lkml.org 
[lkml]   [2014]   [Feb]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH v12 9/9] Add 32 bit VDSO time support for 64 bit kernel
From
Date
Am Mittwoch, den 05.02.2014, 14:01 -0800 schrieb Andy Lutomirski:
> On Wed, Feb 5, 2014 at 12:20 AM, <stefani@seibold.net> wrote:
> > From: Stefani Seibold <stefani@seibold.net>
> >
> > This patch add the VDSO time support for the IA32 Emulation Layer.
> >
> > Due the nature of the kernel headers and the LP64 compiler where the
> > size of a long and a pointer differs against a 32 bit compiler, there
> > is some type hacking necessary.
> >
> > The vsyscall_gtod_data struture must be a little bit rearranged, to
> > serve 32- and 64-bit code access:
> >
> > - The seqcount_t was replaced by an unsigned, this makes the
> > vsyscall_gtod_data intedepend of kernel configuration and internal functions.
> > - The structure is now packed, so it can accessed from 32- und 64- bit
> > code at the same time.
> > - The inner struct clock was removed, to make packing of the while
> > struct easier.
> >
> > The "unsigned seq" would be handled by functions derivated from seqcount_t.
> >
> > Signed-off-by: Stefani Seibold <stefani@seibold.net>
> > ---
> > arch/x86/include/asm/vgtod.h | 20 +++---
> > arch/x86/kernel/vsyscall_gtod.c | 26 +++++--
> > arch/x86/vdso/vclock_gettime.c | 129 ++++++++++++++++++++++++----------
> > arch/x86/vdso/vdso32/vclock_gettime.c | 11 +++
> > include/uapi/linux/time.h | 2 +-
> > 5 files changed, 132 insertions(+), 56 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/vgtod.h b/arch/x86/include/asm/vgtod.h
> > index 46e24d3..2567b02 100644
> > --- a/arch/x86/include/asm/vgtod.h
> > +++ b/arch/x86/include/asm/vgtod.h
> > @@ -4,16 +4,18 @@
> > #include <asm/vsyscall.h>
> > #include <linux/clocksource.h>
> >
> > -struct vsyscall_gtod_data {
> > - seqcount_t seq;
> > +/*
> > + * vsyscall_gtod_data will be accessed by 32 and 64 bit code at the same time
> > + * so the structure must be packed
> > + */
> > +struct __attribute__((packed)) vsyscall_gtod_data {
> > + unsigned seq;
>
> Is that actually true? At least in the part you're changing,
> everything looks like it's aligned correctly.
>

Yes, i know. But for convenient this is less error prone when modifying
the structure. I can kick it out if you insist.

> It's almost certainly having some kind of BUILD_BUG_ON that will catch
> the case where this structure's size changes. I suspect that some
> kind of asm-offsets magic can be used for this.
>

This kind af ASM magic wan't work, because the code will be compiled
with -m32 for a 32 bit VDSO but will access a structure generated with a
64 bit compiler.

> >
> > - struct { /* extract of a clocksource struct */
> > - int vclock_mode;
> > - cycle_t cycle_last;
> > - cycle_t mask;
> > - u32 mult;
> > - u32 shift;
> > - } clock;
> > + int vclock_mode;
> > + cycle_t cycle_last;
> > + cycle_t mask;
> > + u32 mult;
> > + u32 shift;
> >
> > /* open coded 'struct timespec' */
> > time_t wall_time_sec;
> > diff --git a/arch/x86/kernel/vsyscall_gtod.c b/arch/x86/kernel/vsyscall_gtod.c
> > index 91862a4..ca48248 100644
> > --- a/arch/x86/kernel/vsyscall_gtod.c
> > +++ b/arch/x86/kernel/vsyscall_gtod.c
> > @@ -16,6 +16,18 @@
> >
> > DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data);
> >
> > +static inline void gtod_write_begin(unsigned *s)
> > +{
> > + ++*s;
> > + smp_wmb();
> > +}
> > +
> > +static inline void gtod_write_end(unsigned *s)
> > +{
> > + smp_wmb();
> > + ++*s;
> > +}
> > +
>
> Someone else should probably comment on the style for this. Maybe
> this should live in a header somewhere. IMO if it's called
> gtod_write_begin, it should take a pointer to gtod as a parameter.
>

Fixed in the next version.

>
> > +struct api_timeval {
> > + long tv_sec; /* seconds */
> > + long tv_usec; /* microseconds */
> > +};
> > +
> > +struct api_timespec {
> > + long tv_sec; /* seconds */
> > + long tv_nsec; /* nanoseconds */
> > +};
>
> Did you address my question about why there are two versions of this?
> Shouldn't it just match the userspace headers regardless of what the
> host kernel is?
>

Again, -m32 compilation. The timeval and timespec are __kernel_long_t,
which is defined as "long long" in a case of 32 bit VDSO for 64 bit
kernel.

The vsyscall_gtod_data use the kernel defined timeval and timespec, so
for 32 bit a conversation is needed.

The only way to prevent this kind of hack is to replace the struct
timespec and time_t members of vsyscall_gtod_data by a compiler independ
definition.

In this case it would be also necessary to have an own copy of
timespec_add_ns() function for the VDSO.

> You have *still* not responded to my objection to the unconditional fixmaps.
>

I have responded this objection on monday. If i enable the sysctl
interface i need this mapping too.

- Stefani




\
 
 \ /
  Last update: 2014-02-06 13:41    [W:0.055 / U:0.236 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site