[PATCH 1/2] pdc202xx_new: fix PCI refcounting

[PATCH 1/2] pdc202xx_new: fix PCI refcounting

[Sergei Shtylyov]

The driver erroneously "lets go" the mate IDE chip in init_setup_pdc20270()
when ide_setup_pci_devices() call succeeds -- fix this, and drop a couple of
useless assignments in this function while at it...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>

---
This patch is against the current Linus' tree, it has only been compile tested
since I do not have PDC2027x chips (and even less so behind DC21150 bridge :-).

 drivers/ide/pci/pdc202xx_new.c |    9 +++++----
 1 files changed, 5 insertions(+), 4 deletions(-)

Index: linux-2.6/drivers/ide/pci/pdc202xx_new.c
===================================================================
--- linux-2.6.orig/drivers/ide/pci/pdc202xx_new.c
+++ linux-2.6/drivers/ide/pci/pdc202xx_new.c
@@ -9,7 +9,7 @@
  *  Split from:
  *  linux/drivers/ide/pdc202xx.c	Version 0.35	Mar. 30, 2002
  *  Copyright (C) 1998-2002		Andre Hedrick <andre@linux-ide.org>
- *  Copyright (C) 2005-2006		MontaVista Software, Inc.
+ *  Copyright (C) 2005-2007		MontaVista Software, Inc.
  *  Portions Copyright (C) 1999 Promise Technology, Inc.
  *  Author: Frank Tiernan (frankt@promise.com)
  *  Released under terms of General Public License
@@ -521,7 +521,7 @@ static int __devinit init_setup_pdcnew(s
 static int __devinit init_setup_pdc20270(struct pci_dev *dev,
 					 ide_pci_device_t *d)
 {
-	struct pci_dev *findev = NULL;
+	struct pci_dev *findev;
 	int ret;
 
 	if ((dev->bus->self &&
@@ -529,7 +529,7 @@ static int __devinit init_setup_pdc20270
 	    (dev->bus->self->device == PCI_DEVICE_ID_DEC_21150)) {
 		if (PCI_SLOT(dev->devfn) & 2)
 			return -ENODEV;
-		d->extra = 0;
+
 		while ((findev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, findev)) != NULL) {
 			if ((findev->vendor == dev->vendor) &&
 			    (findev->device == dev->device) &&
@@ -538,7 +538,8 @@ static int __devinit init_setup_pdc20270
 					findev->irq = dev->irq;
 				}
 				ret = ide_setup_pci_devices(dev, findev, d);
-				pci_dev_put(findev);
+				if (ret < 0)
+					pci_dev_put(findev);
 				return ret;
 			}
 		}

Re: [PATCH 1/2] pdc202xx_new: fix PCI refcounting

[Bartlomiej Zolnierkiewicz]

On Saturday 18 August 2007, Sergei Shtylyov wrote:
> The driver erroneously "lets go" the mate IDE chip in init_setup_pdc20270()
> when ide_setup_pci_devices() call succeeds -- fix this, and drop a couple of
> useless assignments in this function while at it...
> 
> Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>

applied but

> ---
> This patch is against the current Linus' tree, it has only been compile tested
> since I do not have PDC2027x chips (and even less so behind DC21150 bridge :-).
> 
>  drivers/ide/pci/pdc202xx_new.c |    9 +++++----
>  1 files changed, 5 insertions(+), 4 deletions(-)
> 
> Index: linux-2.6/drivers/ide/pci/pdc202xx_new.c
> ===================================================================
> --- linux-2.6.orig/drivers/ide/pci/pdc202xx_new.c
> +++ linux-2.6/drivers/ide/pci/pdc202xx_new.c
> @@ -9,7 +9,7 @@
>   *  Split from:
>   *  linux/drivers/ide/pdc202xx.c	Version 0.35	Mar. 30, 2002
>   *  Copyright (C) 1998-2002		Andre Hedrick <andre@linux-ide.org>
> - *  Copyright (C) 2005-2006		MontaVista Software, Inc.
> + *  Copyright (C) 2005-2007		MontaVista Software, Inc.
>   *  Portions Copyright (C) 1999 Promise Technology, Inc.
>   *  Author: Frank Tiernan (frankt@promise.com)
>   *  Released under terms of General Public License
> @@ -521,7 +521,7 @@ static int __devinit init_setup_pdcnew(s
>  static int __devinit init_setup_pdc20270(struct pci_dev *dev,
>  					 ide_pci_device_t *d)
>  {
> -	struct pci_dev *findev = NULL;
> +	struct pci_dev *findev;

drivers/ide/pci/pdc202xx_new.c: In function ‘init_setup_pdc20270’:
drivers/ide/pci/pdc202xx_new.c:530: warning: ‘findev’ may be used uninitialized in this function

with gcc version 4.1.2 20070626 (Red Hat 4.1.2-13)

Seems to be gcc problem but I reverted this chunk.

Update: after applying patch #2/2 and fixing the reject caused by the above change
warning disappears (patch #2/2 besides renaming "findev" to "dev2" moves the code
around making it easier for gcc to deduce what is going on :).

>  	int ret;
>  
>  	if ((dev->bus->self &&
> @@ -529,7 +529,7 @@ static int __devinit init_setup_pdc20270
>  	    (dev->bus->self->device == PCI_DEVICE_ID_DEC_21150)) {
>  		if (PCI_SLOT(dev->devfn) & 2)
>  			return -ENODEV;
> -		d->extra = 0;
> +
>  		while ((findev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, findev)) != NULL) {
>  			if ((findev->vendor == dev->vendor) &&
>  			    (findev->device == dev->device) &&
> @@ -538,7 +538,8 @@ static int __devinit init_setup_pdc20270
>  					findev->irq = dev->irq;
>  				}
>  				ret = ide_setup_pci_devices(dev, findev, d);
> -				pci_dev_put(findev);
> +				if (ret < 0)
> +					pci_dev_put(findev);
>  				return ret;
>  			}
>  		}

Re: [PATCH 1/2] pdc202xx_new: fix PCI refcounting

[Sergei Shtylyov]

Bartlomiej Zolnierkiewicz wrote:

>>The driver erroneously "lets go" the mate IDE chip in init_setup_pdc20270()
>>when ide_setup_pci_devices() call succeeds -- fix this, and drop a couple of
>>useless assignments in this function while at it...

>>Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>

> applied but

>>---
>>This patch is against the current Linus' tree, it has only been compile tested
>>since I do not have PDC2027x chips (and even less so behind DC21150 bridge :-).

>>Index: linux-2.6/drivers/ide/pci/pdc202xx_new.c
>>===================================================================
>>--- linux-2.6.orig/drivers/ide/pci/pdc202xx_new.c
>>+++ linux-2.6/drivers/ide/pci/pdc202xx_new.c
[...]
>>@@ -521,7 +521,7 @@ static int __devinit init_setup_pdcnew(s
>> static int __devinit init_setup_pdc20270(struct pci_dev *dev,
>> 					 ide_pci_device_t *d)
>> {
>>-	struct pci_dev *findev = NULL;
>>+	struct pci_dev *findev;

> drivers/ide/pci/pdc202xx_new.c: In function ‘init_setup_pdc20270’:
> drivers/ide/pci/pdc202xx_new.c:530: warning: ‘findev’ may be used uninitialized in this function

> with gcc version 4.1.2 20070626 (Red Hat 4.1.2-13)

> Seems to be gcc problem

    No, it must be all those pills I'm taking... ;-)

> but I reverted this chunk.

    And you were absolutely right -- findev must be NULL for the search to go 
thru all the PCI devices. :-/

> Update: after applying patch #2/2 and fixing the reject caused by the above change
> warning disappears (patch #2/2 besides renaming "findev" to "dev2" moves the code
> around making it easier for gcc to deduce what is going on :).

    Right, because pci_get_slot() is not a "device scanner" and so doesn't 
need the starting node.
    I guess there's no need to recast now?

WBR, Sergei