Re: [PATCH RFC] allow constant folding in msecs_to_jiffies where possible for gcc

[Joe Perches]

On Thu, 2015-04-02 at 08:37 +0200, Nicholas Mc Guire wrote:
> A number of cleanup patches where switching var * HZ / 1000
> constructs to msecs_to_jiffies(var) to ensure that all corener
> cases are handled properly. The downside of this though is that
> it now uses a function call and also was not performing
> constant folding where it was originally possible.
> 
> msecs_to_jiffies() will calculate jiffies even if constants are
> passed in that could be handled by constant folding at compile time
> using __builtin_constant_p() gcc can optimize the constant case
> again.

OK, but how could this actually work?

This code isn't visible in a module at compile time, so how
could the compiler actually optimize the function call away?

Look at the disassembly for any object that uses
msecs_to_jiffies(<constant>).

For instance: drivers/bluetooth/hci_vhci.o

	schedule_delayed_work(&data->open_timeout, msecs_to_jiffies(1000));
 60d:	bf e8 03 00 00       	mov    $0x3e8,%edi
 612:	e8 00 00 00 00       	callq  617 <vhci_open+0xc7>
			613: R_X86_64_PC32	msecs_to_jiffies-0x4

You need to make this code either a macro or static inline
in jiffies.h, and for that the kernel/timeconst.h file needs
to be available.



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/