linux-arm-kernel.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] drivers: i2c: exynos5: irq spinlock rt-safe
@ 2015-08-25  7:52 Anders Roxell
  2015-08-25 21:51 ` Thomas Gleixner
  2015-09-03 21:06 ` [PATCHv2] " Anders Roxell
  0 siblings, 2 replies; 5+ messages in thread
From: Anders Roxell @ 2015-08-25  7:52 UTC (permalink / raw)
  To: linux-arm-kernel

The exynos5_i2c_message_start enables interrupts while holding the i2c
lock which is sought by the irq handler. If an IRQ is received before
this lock is released then a deadlock occurs.

This is only seen on an RT patched kernel, due to the transformation of
spinlocks into sleeping locks. By using raw_spinlocks here the same code
can run in mainline or an RT patched kernel. No change for !RT kenrels

[   10.992238] kernel BUG at ../kernel/locking/rtmutex.c:998!
[   10.992243] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
[   10.992250] Modules linked in:
[   10.992258] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.1.5-rt5
[   10.992263] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
[   10.992268] task: ed880000 ti: ed888000 task.ti: ed888000
[   10.992281] PC is at rt_spin_lock_slowlock+0xa4/0x2ec
[   10.992288] LR is at rt_spin_lock_slowlock+0x54/0x2ec
[   10.992296] pc : [<c099e1bc>]    lr : [<c099e16c>]    psr: 60000193
[   10.992296] sp : ed889a28  ip : ed880001  fp : 00000089
[   10.992300] r10: ed889a28  r9 : c0f55654  r8 : 00000060
[   10.992305] r7 : ed880000  r6 : 00000000  r5 : 00000001  r4 : ed9f7288
[   10.992310] r3 : ed880000  r2 : 00000000  r1 : ed880000  r0 : 00000000
[   10.992316] Flags: nZCv  IRQs off  FIQs on  Mode SVC_32  ISA ARM
Segment kernel
...
...
[   10.992662] [<c099e1bc>] (rt_spin_lock_slowlock) from [<c07bc794>]
(exynos5_i2c_irq+0x20/0x2b0)
[   10.992678] [<c07bc794>] (exynos5_i2c_irq) from [<c028c8b4>]
(handle_irq_event_percpu+0x68/0x158)
[   10.992690] [<c028c8b4>] (handle_irq_event_percpu) from [<c028ca0c>]
(handle_irq_event+0x68/0xa8)
[   10.992702] [<c028ca0c>] (handle_irq_event) from [<c028f9c4>]
(handle_fasteoi_irq+0x11c/0x1d4)
[   10.992713] [<c028f9c4>] (handle_fasteoi_irq) from [<c028c17c>]
(generic_handle_irq+0x20/0x30)
[   10.992724] [<c028c17c>] (generic_handle_irq) from [<c028c290>]
(__handle_domain_irq+0x6c/0xe4)
[   10.992734] [<c028c290>] (__handle_domain_irq) from [<c020a71c>]
(gic_handle_irq+0x2c/0x68)
[   10.992744] [<c020a71c>] (gic_handle_irq) from [<c0214140>]
(__irq_svc+0x40/0x88)
[   10.992749] Exception stack(0xed889b28 to 0xed889b70)
...
...

Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
---
 drivers/i2c/busses/i2c-exynos5.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/i2c/busses/i2c-exynos5.c b/drivers/i2c/busses/i2c-exynos5.c
index b29c750..b12e77e 100644
--- a/drivers/i2c/busses/i2c-exynos5.c
+++ b/drivers/i2c/busses/i2c-exynos5.c
@@ -170,7 +170,7 @@ struct exynos5_i2c {
 	struct device		*dev;
 	int			state;
 
-	spinlock_t		lock;		/* IRQ synchronization */
+	raw_spinlock_t		lock;		/* IRQ synchronization */
 
 	/*
 	 * Since the TRANS_DONE bit is cleared on read, and we may read it
@@ -433,7 +433,7 @@ static irqreturn_t exynos5_i2c_irq(int irqno, void *dev_id)
 
 	i2c->state = -EINVAL;
 
-	spin_lock(&i2c->lock);
+	raw_spin_lock(&i2c->lock);
 
 	int_status = readl(i2c->regs + HSI2C_INT_STATUS);
 	writel(int_status, i2c->regs + HSI2C_INT_STATUS);
@@ -521,7 +521,7 @@ static irqreturn_t exynos5_i2c_irq(int irqno, void *dev_id)
 		complete(&i2c->msg_complete);
 	}
 
-	spin_unlock(&i2c->lock);
+	raw_spin_unlock(&i2c->lock);
 
 	return IRQ_HANDLED;
 }
@@ -610,7 +610,7 @@ static void exynos5_i2c_message_start(struct exynos5_i2c *i2c, int stop)
 	 * Enable interrupts before starting the transfer so that we don't
 	 * miss any INT_I2C interrupts.
 	 */
-	spin_lock_irqsave(&i2c->lock, flags);
+	raw_spin_lock_irqsave(&i2c->lock, flags);
 	writel(int_en, i2c->regs + HSI2C_INT_ENABLE);
 
 	if (stop == 1)
@@ -618,7 +618,7 @@ static void exynos5_i2c_message_start(struct exynos5_i2c *i2c, int stop)
 	i2c_auto_conf |= i2c->msg->len;
 	i2c_auto_conf |= HSI2C_MASTER_RUN;
 	writel(i2c_auto_conf, i2c->regs + HSI2C_AUTO_CONF);
-	spin_unlock_irqrestore(&i2c->lock, flags);
+	raw_spin_unlock_irqrestore(&i2c->lock, flags);
 }
 
 static int exynos5_i2c_xfer_msg(struct exynos5_i2c *i2c,
@@ -763,7 +763,7 @@ static int exynos5_i2c_probe(struct platform_device *pdev)
 	/* Clear pending interrupts from u-boot or misc causes */
 	exynos5_i2c_clr_pend_irq(i2c);
 
-	spin_lock_init(&i2c->lock);
+	raw_spin_lock_init(&i2c->lock);
 	init_completion(&i2c->msg_complete);
 
 	i2c->irq = ret = platform_get_irq(pdev, 0);
-- 
2.1.4

^ permalink raw reply related	[flat|nested] 5+ messages in thread

* [PATCH] drivers: i2c: exynos5: irq spinlock rt-safe
  2015-08-25  7:52 [PATCH] drivers: i2c: exynos5: irq spinlock rt-safe Anders Roxell
@ 2015-08-25 21:51 ` Thomas Gleixner
  2015-08-31 21:28   ` Anders Roxell
  2015-09-03 21:06 ` [PATCHv2] " Anders Roxell
  1 sibling, 1 reply; 5+ messages in thread
From: Thomas Gleixner @ 2015-08-25 21:51 UTC (permalink / raw)
  To: linux-arm-kernel

On Tue, 25 Aug 2015, Anders Roxell wrote:

> The exynos5_i2c_message_start enables interrupts while holding the i2c
> lock which is sought by the irq handler. If an IRQ is received before
> this lock is released then a deadlock occurs.

That's crap. The interrupt handler runs in an irq thread as RT forces
all normal interrupts to that.

Thanks,

	tglx

 

^ permalink raw reply	[flat|nested] 5+ messages in thread

* [PATCH] drivers: i2c: exynos5: irq spinlock rt-safe
  2015-08-25 21:51 ` Thomas Gleixner
@ 2015-08-31 21:28   ` Anders Roxell
  0 siblings, 0 replies; 5+ messages in thread
From: Anders Roxell @ 2015-08-31 21:28 UTC (permalink / raw)
  To: linux-arm-kernel

On 2015-08-25 23:51, Thomas Gleixner wrote:
> On Tue, 25 Aug 2015, Anders Roxell wrote:
> 
> > The exynos5_i2c_message_start enables interrupts while holding the i2c
> > lock which is sought by the irq handler. If an IRQ is received before
> > this lock is released then a deadlock occurs.
> 
> That's crap. The interrupt handler runs in an irq thread as RT forces
> all normal interrupts to that.

You are right, I will send a v2 with a better changelog.

Cheers,
Anders

^ permalink raw reply	[flat|nested] 5+ messages in thread

* [PATCHv2] drivers: i2c: exynos5: irq spinlock rt-safe
  2015-08-25  7:52 [PATCH] drivers: i2c: exynos5: irq spinlock rt-safe Anders Roxell
  2015-08-25 21:51 ` Thomas Gleixner
@ 2015-09-03 21:06 ` Anders Roxell
  2015-09-03 21:31   ` Thomas Gleixner
  1 sibling, 1 reply; 5+ messages in thread
From: Anders Roxell @ 2015-09-03 21:06 UTC (permalink / raw)
  To: linux-arm-kernel

The current spin_lock_irqsave() protects the enabling of the device
interrupt.  In order to prevent recursion in the case of sleeping
spinlocks (e.g. with RT patch, stacktrace below), convert the spin lock
to a raw spin lock. No change for !RT kernels.

[   10.992238] kernel BUG at ../kernel/locking/rtmutex.c:998!
[   10.992243] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
[   10.992250] Modules linked in:
[   10.992258] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.1.5-rt5
[   10.992263] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
[   10.992268] task: ed880000 ti: ed888000 task.ti: ed888000
[   10.992281] PC is at rt_spin_lock_slowlock+0xa4/0x2ec
[   10.992288] LR is at rt_spin_lock_slowlock+0x54/0x2ec
[   10.992296] pc : [<c099e1bc>]    lr : [<c099e16c>]    psr: 60000193
[   10.992296] sp : ed889a28  ip : ed880001  fp : 00000089
[   10.992300] r10: ed889a28  r9 : c0f55654  r8 : 00000060
[   10.992305] r7 : ed880000  r6 : 00000000  r5 : 00000001  r4 : ed9f7288
[   10.992310] r3 : ed880000  r2 : 00000000  r1 : ed880000  r0 : 00000000
[   10.992316] Flags: nZCv  IRQs off  FIQs on  Mode SVC_32  ISA ARM
Segment kernel
...
...
[   10.992662] [<c099e1bc>] (rt_spin_lock_slowlock) from [<c07bc794>]
(exynos5_i2c_irq+0x20/0x2b0)
[   10.992678] [<c07bc794>] (exynos5_i2c_irq) from [<c028c8b4>]
(handle_irq_event_percpu+0x68/0x158)
[   10.992690] [<c028c8b4>] (handle_irq_event_percpu) from [<c028ca0c>]
(handle_irq_event+0x68/0xa8)
[   10.992702] [<c028ca0c>] (handle_irq_event) from [<c028f9c4>]
(handle_fasteoi_irq+0x11c/0x1d4)
[   10.992713] [<c028f9c4>] (handle_fasteoi_irq) from [<c028c17c>]
(generic_handle_irq+0x20/0x30)
[   10.992724] [<c028c17c>] (generic_handle_irq) from [<c028c290>]
(__handle_domain_irq+0x6c/0xe4)
[   10.992734] [<c028c290>] (__handle_domain_irq) from [<c020a71c>]
(gic_handle_irq+0x2c/0x68)
[   10.992744] [<c020a71c>] (gic_handle_irq) from [<c0214140>]
(__irq_svc+0x40/0x88)
[   10.992749] Exception stack(0xed889b28 to 0xed889b70)
...
...

Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
---
 drivers/i2c/busses/i2c-exynos5.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/i2c/busses/i2c-exynos5.c b/drivers/i2c/busses/i2c-exynos5.c
index b29c750..b12e77e 100644
--- a/drivers/i2c/busses/i2c-exynos5.c
+++ b/drivers/i2c/busses/i2c-exynos5.c
@@ -170,7 +170,7 @@ struct exynos5_i2c {
 	struct device		*dev;
 	int			state;
 
-	spinlock_t		lock;		/* IRQ synchronization */
+	raw_spinlock_t		lock;		/* IRQ synchronization */
 
 	/*
 	 * Since the TRANS_DONE bit is cleared on read, and we may read it
@@ -433,7 +433,7 @@ static irqreturn_t exynos5_i2c_irq(int irqno, void *dev_id)
 
 	i2c->state = -EINVAL;
 
-	spin_lock(&i2c->lock);
+	raw_spin_lock(&i2c->lock);
 
 	int_status = readl(i2c->regs + HSI2C_INT_STATUS);
 	writel(int_status, i2c->regs + HSI2C_INT_STATUS);
@@ -521,7 +521,7 @@ static irqreturn_t exynos5_i2c_irq(int irqno, void *dev_id)
 		complete(&i2c->msg_complete);
 	}
 
-	spin_unlock(&i2c->lock);
+	raw_spin_unlock(&i2c->lock);
 
 	return IRQ_HANDLED;
 }
@@ -610,7 +610,7 @@ static void exynos5_i2c_message_start(struct exynos5_i2c *i2c, int stop)
 	 * Enable interrupts before starting the transfer so that we don't
 	 * miss any INT_I2C interrupts.
 	 */
-	spin_lock_irqsave(&i2c->lock, flags);
+	raw_spin_lock_irqsave(&i2c->lock, flags);
 	writel(int_en, i2c->regs + HSI2C_INT_ENABLE);
 
 	if (stop == 1)
@@ -618,7 +618,7 @@ static void exynos5_i2c_message_start(struct exynos5_i2c *i2c, int stop)
 	i2c_auto_conf |= i2c->msg->len;
 	i2c_auto_conf |= HSI2C_MASTER_RUN;
 	writel(i2c_auto_conf, i2c->regs + HSI2C_AUTO_CONF);
-	spin_unlock_irqrestore(&i2c->lock, flags);
+	raw_spin_unlock_irqrestore(&i2c->lock, flags);
 }
 
 static int exynos5_i2c_xfer_msg(struct exynos5_i2c *i2c,
@@ -763,7 +763,7 @@ static int exynos5_i2c_probe(struct platform_device *pdev)
 	/* Clear pending interrupts from u-boot or misc causes */
 	exynos5_i2c_clr_pend_irq(i2c);
 
-	spin_lock_init(&i2c->lock);
+	raw_spin_lock_init(&i2c->lock);
 	init_completion(&i2c->msg_complete);
 
 	i2c->irq = ret = platform_get_irq(pdev, 0);
-- 
2.1.4

^ permalink raw reply related	[flat|nested] 5+ messages in thread

* [PATCHv2] drivers: i2c: exynos5: irq spinlock rt-safe
  2015-09-03 21:06 ` [PATCHv2] " Anders Roxell
@ 2015-09-03 21:31   ` Thomas Gleixner
  0 siblings, 0 replies; 5+ messages in thread
From: Thomas Gleixner @ 2015-09-03 21:31 UTC (permalink / raw)
  To: linux-arm-kernel

On Thu, 3 Sep 2015, Anders Roxell wrote:

> The current spin_lock_irqsave() protects the enabling of the device
> interrupt.  In order to prevent recursion in the case of sleeping
> spinlocks (e.g. with RT patch, stacktrace below), convert the spin lock
> to a raw spin lock. No change for !RT kernels.
> 
> [   10.992238] kernel BUG at ../kernel/locking/rtmutex.c:998!
> [   10.992243] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
> [   10.992250] Modules linked in:
> [   10.992258] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.1.5-rt5
> [   10.992263] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
> [   10.992268] task: ed880000 ti: ed888000 task.ti: ed888000
> [   10.992281] PC is at rt_spin_lock_slowlock+0xa4/0x2ec
> [   10.992288] LR is at rt_spin_lock_slowlock+0x54/0x2ec
> [   10.992296] pc : [<c099e1bc>]    lr : [<c099e16c>]    psr: 60000193
> [   10.992296] sp : ed889a28  ip : ed880001  fp : 00000089
> [   10.992300] r10: ed889a28  r9 : c0f55654  r8 : 00000060
> [   10.992305] r7 : ed880000  r6 : 00000000  r5 : 00000001  r4 : ed9f7288
> [   10.992310] r3 : ed880000  r2 : 00000000  r1 : ed880000  r0 : 00000000
> [   10.992316] Flags: nZCv  IRQs off  FIQs on  Mode SVC_32  ISA ARM
> Segment kernel
> ...
> ...
> [   10.992662] [<c099e1bc>] (rt_spin_lock_slowlock) from [<c07bc794>]
> (exynos5_i2c_irq+0x20/0x2b0)
> [   10.992678] [<c07bc794>] (exynos5_i2c_irq) from [<c028c8b4>]
> (handle_irq_event_percpu+0x68/0x158)
> [   10.992690] [<c028c8b4>] (handle_irq_event_percpu) from [<c028ca0c>]
> (handle_irq_event+0x68/0xa8)
> [   10.992702] [<c028ca0c>] (handle_irq_event) from [<c028f9c4>]
> (handle_fasteoi_irq+0x11c/0x1d4)
> [   10.992713] [<c028f9c4>] (handle_fasteoi_irq) from [<c028c17c>]
> (generic_handle_irq+0x20/0x30)
> [   10.992724] [<c028c17c>] (generic_handle_irq) from [<c028c290>]
> (__handle_domain_irq+0x6c/0xe4)
> [   10.992734] [<c028c290>] (__handle_domain_irq) from [<c020a71c>]
> (gic_handle_irq+0x2c/0x68)
> [   10.992744] [<c020a71c>] (gic_handle_irq) from [<c0214140>]
> (__irq_svc+0x40/0x88)
> [   10.992749] Exception stack(0xed889b28 to 0xed889b70)

What you completely fail to explain how this can happen at all.

On RT the interrupt of that device is force threaded, but the handler
is called in hard interrupt context, which is wrong to begin with.

While the interrupt request in this driver is bogus:

        ret = devm_request_irq(&pdev->dev, i2c->irq, exynos5_i2c_irq,
                                IRQF_NO_SUSPEND | IRQF_ONESHOT,
                                dev_name(&pdev->dev), i2c);

because it sets the ONESHOT flag for no reason, it does not set the
NOTRHEAD flag which would prevent force threading.

Hmm?

	tglx

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2015-09-03 21:31 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2015-08-25  7:52 [PATCH] drivers: i2c: exynos5: irq spinlock rt-safe Anders Roxell
2015-08-25 21:51 ` Thomas Gleixner
2015-08-31 21:28   ` Anders Roxell
2015-09-03 21:06 ` [PATCHv2] " Anders Roxell
2015-09-03 21:31   ` Thomas Gleixner

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).