[PATCH 1/1] PCI: fix on-stack pci_device_id's

[PATCH 1/1] PCI: fix on-stack pci_device_id's

[Jiri Slaby]

Commit "PCI: Use pci_device_id on stack for pci_get_subsys/class() to
avoid kmalloc" changed heap allocations to on-stack variables, but it
did not add initialization of other than set members. This causes
random failures during bootup wherever pci device is needed to be
found. Hence the boot just hangs or panics.

This patches fixes it by setting the content of pci_device_id directly
in the initializer.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Feng Tang <feng.tang@intel.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
---
 drivers/pci/search.c | 25 ++++++++++++++-----------
 1 file changed, 14 insertions(+), 11 deletions(-)

diff --git a/drivers/pci/search.c b/drivers/pci/search.c
index 00d486a..bf969ba 100644
--- a/drivers/pci/search.c
+++ b/drivers/pci/search.c
@@ -243,12 +243,12 @@ struct pci_dev *pci_get_subsys(unsigned int vendor, unsigned int device,
 			       unsigned int ss_vendor, unsigned int ss_device,
 			       struct pci_dev *from)
 {
-	struct pci_device_id id;
-
-	id.vendor = vendor;
-	id.device = device;
-	id.subvendor = ss_vendor;
-	id.subdevice = ss_device;
+	struct pci_device_id id = {
+		.vendor = vendor,
+		.device = device,
+		.subvendor = ss_vendor,
+		.subdevice = ss_device,
+	};
 
 	return pci_get_dev_by_id(&id, from);
 }
@@ -289,11 +289,14 @@ pci_get_device(unsigned int vendor, unsigned int device, struct pci_dev *from)
  */
 struct pci_dev *pci_get_class(unsigned int class, struct pci_dev *from)
 {
-	struct pci_device_id id;
-
-	id.vendor = id.device = id.subvendor = id.subdevice = PCI_ANY_ID;
-	id.class_mask = PCI_ANY_ID;
-	id.class = class;
+	struct pci_device_id id = {
+		.vendor = PCI_ANY_ID,
+		.device = PCI_ANY_ID,
+		.subvendor = PCI_ANY_ID,
+		.subdevice = PCI_ANY_ID,
+		.class_mask = PCI_ANY_ID,
+		.class = class,
+	};
 
 	return pci_get_dev_by_id(&id, from);
 }
-- 
1.7.11.5

Re: [PATCH 1/1] PCI: fix on-stack pci_device_id's

[Bjorn Helgaas]

On Mon, Sep 10, 2012 at 1:36 PM, Jiri Slaby <jslaby@suse.cz> wrote:
> Commit "PCI: Use pci_device_id on stack for pci_get_subsys/class() to
> avoid kmalloc" changed heap allocations to on-stack variables, but it
> did not add initialization of other than set members. This causes
> random failures during bootup wherever pci device is needed to be
> found. Hence the boot just hangs or panics.
>
> This patches fixes it by setting the content of pci_device_id directly
> in the initializer.

Nice!  I fixed this already by adding a memset(), but I like your way
better.  I knew we could initialize members of a structure on the
stack, but I didn't know C guaranteed that uninitialized members would
be implicitly initialized also.  But apparently it does:

http://stackoverflow.com/questions/10828294/c-and-c-partial-initialization-of-automatic-structure

> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Feng Tang <feng.tang@intel.com>
> Cc: Bjorn Helgaas <bhelgaas@google.com>
> Cc: Fengguang Wu <fengguang.wu@intel.com>
> ---
>  drivers/pci/search.c | 25 ++++++++++++++-----------
>  1 file changed, 14 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/pci/search.c b/drivers/pci/search.c
> index 00d486a..bf969ba 100644
> --- a/drivers/pci/search.c
> +++ b/drivers/pci/search.c
> @@ -243,12 +243,12 @@ struct pci_dev *pci_get_subsys(unsigned int vendor, unsigned int device,
>                                unsigned int ss_vendor, unsigned int ss_device,
>                                struct pci_dev *from)
>  {
> -       struct pci_device_id id;
> -
> -       id.vendor = vendor;
> -       id.device = device;
> -       id.subvendor = ss_vendor;
> -       id.subdevice = ss_device;
> +       struct pci_device_id id = {
> +               .vendor = vendor,
> +               .device = device,
> +               .subvendor = ss_vendor,
> +               .subdevice = ss_device,
> +       };
>
>         return pci_get_dev_by_id(&id, from);
>  }
> @@ -289,11 +289,14 @@ pci_get_device(unsigned int vendor, unsigned int device, struct pci_dev *from)
>   */
>  struct pci_dev *pci_get_class(unsigned int class, struct pci_dev *from)
>  {
> -       struct pci_device_id id;
> -
> -       id.vendor = id.device = id.subvendor = id.subdevice = PCI_ANY_ID;
> -       id.class_mask = PCI_ANY_ID;
> -       id.class = class;
> +       struct pci_device_id id = {
> +               .vendor = PCI_ANY_ID,
> +               .device = PCI_ANY_ID,
> +               .subvendor = PCI_ANY_ID,
> +               .subdevice = PCI_ANY_ID,
> +               .class_mask = PCI_ANY_ID,
> +               .class = class,
> +       };
>
>         return pci_get_dev_by_id(&id, from);
>  }
> --
> 1.7.11.5
>
>

Re: [PATCH 1/1] PCI: fix on-stack pci_device_id's

[Fengguang Wu]

On Mon, Sep 10, 2012 at 02:44:32PM -0600, Bjorn Helgaas wrote:
> On Mon, Sep 10, 2012 at 1:36 PM, Jiri Slaby <jslaby@suse.cz> wrote:
> > Commit "PCI: Use pci_device_id on stack for pci_get_subsys/class() to
> > avoid kmalloc" changed heap allocations to on-stack variables, but it
> > did not add initialization of other than set members. This causes
> > random failures during bootup wherever pci device is needed to be
> > found. Hence the boot just hangs or panics.
> >
> > This patches fixes it by setting the content of pci_device_id directly
> > in the initializer.
> 
> Nice!  I fixed this already by adding a memset(), but I like your way
> better.  I knew we could initialize members of a structure on the
> stack, but I didn't know C guaranteed that uninitialized members would
> be implicitly initialized also.  But apparently it does:
> 
> http://stackoverflow.com/questions/10828294/c-and-c-partial-initialization-of-automatic-structure

Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>

For the record, I proposed the same initializer-style fix in the very
beginning that you didn't realize its beauty until now ;-)

Thanks,
Fengguang

Re: [PATCH 1/1] PCI: fix on-stack pci_device_id's

[Bjorn Helgaas]

On Mon, Sep 10, 2012 at 4:55 PM, Fengguang Wu <fengguang.wu@intel.com> wrote:
> On Mon, Sep 10, 2012 at 02:44:32PM -0600, Bjorn Helgaas wrote:
>> On Mon, Sep 10, 2012 at 1:36 PM, Jiri Slaby <jslaby@suse.cz> wrote:
>> > Commit "PCI: Use pci_device_id on stack for pci_get_subsys/class() to
>> > avoid kmalloc" changed heap allocations to on-stack variables, but it
>> > did not add initialization of other than set members. This causes
>> > random failures during bootup wherever pci device is needed to be
>> > found. Hence the boot just hangs or panics.
>> >
>> > This patches fixes it by setting the content of pci_device_id directly
>> > in the initializer.
>>
>> Nice!  I fixed this already by adding a memset(), but I like your way
>> better.  I knew we could initialize members of a structure on the
>> stack, but I didn't know C guaranteed that uninitialized members would
>> be implicitly initialized also.  But apparently it does:
>>
>> http://stackoverflow.com/questions/10828294/c-and-c-partial-initialization-of-automatic-structure
>
> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
>
> For the record, I proposed the same initializer-style fix in the very
> beginning that you didn't realize its beauty until now ;-)

Yep, I credited you :)  Sorry I didn't look closely enough to begin with.