iphase: BUG: sleeping function called from invalid context

PATCH In review Jorge Boncompte [DTI2] 2 participants 4 messages Jun 5, 2008 – Jun 15, 2008

Subsystems (3)

Networking Network drivers ATM

lore

Share & tools

lore

Jorge Boncompte [DTI2]

author

2008-06-05 16:12 UTC #0 lore raw

PATCH 1/1 iphase: BUG: sleeping function called from invalid context

iphase driver calls ioremap under spinlock_irqsave in his
initialization function. The spinlock seems to be there just for the sole purpose of preventing initializing multiple cards at once. So I think that removing it should be fine.

Signed-off-by Jorge Boncompte [DTI2] <[email protected]>

---
 drivers/atm/iphase.c |    6 ------
 drivers/atm/iphase.h |    2 +-
 2 files changed, 1 insertions(+), 7 deletions(-)

drivers/atm/iphase.c +0 -6

index 5c28ca7..a51b7cb 100644
@@ -3171,7 +3171,6 @@ static int __devinit ia_init_one(struct pci_dev *pdev,
 {
 	struct atm_dev *dev;
 	IADEV *iadev;
-        unsigned long flags;
 	int ret;
 
 	iadev = kzalloc(sizeof(*iadev), GFP_KERNEL);

@@ -3201,19 +3200,14 @@ static int __devinit ia_init_one(struct pci_dev *pdev,
 	ia_dev[iadev_count] = iadev;
 	_ia_dev[iadev_count] = dev;
 	iadev_count++;
-	spin_lock_init(&iadev->misc_lock);
-	/* First fixes first. I don't want to think about this now. */
-	spin_lock_irqsave(&iadev->misc_lock, flags);
 	if (ia_init(dev) || ia_start(dev)) {
 		IF_INIT(printk("IA register failed!\n");)
 		iadev_count--;
 		ia_dev[iadev_count] = NULL;
 		_ia_dev[iadev_count] = NULL;
-		spin_unlock_irqrestore(&iadev->misc_lock, flags);
 		ret = -EINVAL;
 		goto err_out_deregister_dev;
 	}
-	spin_unlock_irqrestore(&iadev->misc_lock, flags);
 	IF_EVENT(printk("iadev_count = %d\n", iadev_count);)
 
 	iadev->next_board = ia_boards;

drivers/atm/iphase.h +1 -1

index b2cd20f..f3457a9 100644
@@ -1022,7 +1022,7 @@ typedef struct iadev_t {
 	struct dle_q rx_dle_q;
 	struct free_desc_q *rx_free_desc_qhead;
 	struct sk_buff_head rx_dma_q;
-        spinlock_t rx_lock, misc_lock;
+        spinlock_t rx_lock;
 	struct atm_vcc **rx_open;	/* list of all open VCs */
         u16 num_rx_desc, rx_buf_sz, rxing;
         u32 rx_pkt_ram, rx_tmp_cnt;

Chas Williams (CONTRACTOR)

2008-06-14 21:06 UTC #1 lore raw

Quoted 4 lines · In message <[email protected]>,"Jorge Boncompte [DTI2]" writes:

iphase driver calls ioremap under spinlock_irqsave in his
initialization function. The spinlock seems to be there just for the sole purpose of preventing initializing multiple cards at once. So I think that removing it should be fine.

the lock is necessary to prevent multiple cards from modifying iadev_count at the wrong time since iadev_count is incremented and decremented. it might be simpler to just make a local copy of iadev_count and just always increment iadev_count. if you fail to init a particular card, you will just leave a hole in ia_dev[]. something like the attached.

also, ia_remove_one() is broken since it assumes that the cards will be removed in the same order that they were installed.

even better would be to rewrite this to avoid the iadev[] array entirely and use a linked list.

drivers/atm/iphase.c +8 ~~-15~~

index 5c28ca7..64d1c97 100644
@@ -3171,7 +3171,6 @@ static int __devinit ia_init_one(struct pci_dev *pdev,
 {  
 	struct atm_dev *dev;  
 	IADEV *iadev;  
-        unsigned long flags;
 	int ret;
 
 	iadev = kzalloc(sizeof(*iadev), GFP_KERNEL);

@@ -3180,6 +3179,7 @@ static int __devinit ia_init_one(struct pci_dev *pdev,
 		goto err_out;
 	}
 
+	iadev->iadev_count = iadev_count++;
 	iadev->pci = pdev;
 
 	IF_INIT(printk("ia detected at bus:%d dev: %d function:%d\n",

@@ -3198,23 +3198,17 @@ static int __devinit ia_init_one(struct pci_dev *pdev,
 	IF_INIT(printk("dev_id = 0x%x iadev->LineRate = %d \n", (u32)dev,
 		iadev->LineRate);)
 
-	ia_dev[iadev_count] = iadev;
-	_ia_dev[iadev_count] = dev;
-	iadev_count++;
-	spin_lock_init(&iadev->misc_lock);
-	/* First fixes first. I don't want to think about this now. */
-	spin_lock_irqsave(&iadev->misc_lock, flags); 
+	ia_dev[iadev->iadev_count] = iadev;
+	_ia_dev[iadev->iadev_count] = dev;
 	if (ia_init(dev) || ia_start(dev)) {  
 		IF_INIT(printk("IA register failed!\n");)
 		iadev_count--;
-		ia_dev[iadev_count] = NULL;
-		_ia_dev[iadev_count] = NULL;
-		spin_unlock_irqrestore(&iadev->misc_lock, flags); 
+		ia_dev[iadev->iadev_count] = NULL;
+		_ia_dev[iadev->iadev_count] = NULL;
 		ret = -EINVAL;
 		goto err_out_deregister_dev;
 	}
-	spin_unlock_irqrestore(&iadev->misc_lock, flags); 
-	IF_EVENT(printk("iadev_count = %d\n", iadev_count);)
+	IF_EVENT(printk("iadev_count = %d\n", iadev->iadev_count);)
 
 	iadev->next_board = ia_boards;  
 	ia_boards = dev;

@@ -3243,9 +3237,8 @@ static void __devexit ia_remove_one(struct pci_dev *pdev)
 
 	/* De-register device */  
       	free_irq(iadev->irq, dev);
-	iadev_count--;
-	ia_dev[iadev_count] = NULL;
-	_ia_dev[iadev_count] = NULL;
+	ia_dev[iadev->iadev_count] = NULL;
+	_ia_dev[iadev->iadev_count] = NULL;
 	IF_EVENT(printk("deregistering iav at (itf:%d)\n", dev->number);)
 	atm_dev_deregister(dev);

drivers/atm/iphase.h +2 -1

index b2cd20f..a0b1982 100644
@@ -1022,7 +1022,7 @@ typedef struct iadev_t {
 	struct dle_q rx_dle_q;  
 	struct free_desc_q *rx_free_desc_qhead;  
 	struct sk_buff_head rx_dma_q;  
-        spinlock_t rx_lock, misc_lock;
+        spinlock_t rx_lock;
 	struct atm_vcc **rx_open;	/* list of all open VCs */  
         u16 num_rx_desc, rx_buf_sz, rxing;
         u32 rx_pkt_ram, rx_tmp_cnt;

@@ -1064,6 +1064,7 @@ typedef struct iadev_t {
         struct testTable_t **testTable;
 	dma_addr_t tx_dle_dma;
 	dma_addr_t rx_dle_dma;
+	int iadev_count;
 } IADEV;

Jorge Boncompte [DTI2]

author

2008-06-15 13:00 UTC #2 lore raw

Quoted 9 lines · Chas Williams (CONTRACTOR)

Quoted 2 lines · In message <[email protected]>,"Jorge Boncompte [DTI2]" writes:
iphase driver calls ioremap under spinlock_irqsave in his
initialization function. The spinlock seems to be there just for the sole purpose of preventing initializing multiple cards at once. So I think that removing it should be fine.
the lock is necessary to prevent multiple cards from modifying iadev_count at the wrong time since iadev_count is incremented and decremented. it might be simpler to just make a local copy of iadev_count and just always increment iadev_count. if you fail to init a particular card, you will just leave a hole in ia_dev[]. something like the attached.

I thought that device initialization inside a driver is serialized. isn't it?

Quoted 6 lines

also, ia_remove_one() is broken since it assumes that the cards will be removed in the same order that they were installed.
even better would be to rewrite this to avoid the iadev[] array entirely and use a linked list.

I have a patch that removes iadev and iadev_count enterely that i did not sent because i thought it wouldn't be
considered a bug fix. I'll sent it this week.

Thanks for your review,

-Jorge

Signature · 10 lines

==============================================================
Jorge Boncompte - Ingenieria y Gestion de RED
DTI2 - Desarrollo de la Tecnologia de las Comunicaciones
--------------------------------------------------------------
C/ Abogado Enriquez Barrios, 5   14004 CORDOBA (SPAIN)
Tlf: +34 957 761395 / FAX: +34 957 450380
==============================================================
- Sin pistachos no hay Rock & Roll...
- Without wicker a basket cannot be made.
==============================================================

Chas Williams (CONTRACTOR)

2008-06-15 14:47 UTC #3 lore raw

Quoted 7 lines · In message <[email protected]>,"Jorge Boncompte [DTI2]" writes:

Quoted 5 lines
the lock is necessary to prevent multiple cards from modifying iadev_count at the wrong time since iadev_count is incremented and decremented. it might be simpler to just make a local copy of iadev_count and just always increment iadev_count. if you fail to init a particular card, you will just leave a hole in ia_dev[]. something like the attached.
I thought that device initialization inside a driver is serialized. isn't it?

i honestly dont know. even if it is serialized, that code is just hideous. if you had a system with hotplug support, it wouldnt work right.

Quoted 2 lines

I have a patch that removes iadev and iadev_count enterely that i did not sent because i thought it wouldn't be
considered a bug fix. I'll sent it this week.

i would certainly prefer that version. no matter, how you fix the problem it is a bug fix.