[PATCH] Add PCI_DEVICE_TABLE macro

[PATCH] Add PCI_DEVICE_TABLE macro

[Jonas Bonn]

The definitions of struct pci_device_id arrays should generally follow
the same pattern across the entire kernel.  This macro defines this
array as static const and puts it into the __devinitconst section.

Signed-off-by: Jonas Bonn <jonas@southpole.se>
---
 include/linux/pci.h |    8 ++++++++
 1 files changed, 8 insertions(+), 0 deletions(-)

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 87195b6..487d31c 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -389,6 +389,14 @@ struct pci_driver {
 #define	to_pci_driver(drv) container_of(drv, struct pci_driver, driver)
 
 /**
+ * PCI_DEVICE_TABLE - macro used to describe a pci device table
+ * 
+ * This macro is used to create a struct pci_device_id array in a generic
+ * manner.
+ */
+#define PCI_DEVICE_TABLE(_table) static const struct pci_device_id _table[] __devinitconst
+
+/**
  * PCI_DEVICE - macro used to describe a specific pci device
  * @vend: the 16 bit PCI Vendor ID
  * @dev: the 16 bit PCI Device ID
-- 
1.5.3.8

Re: [PATCH] Add PCI_DEVICE_TABLE macro

[Jonas Bonn]

[-- Attachment #1: Type: text/plain, Size: 1403 bytes --]

I'm a bit uncertain whether the definition really should include the
'static' modifier... for most definitions of these tables this is ok,
but there are a couple of cases where it should not be static, so the
line would need to be open-coded again...

/Jonas

On Sun, 2008-02-17 at 13:10 +0100, Jonas Bonn wrote:
> The definitions of struct pci_device_id arrays should generally follow
> the same pattern across the entire kernel.  This macro defines this
> array as static const and puts it into the __devinitconst section.
> 
> Signed-off-by: Jonas Bonn <jonas@southpole.se>
> ---
>  include/linux/pci.h |    8 ++++++++
>  1 files changed, 8 insertions(+), 0 deletions(-)
> 
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 87195b6..487d31c 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -389,6 +389,14 @@ struct pci_driver {
>  #define	to_pci_driver(drv) container_of(drv, struct pci_driver, driver)
>  
>  /**
> + * PCI_DEVICE_TABLE - macro used to describe a pci device table
> + * 
> + * This macro is used to create a struct pci_device_id array in a generic
> + * manner.
> + */
> +#define PCI_DEVICE_TABLE(_table) static const struct pci_device_id _table[] __devinitconst
> +
> +/**
>   * PCI_DEVICE - macro used to describe a specific pci device
>   * @vend: the 16 bit PCI Vendor ID
>   * @dev: the 16 bit PCI Device ID

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [PATCH] Add PCI_DEVICE_TABLE macro

[Greg KH]

On Sun, Feb 17, 2008 at 01:13:24PM +0100, Jonas Bonn wrote:
> I'm a bit uncertain whether the definition really should include the
> 'static' modifier... for most definitions of these tables this is ok,
> but there are a couple of cases where it should not be static, so the
> line would need to be open-coded again...

Yes, please don't make it static.

And again, what does this buy us?

thanks,

greg k-h

Re: [PATCH] Add PCI_DEVICE_TABLE macro

[Jonas Bonn]

> 
> And again, what does this buy us?
> 

Clarity and simplicity, I hope... there are a bunch of definitions 
scattered about the kernel that omit the __devinitdata modifier despite 
the documentation stating that it should always be there.  The 
definition really should have been const, which wasn't possible before 
but has become so with the addition of the __devinitconst attribute.

Furthermore, there are definitions that use "const" and __devinitdata, 
which is explicitly wrong but the compiler doesn't catch section 
mismatches if there's only one such one case in the module (which is 
often the case).

Adding the __devinitconst modifier where there was nothing before buys 
us memory.  Adding the const modifier gives the compiler a chance to do 
its thing.  Changing __devinitdata to __devinitconst where it was wrong 
actually fixes some compiler errors in older (mid-release) kernels that 
were patched over by "removing" the section attribute altogether (which 
wastes memory).

Adding the macro (Olof's idea, not mine) makes it pretty difficult to 
get this definition wrong... I'll do the rest of the cleanup, but I need 
to know whether it's better to use a macro like this, or to open code 
the definitions.  I prefer the macro approach...

Hope this makes some sense...

/Jonas

Re: [PATCH] Add PCI_DEVICE_TABLE macro

[Greg KH]

On Mon, Feb 18, 2008 at 08:34:42AM +0100, Jonas Bonn wrote:
>> And again, what does this buy us?
>
> Clarity and simplicity, I hope... there are a bunch of definitions 
> scattered about the kernel that omit the __devinitdata modifier despite the 
> documentation stating that it should always be there.  The definition 
> really should have been const, which wasn't possible before but has become 
> so with the addition of the __devinitconst attribute.
>
> Furthermore, there are definitions that use "const" and __devinitdata, 
> which is explicitly wrong but the compiler doesn't catch section mismatches 
> if there's only one such one case in the module (which is often the case).
>
> Adding the __devinitconst modifier where there was nothing before buys us 
> memory.  Adding the const modifier gives the compiler a chance to do its 
> thing.  Changing __devinitdata to __devinitconst where it was wrong 
> actually fixes some compiler errors in older (mid-release) kernels that 
> were patched over by "removing" the section attribute altogether (which 
> wastes memory).
>
> Adding the macro (Olof's idea, not mine) makes it pretty difficult to get 
> this definition wrong... I'll do the rest of the cleanup, but I need to 
> know whether it's better to use a macro like this, or to open code the 
> definitions.  I prefer the macro approach...
>
> Hope this makes some sense...

Ok, yes it does, thanks for the explaination.

Please, can you add this very good text to the changelog entry for the
addition of the macro, and to the documentation somewhere?  I'd be glad
to take the patch if that was done.

thanks,

greg k-h

Re: [PATCH] Add PCI_DEVICE_TABLE macro

[Jeff Garzik]

Greg KH wrote:
> On Mon, Feb 18, 2008 at 08:34:42AM +0100, Jonas Bonn wrote:
>>> And again, what does this buy us?
>> Clarity and simplicity, I hope... there are a bunch of definitions 
>> scattered about the kernel that omit the __devinitdata modifier despite the 
>> documentation stating that it should always be there.  The definition 
>> really should have been const, which wasn't possible before but has become 
>> so with the addition of the __devinitconst attribute.
>>
>> Furthermore, there are definitions that use "const" and __devinitdata, 
>> which is explicitly wrong but the compiler doesn't catch section mismatches 
>> if there's only one such one case in the module (which is often the case).
>>
>> Adding the __devinitconst modifier where there was nothing before buys us 
>> memory.  Adding the const modifier gives the compiler a chance to do its 
>> thing.  Changing __devinitdata to __devinitconst where it was wrong 
>> actually fixes some compiler errors in older (mid-release) kernels that 
>> were patched over by "removing" the section attribute altogether (which 
>> wastes memory).
>>
>> Adding the macro (Olof's idea, not mine) makes it pretty difficult to get 
>> this definition wrong... I'll do the rest of the cleanup, but I need to 
>> know whether it's better to use a macro like this, or to open code the 
>> definitions.  I prefer the macro approach...
>>
>> Hope this makes some sense...
> 
> Ok, yes it does, thanks for the explaination.
> 
> Please, can you add this very good text to the changelog entry for the
> addition of the macro, and to the documentation somewhere?  I'd be glad
> to take the patch if that was done.

I would suggest having a DECLARE_ prefix in there, like other subsystems 
do...

	Jeff

[PATCH] Add DECLARE_PCI_DEVICE_TABLE macro

[Jonas Bonn]

The definitions of struct pci_device_id arrays should generally follow
the same pattern across the entire kernel.  This macro defines this
array as const and puts it into the __devinitconst section.

There are currently many definitions scattered about the kernel that
omit the __devinitdata modifier despite the documentation stating that
it should always be there.  These definitions really also should have
been const, which wasn't possible before but has become so with the
addition of the __devinitconst attribute.

Furthermore, there are definitions that use "const" and __devinitdata,
which is explicitly wrong but the compiler doesn't catch section
mismatches if there's only one such one case in the module (which is
often the case).

Adding the __devinitconst modifier where there was nothing before buys
us memory.  Adding the const modifier gives the compiler a chance to do
its thing.  Changing __devinitdata to __devinitconst where it was wrong
actually fixes some compiler errors in older (mid-release) kernels that
were patched over by "removing" the section attribute altogether (which
wastes memory).

This macro makes it pretty difficult to get this definition wrong in
the future...

Signed-off-by: Jonas Bonn <jonas@southpole.se>
---
 Documentation/pci.txt |    6 ++++--
 include/linux/pci.h   |    9 +++++++++
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/Documentation/pci.txt b/Documentation/pci.txt
index 72b20c6..bb7bd27 100644
--- a/Documentation/pci.txt
+++ b/Documentation/pci.txt
@@ -123,7 +123,8 @@ initialization with a pointer to a structure describing the driver
 
 
 The ID table is an array of struct pci_device_id entries ending with an
-all-zero entry.  Each entry consists of:
+all-zero entry; use of the macro DECLARE_PCI_DEVICE_TABLE is the preferred
+method of declaring the table.  Each entry consists of:
 
 	vendor,device	Vendor and device ID to match (or PCI_ANY_ID)
 
@@ -191,7 +192,8 @@ Tips on when/where to use the above attributes:
 
 	o Do not mark the struct pci_driver.
 
-	o The ID table array should be marked __devinitdata.
+	o The ID table array should be marked __devinitconst; this is done
+	  automatically if the table is declared with DECLARE_PCI_DEVICE_TABLE().
 
 	o The probe() and remove() functions should be marked __devinit
 	  and __devexit respectively.  All initialization functions
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 87195b6..a41a484 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -389,6 +389,15 @@ struct pci_driver {
 #define	to_pci_driver(drv) container_of(drv, struct pci_driver, driver)
 
 /**
+ * DECLARE_PCI_DEVICE_TABLE - macro used to describe a pci device table
+ * @_table: device table name
+ *
+ * This macro is used to create a struct pci_device_id array (a device table) 
+ * in a generic manner.
+ */
+#define DECLARE_PCI_DEVICE_TABLE(_table) const struct pci_device_id _table[] __devinitconst
+
+/**
  * PCI_DEVICE - macro used to describe a specific pci device
  * @vend: the 16 bit PCI Vendor ID
  * @dev: the 16 bit PCI Device ID
-- 
1.5.3.8

Re: [PATCH] Add PCI_DEVICE_TABLE macro

[Randy Dunlap]

On Sun, 17 Feb 2008 13:10:05 +0100 Jonas Bonn wrote:

> The definitions of struct pci_device_id arrays should generally follow
> the same pattern across the entire kernel.  This macro defines this
> array as static const and puts it into the __devinitconst section.
> 
> Signed-off-by: Jonas Bonn <jonas@southpole.se>
> ---
>  include/linux/pci.h |    8 ++++++++
>  1 files changed, 8 insertions(+), 0 deletions(-)
> 
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 87195b6..487d31c 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -389,6 +389,14 @@ struct pci_driver {
>  #define	to_pci_driver(drv) container_of(drv, struct pci_driver, driver)
>  
>  /**
> + * PCI_DEVICE_TABLE - macro used to describe a pci device table
> + * 
> + * This macro is used to create a struct pci_device_id array in a generic
> + * manner.
> + */
> +#define PCI_DEVICE_TABLE(_table) static const struct pci_device_id _table[] __devinitconst
> +
> +/**
>   * PCI_DEVICE - macro used to describe a specific pci device
>   * @vend: the 16 bit PCI Vendor ID
>   * @dev: the 16 bit PCI Device ID
> -- 

Missing kernel-doc notation for the _table parameter.  See the
parameters for PCI_DEVICE() below for an example.

---
~Randy

[PATCH] Add PCI_DEVICE_TABLE macro

[Jonas Bonn]

The definitions of struct pci_device_id arrays should generally follow
the same pattern across the entire kernel.  This macro defines this
array as const and puts it into the __devinitconst section.

Signed-off-by: Jonas Bonn <jonas@southpole.se>
---
 include/linux/pci.h |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 87195b6..c7a91b1 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -389,6 +389,15 @@ struct pci_driver {
 #define	to_pci_driver(drv) container_of(drv, struct pci_driver, driver)
 
 /**
+ * PCI_DEVICE_TABLE - macro used to describe a pci device table
+ * @_table: device table name
+ *
+ * This macro is used to create a struct pci_device_id array (a device table) 
+ * in a generic manner.
+ */
+#define PCI_DEVICE_TABLE(_table) const struct pci_device_id _table[] __devinitconst
+
+/**
  * PCI_DEVICE - macro used to describe a specific pci device
  * @vend: the 16 bit PCI Vendor ID
  * @dev: the 16 bit PCI Device ID
-- 
1.5.3.8

pci_device_id cleanups

[Jonas Bonn]

The PCI_DEVICE_TABLE patch I sent earlier doesn't necessarily make much sense by itself... here is a set of patches that apply this macro, in turn moving a lot of this data into __devinitconst which is discardable in certain situations.  Hopefully the benefit of this approach is a bit clearer now.

 drivers/net/3c59x.c               |    2 +-
 drivers/net/amd8111e.c            |    2 +-
 drivers/net/arcnet/com20020-pci.c |    2 +-
 drivers/net/defxx.c               |    2 +-
 drivers/net/hamachi.c             |    2 +-
 drivers/net/niu.c                 |    2 +-
 drivers/net/pasemi_mac.c          |    2 +-
 drivers/net/sk98lin/skge.c        |    2 +-
 drivers/net/skfp/skfddi.c         |    2 +-
 drivers/net/starfire.c            |    2 +-
 drivers/net/sunhme.c              |    2 +-
 drivers/net/tlan.c                |    2 +-
 drivers/net/wan/dscc4.c           |    2 +-
 drivers/net/wan/lmc/lmc_main.c    |    2 +-
 include/linux/pci.h               |    9 +++++++++
 15 files changed, 23 insertions(+), 14 deletions(-)

[PATCH] Add PCI_DEVICE_TABLE macro

[Jonas Bonn]

The definitions of struct pci_device_id arrays should generally follow
the same pattern across the entire kernel.  This macro defines this
array as const and puts it into the __devinitconst section.

Signed-off-by: Jonas Bonn <jonas@southpole.se>
---
 include/linux/pci.h |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 87195b6..c7a91b1 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -389,6 +389,15 @@ struct pci_driver {
 #define	to_pci_driver(drv) container_of(drv, struct pci_driver, driver)
 
 /**
+ * PCI_DEVICE_TABLE - macro used to describe a pci device table
+ * @_table: device table name
+ *
+ * This macro is used to create a struct pci_device_id array (a device table) 
+ * in a generic manner.
+ */
+#define PCI_DEVICE_TABLE(_table) const struct pci_device_id _table[] __devinitconst
+
+/**
  * PCI_DEVICE - macro used to describe a specific pci device
  * @vend: the 16 bit PCI Vendor ID
  * @dev: the 16 bit PCI Device ID
-- 
1.5.3.8