Skip to contentSkip to conversation

sleeping while atomic in dwc3_gadget_start

DISCUSSION Stephen Boyd 2 participants 6 messages Jun 26, 2013 – Jul 1, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi,

I'm getting the folllowing BUG message on bootup with 3.10-rc5

BUG: sleeping function called from invalid context at mm/slub.c:926
in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc5-gee3e35b-09316-ge78f3b35 #643
[<c0014220>] (unwind_backtrace+0x0/0x120) from [<c001212c>] (show_stack+0x10/0x14)
[<c001212c>] (show_stack+0x10/0x14) from [<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210)
[<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210) from [<c00e0c60>] (request_threaded_irq+0x88/0x11c)
[<c00e0c60>] (request_threaded_irq+0x88/0x11c) from [<c03bf53c>] (dwc3_gadget_start+0x198/0x200)
[<c03bf53c>] (dwc3_gadget_start+0x198/0x200) from [<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8)
[<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8) from [<c03f7b50>] (usb_gadget_probe_driver+0x8c/0xb8)

and I suspect this problem was introduced in commit 8698e2acf (usb: dwc3: gadget: introduce and use enable/disable irq methods). Is there a fix for this problem? Can we just move the irq request outside the spinlock?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
#1 lore raw
Quoted 14 lines · Stephen Boyd

Hi,

I'm getting the folllowing BUG message on bootup with 3.10-rc5

BUG: sleeping function called from invalid context at mm/slub.c:926
in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc5-gee3e35b-09316-ge78f3b35 #643
[<c0014220>] (unwind_backtrace+0x0/0x120) from [<c001212c>] (show_stack+0x10/0x14)
[<c001212c>] (show_stack+0x10/0x14) from [<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210)
[<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210) from [<c00e0c60>] (request_threaded_irq+0x88/0x11c)
[<c00e0c60>] (request_threaded_irq+0x88/0x11c) from [<c03bf53c>] (dwc3_gadget_start+0x198/0x200)
[<c03bf53c>] (dwc3_gadget_start+0x198/0x200) from [<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8)
[<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8) from [<c03f7b50>] (usb_gadget_probe_driver+0x8c/0xb8)
Quoted 4 lines

and I suspect this problem was introduced in commit 8698e2acf (usb: dwc3: gadget: introduce and use enable/disable irq methods). Is there a fix for this problem? Can we just move the irq request outside the spinlock?

nice :-)

how about this ?

drivers/usb/dwc3/gadget.c +15 -14
index b5e5b35..a8e6b8b 100644
@@ -1508,6 +1508,15 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 	int			irq;
 	u32			reg;
 
+	irq = platform_get_irq(to_platform_device(dwc->dev), 0);
+	ret = request_threaded_irq(irq, dwc3_interrupt, dwc3_thread_interrupt,
+			IRQF_SHARED | IRQF_ONESHOT, "dwc3", dwc);
+	if (ret) {
+		dev_err(dwc->dev, "failed to request irq #%d --> %d\n",
+				irq, ret);
+		goto err0;
+	}
+
 	spin_lock_irqsave(&dwc->lock, flags);
 
 	if (dwc->gadget_driver) {
@@ -1515,7 +1524,7 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 				dwc->gadget.name,
 				dwc->gadget_driver->driver.name);
 		ret = -EBUSY;
-		goto err0;
+		goto err1;
 	}
 
 	dwc->gadget_driver	= driver;
@@ -1551,41 +1560,33 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, false);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
-		goto err0;
+		goto err1;
 	}
 
 	dep = dwc->eps[1];
 	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, false);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
-		goto err1;
+		goto err2;
 	}
 
 	/* begin to receive SETUP packets */
 	dwc->ep0state = EP0_SETUP_PHASE;
 	dwc3_ep0_out_start(dwc);
 
-	irq = platform_get_irq(to_platform_device(dwc->dev), 0);
-	ret = request_threaded_irq(irq, dwc3_interrupt, dwc3_thread_interrupt,
-			IRQF_SHARED | IRQF_ONESHOT, "dwc3", dwc);
-	if (ret) {
-		dev_err(dwc->dev, "failed to request irq #%d --> %d\n",
-				irq, ret);
-		goto err1;
-	}
-
 	dwc3_gadget_enable_irq(dwc);
 
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
 	return 0;
 
-err1:
+err2:
 	__dwc3_gadget_ep_disable(dwc->eps[0]);
 
-err0:
+err1:
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
+err0:
 	return ret;
 }
 
Signature · 1 line
balbi
#2 lore raw
Quoted 15 lines · Felipe Balbi
Quoted 14 lines · Stephen Boyd

Hi,

I'm getting the folllowing BUG message on bootup with 3.10-rc5

BUG: sleeping function called from invalid context at mm/slub.c:926
in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc5-gee3e35b-09316-ge78f3b35 #643
[<c0014220>] (unwind_backtrace+0x0/0x120) from [<c001212c>] (show_stack+0x10/0x14)
[<c001212c>] (show_stack+0x10/0x14) from [<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210)
[<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210) from [<c00e0c60>] (request_threaded_irq+0x88/0x11c)
[<c00e0c60>] (request_threaded_irq+0x88/0x11c) from [<c03bf53c>] (dwc3_gadget_start+0x198/0x200)
[<c03bf53c>] (dwc3_gadget_start+0x198/0x200) from [<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8)
[<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8) from [<c03f7b50>] (usb_gadget_probe_driver+0x8c/0xb8)
Quoted 7 lines
Quoted 4 lines

and I suspect this problem was introduced in commit 8698e2acf (usb: dwc3: gadget: introduce and use enable/disable irq methods). Is there a fix for this problem? Can we just move the irq request outside the spinlock?

nice :-)

how about this ?

If start fails do you call stop? I believe the answer is no, so we'll need to free_irq() somewhere along the error path. Or we can request it after the spin_unlock()?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
#3 lore raw

Hi,

Quoted 20 lines · Stephen Boyd
Quoted 19 lines · Felipe Balbi
Quoted 18 lines · Stephen Boyd

Hi,

I'm getting the folllowing BUG message on bootup with 3.10-rc5

BUG: sleeping function called from invalid context at mm/slub.c:926
in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc5-gee3e35b-09316-ge78f3b35 #643
[<c0014220>] (unwind_backtrace+0x0/0x120) from [<c001212c>] (show_stack+0x10/0x14)
[<c001212c>] (show_stack+0x10/0x14) from [<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210)
[<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210) from [<c00e0c60>] (request_threaded_irq+0x88/0x11c)
[<c00e0c60>] (request_threaded_irq+0x88/0x11c) from [<c03bf53c>] (dwc3_gadget_start+0x198/0x200)
[<c03bf53c>] (dwc3_gadget_start+0x198/0x200) from [<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8)
[<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8) from [<c03f7b50>] (usb_gadget_probe_driver+0x8c/0xb8)

and I suspect this problem was introduced in commit 8698e2acf (usb: dwc3: gadget: introduce and use enable/disable irq methods). Is there a fix for this problem? Can we just move the irq request outside the spinlock?

Quoted 7 lines
Quoted 3 lines

nice :-)

how about this ?

If start fails do you call stop? I believe the answer is no, so we'll need to free_irq() somewhere along the error path. Or we can request it after the spin_unlock()?

good point here's v2:

From 1a7896c5481646e533ca09ba60918831d9849ed8 Mon Sep 17 00:00:00 2001
From: Felipe Balbi <[email protected]>
Date: Thu, 27 Jun 2013 10:00:18 +0300
Subject: [PATCH] usb: dwc3: gadget: don't request IRQs in atomic

We cannot request an IRQ with spinlocks held
as that would trigger a sleeping inside
spinlock warning.

---
 drivers/usb/dwc3/gadget.c | 37 ++++++++++++++++++++-----------------
 1 file changed, 20 insertions(+), 17 deletions(-)
drivers/usb/dwc3/gadget.c +20 -17
index b5e5b35..bb0d312c 100644
@@ -1508,6 +1508,15 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 	int			irq;
 	u32			reg;
 
+	irq = platform_get_irq(to_platform_device(dwc->dev), 0);
+	ret = request_threaded_irq(irq, dwc3_interrupt, dwc3_thread_interrupt,
+			IRQF_SHARED | IRQF_ONESHOT, "dwc3", dwc);
+	if (ret) {
+		dev_err(dwc->dev, "failed to request irq #%d --> %d\n",
+				irq, ret);
+		goto err0;
+	}
+
 	spin_lock_irqsave(&dwc->lock, flags);
 
 	if (dwc->gadget_driver) {
@@ -1515,7 +1524,7 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 				dwc->gadget.name,
 				dwc->gadget_driver->driver.name);
 		ret = -EBUSY;
-		goto err0;
+		goto err1;
 	}
 
 	dwc->gadget_driver	= driver;
@@ -1551,41 +1560,35 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, false);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
-		goto err0;
+		goto err1;
 	}
 
 	dep = dwc->eps[1];
 	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, false);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
-		goto err1;
+		goto err2;
 	}
 
 	/* begin to receive SETUP packets */
 	dwc->ep0state = EP0_SETUP_PHASE;
 	dwc3_ep0_out_start(dwc);
 
-	irq = platform_get_irq(to_platform_device(dwc->dev), 0);
-	ret = request_threaded_irq(irq, dwc3_interrupt, dwc3_thread_interrupt,
-			IRQF_SHARED | IRQF_ONESHOT, "dwc3", dwc);
-	if (ret) {
-		dev_err(dwc->dev, "failed to request irq #%d --> %d\n",
-				irq, ret);
-		goto err1;
-	}
-
 	dwc3_gadget_enable_irq(dwc);
 
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
 	return 0;
 
-err1:
+err2:
 	__dwc3_gadget_ep_disable(dwc->eps[0]);
 
-err0:
+err1:
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
+err0:
+	free_irq(irq, dwc);
+
 	return ret;
 }
 
@@ -1599,9 +1602,6 @@ static int dwc3_gadget_stop(struct usb_gadget *g,
 	spin_lock_irqsave(&dwc->lock, flags);
 
 	dwc3_gadget_disable_irq(dwc);
-	irq = platform_get_irq(to_platform_device(dwc->dev), 0);
-	free_irq(irq, dwc);
-
 	__dwc3_gadget_ep_disable(dwc->eps[0]);
 	__dwc3_gadget_ep_disable(dwc->eps[1]);
 
@@ -1609,6 +1609,9 @@ static int dwc3_gadget_stop(struct usb_gadget *g,
 
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
+	irq = platform_get_irq(to_platform_device(dwc->dev), 0);
+	free_irq(irq, dwc);
+
 	return 0;
 }
 
Signature · 5 lines
1.8.2.1


-- 
balbi
#4 lore raw
Quoted 23 lines · Felipe Balbi

Hi,

Quoted 20 lines · Stephen Boyd
Quoted 19 lines · Felipe Balbi
Quoted 18 lines · Stephen Boyd

Hi,

I'm getting the folllowing BUG message on bootup with 3.10-rc5

BUG: sleeping function called from invalid context at mm/slub.c:926
in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc5-gee3e35b-09316-ge78f3b35 #643
[<c0014220>] (unwind_backtrace+0x0/0x120) from [<c001212c>] (show_stack+0x10/0x14)
[<c001212c>] (show_stack+0x10/0x14) from [<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210)
[<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210) from [<c00e0c60>] (request_threaded_irq+0x88/0x11c)
[<c00e0c60>] (request_threaded_irq+0x88/0x11c) from [<c03bf53c>] (dwc3_gadget_start+0x198/0x200)
[<c03bf53c>] (dwc3_gadget_start+0x198/0x200) from [<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8)
[<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8) from [<c03f7b50>] (usb_gadget_probe_driver+0x8c/0xb8)

and I suspect this problem was introduced in commit 8698e2acf (usb: dwc3: gadget: introduce and use enable/disable irq methods). Is there a fix for this problem? Can we just move the irq request outside the spinlock?

Quoted 7 lines
Quoted 6 lines
Quoted 3 lines

nice :-)

how about this ?

If start fails do you call stop? I believe the answer is no, so we'll need to free_irq() somewhere along the error path. Or we can request it after the spin_unlock()?

good point here's v2:

Ok looks good to me. I hope that platform_get_irq() doesn't fail, otherwise we're in for a nasty surprise. Maybe we should add a check in request_irq() for that case.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
#5 lore raw
Quoted 27 lines · Stephen Boyd
Quoted 26 lines · Felipe Balbi

Hi,

Quoted 23 lines · Stephen Boyd
Quoted 22 lines · Felipe Balbi
Quoted 14 lines · Stephen Boyd

Hi,

I'm getting the folllowing BUG message on bootup with 3.10-rc5

BUG: sleeping function called from invalid context at mm/slub.c:926
in_atomic(): 1, irqs_disabled(): 128, pid: 1, name: swapper/0
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc5-gee3e35b-09316-ge78f3b35 #643
[<c0014220>] (unwind_backtrace+0x0/0x120) from [<c001212c>] (show_stack+0x10/0x14)
[<c001212c>] (show_stack+0x10/0x14) from [<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210)
[<c0143750>] (kmem_cache_alloc_trace+0x3c/0x210) from [<c00e0c60>] (request_threaded_irq+0x88/0x11c)
[<c00e0c60>] (request_threaded_irq+0x88/0x11c) from [<c03bf53c>] (dwc3_gadget_start+0x198/0x200)
[<c03bf53c>] (dwc3_gadget_start+0x198/0x200) from [<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8)
[<c03f7a5c>] (udc_bind_to_driver+0x70/0xd8) from [<c03f7b50>] (usb_gadget_probe_driver+0x8c/0xb8)
Quoted 4 lines

and I suspect this problem was introduced in commit 8698e2acf (usb: dwc3: gadget: introduce and use enable/disable irq methods). Is there a fix for this problem? Can we just move the irq request outside the spinlock?

nice :-)

how about this ?

Quoted 8 lines
Quoted 4 lines
Quoted 3 lines

If start fails do you call stop? I believe the answer is no, so we'll need to free_irq() somewhere along the error path. Or we can request it after the spin_unlock()?

good point here's v2:

Ok looks good to me. I hope that platform_get_irq() doesn't fail, otherwise we're in for a nasty surprise. Maybe we should add a check in request_irq() for that case.

I don't think we would ever fall into that situation. And if we do, our data (struct resource or DT) is f-ed up anyway :-)

Signature · 1 line
balbi