Re: [PATCH v13 8/9] Add 32 bit VDSO time support for 32 bit kernel

From: Stefani Seibold
Date: Sun Feb 09 2014 - 03:11:01 EST


Am Samstag, den 08.02.2014, 21:22 -0800 schrieb Andi Kleen:
> > +notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
> > +{
> > + long ret;
> > +
> > + asm(
> > + "push %%ebx \n"
> > + "mov %2,%%ebx \n"
> > + "call VDSO32_vsyscall \n"
> > + "pop %%ebx \n"
> > + : "=a" (ret)
> > + : "0" (__NR_clock_gettime), "d" (clock), "c" (ts)
> > + : "memory");
>
> The push/mov/pop could be just replaced with "b"?

EBX is a reserved as the base register when compiling with -fPIC.
Otherwise it get a "error: can't find a register in class 'BREG' while
reloading 'asm'".

> Also I if this calls syscall.S:__kernel_vsyscall it clobbers more registers
> (ecx at least?) and you need to describe that to gcc with a +

Sine ECX is in use as the second parameter it is still marked as
clobbered.

> Same in the other sys calls.
>

I think it should word... As i know a system call will preserve all
register expect the EAX, which is marked as in and out for this asm()
directive.

> > @@ -310,6 +311,11 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
> > unsigned long addr;
> > int ret = 0;
> > bool compat;
> > + struct vm_area_struct *vma;
> > + extern char __vvar_page;
> > +#ifdef CONFIG_HPET_TIMER
> > + extern unsigned long hpet_address;
> > +#endif
>
> Put this into a header.
>

Okay...

> > +++ b/arch/x86/vdso/vdso32/vclock_gettime.c
> > @@ -0,0 +1,16 @@
> > +#define BUILD_VDSO32
> > +
> > +#ifdef CONFIG_X86_64
> > +
> > +#define _ASM_X86_PAGE_H
> > +
> > +#define __pa(x) 0
> > +#define __va(x) 0
> > +
> > +#undef CONFIG_ILLEGAL_POINTER_VALUE
> > +#define CONFIG_ILLEGAL_POINTER_VALUE 0
>
> This needs some comments.
>
> Why are the __pa/__va defines safe?
>
> The rest of the patchkit looks reasonable to me from a quick read.
> Much better than the first version.
>

This is due the -m32 compilation, i will get a log of
"warning: integer constant is too large for 'unsigned long' type",
because a unsigned long is only 32 bit.

Since the vclocktime.c use a lot of kernel header a little bit hacking
is necessary, because the CONFIG_X86_64 will attract a lot of x86-64
headers, which are not compatible with 32 bit compilation.

The _ASM_X86_PAGE_H prevents the include of arch/x86/include/asm/page.h.

But the inline function phys_to_virt() in arch/x86/include/asm/io.h
depends on the __pa() and __va() functions, which comes from
arch/x86/include/asm/page.h. So i need two dummys for this.

Since non of this is functions is used in arch/x86/vdso/vclock_gettime.c
it is save.

I have no idea how to fix this weaved include file web without change a
lot in many include files.

The define of CONFIG_ILLEGAL_POINTER_VALUE is also to prevent the
""warning: integer constant is too large...".

- Stefani

--
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/