irq from firing tasklets
+ * 2/ stop the tasklet from re-arming irqs
+ */
+ clear_bit(IOAT_RUN, &chan->state);
+
+ /* flush inflight interrupts */
+ switch (device->irq_mode) {
+ case IOAT_MSIX:
+ msix = &device->msix_entries[chan_id];
+ synchronize_irq(msix->vector);
+ break;
+ case IOAT_MSI:
+ case IOAT_INTX:
+ synchronize_irq(pdev->irq);
+ break;
+ default:
+ break;
+ }
+
+ /* flush inflight timers */
+ del_timer_sync(&chan->timer);
+
+ /* flush inflight tasklet runs */
+ tasklet_kill(&chan->cleanup_task);
+
+ /* final cleanup now that everything is quiesced and can't re-arm */
+ device->cleanup_fn((unsigned long) &chan->common);
+}
+
/**
* ioat1_dma_free_chan_resources - release all the descriptors
* @chan: the channel to be cleaned
@@ -379,9 +416,7 @@ static void ioat1_dma_free_chan_resources(struct dma_chan *c)
if (ioat->desccount == 0)
return;
- tasklet_disable(&chan->cleanup_task);
- del_timer_sync(&chan->timer);
- ioat1_cleanup(ioat);
+ ioat_stop(chan);
/* Delay 100ms after reset to allow internal DMA logic to quiesce
* before removing DMA descriptor resources.
@@ -526,8 +561,11 @@ ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest,
static void ioat1_cleanup_event(unsigned long data)
{
struct ioat_dma_chan *ioat = to_ioat_chan((void *) data);
+ struct ioat_chan_common *chan = &ioat->base;
ioat1_cleanup(ioat);
+ if (!test_bit(IOAT_RUN, &chan->state))
+ return;
writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
}
diff --git a/drivers/dma/ioat/dma.h b/drivers/dma/ioat/dma.h
index 11fb877..e982f00 100644
--- a/drivers/dma/ioat/dma.h
+++ b/drivers/dma/ioat/dma.h
@@ -356,6 +356,7 @@ bool ioat_cleanup_preamble(struct ioat_chan_common *chan,
void ioat_kobject_add(struct ioatdma_device *device, struct kobj_type *type);
void ioat_kobject_del(struct ioatdma_device *device);
int ioat_dma_setup_interrupts(struct ioatdma_device *device);
+void ioat_stop(struct ioat_chan_common *chan);
extern const struct sysfs_ops ioat_sysfs_ops;
extern struct ioat_sysfs_entry ioat_version_attr;
extern struct ioat_sysfs_entry ioat_cap_attr;
diff --git a/drivers/dma/ioat/dma_v2.c b/drivers/dma/ioat/dma_v2.c
index 5d3affe..8d10580 100644
--- a/drivers/dma/ioat/dma_v2.c
+++ b/drivers/dma/ioat/dma_v2.c
@@ -190,8 +190,11 @@ static void ioat2_cleanup(struct ioat2_dma_chan *ioat)
void ioat2_cleanup_event(unsigned long data)
{
struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
+ struct ioat_chan_common *chan = &ioat->base;
ioat2_cleanup(ioat);
+ if (!test_bit(IOAT_RUN, &chan->state))
+ return;
writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
}
@@ -553,10 +556,10 @@ int ioat2_alloc_chan_resources(struct dma_chan *c)
ioat->issued = 0;
ioat->tail = 0;
ioat->alloc_order = order;
+ set_bit(IOAT_RUN, &chan->state);
spin_unlock_bh(&ioat->prep_lock);
spin_unlock_bh(&chan->cleanup_lock);
- tasklet_enable(&chan->cleanup_task);
ioat2_start_null_desc(ioat);
/* check that we got off the ground */
@@ -566,7 +569,6 @@ int ioat2_alloc_chan_resources(struct dma_chan *c)
} while (i++ < 20 && !is_ioat_active(status) && !is_ioat_idle(status));
if (is_ioat_active(status) || is_ioat_idle(status)) {
- set_bit(IOAT_RUN, &chan->state);
return 1 << ioat->alloc_order;
} else {
u32 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
@@ -809,11 +811,8 @@ void ioat2_free_chan_resources(struct dma_chan *c)
if (!ioat->ring)
return;
- tasklet_disable(&chan->cleanup_task);
- del_timer_sync(&chan->timer);
- device->cleanup_fn((unsigned long) c);
+ ioat_stop(chan);
device->reset_hw(chan);
- clear_bit(IOAT_RUN, &chan->state);
spin_lock_bh(&chan->cleanup_lock);
spin_lock_bh(&ioat->prep_lock);
diff --git a/drivers/dma/ioat/dma_v3.c b/drivers/dma/ioat/dma_v3.c
index 820817e9..b9b38a1 100644
--- a/drivers/dma/ioat/dma_v3.c
+++ b/drivers/dma/ioat/dma_v3.c
@@ -464,8 +464,11 @@ static void ioat3_cleanup(struct ioat2_dma_chan *ioat)
static void ioat3_cleanup_event(unsigned long data)
{
struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
+ struct ioat_chan_common *chan = &ioat->base;
ioat3_cleanup(ioat);
+ if (!test_bit(IOAT_RUN, &chan->state))
+ return;
writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
}
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
