Thread: weird typeded

Hi all,

Can anyone please explain to me what the following macro definitions are doing?

Code:

#define DECLARE_ARGS(val, low, high)	unsigned long long val
#define EAX_EDX_VAL(val, low, high)	(val)
#define EAX_EDX_RET(val, low, high)	"=A" (val)

They're used in an inline assembly that reads an MSR

Code:

static inline unsigned long long native_read_msr(unsigned int msr)
{
	DECLARE_ARGS(val, low, high);

	asm volatile("rdmsr" : EAX_EDX_RET(val, low, high) : "c" (msr));
	return EAX_EDX_VAL(val, low, high);
}

Thank you all!

I'm not sure this is so good. This bit looks particularly meaningless to me:

Code:

{
	DECLARE_ARGS(val, low, high);

Even if it did anything, where are val, low, and high coming from inside native_read_msr()?

And all these will do will do is return val:

Code:

#define DECLARE_ARGS(val, low, high)	unsigned long long val
#define EAX_EDX_VAL(val, low, high)	(val)

If the first one even compiles, since these are, in fact *NOT* typedefs.

It looks to me like an "abandoned" segment of code; eg. something that was left in by mistake (presuming that the whole thing actually works?).

There is a chance some asm guru could set me straight here about something I'm ignorant of, but I'm pretty sure this is just bad code.

Originally Posted by MK27

This bit looks particularly meaningless to me:

It means:

Code:

unsigned long long val;

Originally Posted by laserlight

It means:

Code:

unsigned long long val;

Exactly. So:

Code:

static inline unsigned long long native_read_msr(unsigned int msr)
{
           unsigned long long val;

That is incomplete. Maybe "purposeless" would have been more accurate than "meaningless" but hopefully the point is clear...

[edit] sorry I'm a little bit mistaken here...

Originally Posted by tabstop

What do you mean incomplete? It's a variable declaration. What exactly were you expecting?

Whoops! I just decided to ignore the "val" on the end.

Which makes that code dependent on this line:

Code:

asm volatile("rdmsr" : EAX_EDX_RET(val, low, high) : "c" (msr));

still, a strange, veritably obfuscated way to write IMO.

I am using GCC 4.2.4

gcc -E will send preprocessor output to your console. Be prepared for lots of output if you have header files. I usually redirect it to a new c file so that I can see it in a syntax colouring editor.

Edit:
Ohhhh, even better
gcc -c -save-temps foo.c
will save preprocessor output to foo.i

Well, if you had bothered to read the surrounding bits of code, you'd see that there is a #ifdef for CONFIG_X86_64 and #else for the #define's that you posted. Since the type of a 64-bit value is different in x86-64 and x86-32, the code inline assembler will have to deal with this fact, and using these macros allow it to achieve that without writing two complete functions.

For reference:
LXR linux/arch/x86/include/asm/msr.h (the quoted function is at line 42).

--
Mats

Originally Posted by MK27

Whoops! I just decided to ignore the "val" on the end.

So I guess there is a sense in which

Code:

#define DECLARE_ARGS(val, low, high)	unsigned long long val

would be useful during development, but either someone forgot to refine this later or they still have big hopes for the future.

It makes much more sense if you have ALL the relevant defines:

Code:

#ifdef CONFIG_X86_64
#define DECLARE_ARGS(val, low, high)    unsigned low, high
#define EAX_EDX_VAL(val, low, high)     ((low) | ((u64)(high) << 32))
#define EAX_EDX_ARGS(val, low, high)    "a" (low), "d" (high)
#define EAX_EDX_RET(val, low, high)     "=a" (low), "=d" (high)
#else
#define DECLARE_ARGS(val, low, high)    unsigned long long val
#define EAX_EDX_VAL(val, low, high)     (val)
#define EAX_EDX_ARGS(val, low, high)    "A" (val)
#define EAX_EDX_RET(val, low, high)     "=A" (val)
#endif

Since in x86_64, a long is 64 bits, and returned in ONE register (EAX), we need to make two unsigned (32-bit) values into a single 64-bit value. Meanwhile, in x86_32, each register holds 32 bits, so we need to return a 64-bit value as two registers (eax, edx). It just so happens also that msr's values are returned in eax, edx. The symbol A in inline assembler represents a 64-bit pair of EAX, EDX.

It is not something missing/extra in there, it's simply that the macros do not use all arguments earch time, instead they are using either val (as a 64-bit two-register integer) or low, high as two 32-bit values that get returned as a single register 64-bit value.

Which is a longer way of explaining what I said in the previous post. Is it because I posted a link to the source rather than typing it out that it was hard to understand this?

--
Mats

Originally Posted by matsp

Which is a longer way of explaining what I said in the previous post. Is it because I posted a link to the source rather than typing it out that it was hard to understand this?

No, I understand now matsp. Thanks!