Re: [PATCH] x86: convert x86_platform_ops to timespec64

[Arnd Bergmann]

On Fri, Oct 13, 2017 at 10:45 PM, Boris Ostrovsky
<boris.ostrovsky@xxxxxxxxxx> wrote:
> On 10/13/2017 02:37 PM, Arnd Bergmann wrote:
>> The x86 platform operations are fairly isolated, so we can
>> change them from using timespec to timespec64. I checked that
>> All the users and callers are safe, and there is only one
>> critical function that is broken beyond 2106:
>>
>> pvclock_read_wallclock() uses a 32-bit number of seconds since
>> the epoch to communicate the boot time between host and guest
>> in a virtual environment. This will work until 2106, but we
>> should ideally find a replacement anyway. I've added a comment
>> about it there.
>>
>> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
>> ---
>>  arch/x86/include/asm/intel_mid_vrtc.h        |  4 ++--
>>  arch/x86/include/asm/mc146818rtc.h           |  4 ++--
>>  arch/x86/include/asm/pvclock.h               |  2 +-
>>  arch/x86/include/asm/x86_init.h              |  6 +++---
>>  arch/x86/kernel/kvmclock.c                   |  4 ++--
>>  arch/x86/kernel/pvclock.c                    | 12 +++++++++---
>>  arch/x86/kernel/rtc.c                        | 16 ++++++++--------
>>  arch/x86/platform/intel-mid/intel_mid_vrtc.c | 10 +++++-----
>>  arch/x86/xen/time.c                          | 10 +++++-----
>>  9 files changed, 37 insertions(+), 31 deletions(-)
>
> Xen bits:
> Reviewed-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>

Thanks!

Since you've looked at it overall, do you have an opinion on the question
how to fix the PV interface to deal with the pvclock_wall_clock overflow?

Should we add a new version now and deprecate the existing one, or
do you think that y2106 is far enough out that we should just ignore the
problem?

> with a couple of nits:
>
>> @@ -136,11 +136,17 @@ void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
>>               rmb();          /* fetch time before checking version */
>>       } while ((wall_clock->version & 1) || (version != wall_clock->version));
>>
>> +     /*
>> +      * Note: wall_clock->sec is a u32 value, so it can only store dates
>> +      * between 1970 and 2106. To allow times beyond that, we need to
>> +      * create a new hypercall interface with an extended pvclock_wall_clock
>> +      * structure like ARM has.
>> +      */
>
> I think this comment block should be moved up above 'now.tv_sec  =
> wall_clock->sec;'

right, changed.

>>       delta = pvclock_clocksource_read(vcpu_time);    /* time since system boot */
>>       delta += now.tv_sec * (u64)NSEC_PER_SEC + now.tv_nsec;
>
> Now that tv_sec is a 64-bit quantity the cast can be dropped.

Ok dropped. In the meantime I had noticed two more problems with the
patch that I did not see earlier when I tested with another patch applied
as well. The kbuild test robot reported the exact same problems, and I've
done a few hundred randconfig builds without the other patch now, so
I'm fairly confident that there are no other problems like those.

I'll follow up with a v2 patch soon.

       Arnd