[PATCH] linux/container_of.h: Add memberof()

[Alejandro Colomar <alx@kernel.org>]

From: Alejandro Colomar <alx.manpages@gmail.com>

Hi!

On 2023-08-18 10:46, Andy Shevchenko wrote:
> On Fri, Aug 18, 2023 at 01:28:42PM +0800, Herbert Xu wrote:
>> On Thu, Aug 17, 2023 at 04:33:17PM +0200, Lucas Segarra Fernandez wrote:
>>>
>>> +static struct pm_status_row pm_event_rows[] = {
>>> +	PM_INFO_REGSET_ENTRY32(event_log[0], EVENT0),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[1], EVENT1),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[2], EVENT2),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[3], EVENT3),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[4], EVENT4),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[5], EVENT5),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[6], EVENT6),
>>> +	PM_INFO_REGSET_ENTRY32(event_log[7], EVENT7),
>>> +};
>>> +
>>> +static_assert(ARRAY_SIZE_OF_FIELD(struct icp_qat_fw_init_admin_pm_info, event_log) ==
>>> +	      ARRAY_SIZE(pm_event_rows));

How about the following?

static_assert(ARRAY_SIZE(pm_event_rows) ==
    ARRAY_SIZE(memberof(struct icp_qat_fw_init_admin_pm_info, event_log)));

It would only need one macro addition, without significant churn.  It's
even less typing.  Below is a scissor patch with the addition of
memberof().

I tried building the kernel, and didn't see any warnings form the patch
below.

Cheers,
Alex

>>
>> Was all of that churn just for this one line?
>>
>> How about simply declaring a macro
>>
>> 	#define QAT_NUMBER_OF_PM_EVENTS 8
>>
>> and then use it for the two arrays:
>>
>> 	static struct pm_status_row pm_event_rows[QAT_NUMBER_OF_PM_EVENTS] = {
>>
>> 	__u32 event_log[QAT_NUMBER_OF_PM_EVENTS];
>>
>> What am I missing?
> 
> Splitting ARRAY_SIZE() is very beneficial on its own.
> The static assert is slightly more robust for the big code then defining
> something that at some point can be missed or miscalculated. Yet we can
> survive with a macro if you thinks it's better.
> 

-----8<------------

Many xxxof_{member,field}() macros make use of the same construction to
refer to a member of a struct without needing a variable of the
structure type.

memberof(T, m) simplifies all of those, avoids possible mistakes in
repetition, adds a meaningful name to the construction, and improves
readability by avoiding too many parentheses together.

It uses a compound literal, which should optimized out by the compiler.
It's a bit simpler to read than the dereference of a casted null
pointer, due to having less parentheses in the implementation.

Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Lucas Segarra Fernandez <lucas.segarra.fernandez@intel.com>
Cc: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
---
 include/linux/container_of.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/include/linux/container_of.h b/include/linux/container_of.h
index 713890c867be..5e762025c780 100644
--- a/include/linux/container_of.h
+++ b/include/linux/container_of.h
@@ -5,7 +5,9 @@
 #include <linux/build_bug.h>
 #include <linux/stddef.h>
 
-#define typeof_member(T, m)	typeof(((T*)0)->m)
+
+#define memberof(T, member)  ((T){}.member)
+#define typeof_member(T, m)  typeof(memberof(T, m))
 
 /**
  * container_of - cast a member of a structure out to the containing structure
-- 
2.40.1