On Mon, Aug 26, 2024 at 12:45:40PM +0200, Christophe Leroy wrote:

Le 26/08/2024 à 10:58, Jason A. Donenfeld a écrit :

quoted

On Mon, Aug 26, 2024 at 10:37:49AM +0200, Christophe Leroy wrote:

quoted

Le 26/08/2024 à 10:07, Jason A. Donenfeld a écrit :

quoted

On Thu, Aug 22, 2024 at 09:13:10AM +0200, Christophe Leroy wrote:

quoted

   
+#define _PAGE_SIZE (1UL << CONFIG_PAGE_SHIFT)
+#define _PAGE_MASK (~(_PAGE_SIZE - 1))

If PAGE_SIZE isn't defined at this point, why not just call it PAGE_SIZE
instead of _PAGE_SIZE? But if that's the case, why not put the vdso
definition of PAGE_SIZE into some vdso header included by this file?

It was working ok on powerpc but on x86 I got:

Seems like there might be some more fiddling to do, then? Or did you
conclude it's impossible?

Maybe someone who knows x86 in details could helps but after a first 
look I gave up because it looks very x86 specific, indeed that's 
x86/asm/vdso/gettimeofday.h that pulls several x86/asm/ headers , and 
the same type of issue might arise for any new architecture coming in.

For me it looked cleaner to just do as commit cffaefd15a8f ("vdso: Use 
CONFIG_PAGE_SHIFT in vdso/datapage.h") and not use PAGE_SIZE at all. But 
I didn't want to directly use (1UL << CONFIG_PAGE_SHIFT) and (~(1UL << 
(CONFIG_PAGE_SHIFT - 1))) in the code directly hence the new macros with 
a leading underscore to avoid any conflict with existing macros.

If that's the case and you can't sort it out or it's impossible, then it
seems like adding #ifndef PAGE_SIZE, #define... inside of some vdso header
(not inside getrandom.c) would make sense.

Jason