Thanks for the reply Richard!
Agree, this is just a short-term workaround. Maybe we can add _TIME_BITS=64 in 
our makefile for all?
https://sourceware.org/glibc/wiki/Y2038ProofnessDesign

-----Original Message-----
From: Richard Cochran <[email protected]> 
Sent: Thursday, May 9, 2019 7:13 AM
To: Vincent Li X <[email protected]>
Cc: [email protected]
Subject: Re: [Linuxptp-devel] Year 2038 problem on 32-bit machine

On Tue, Apr 23, 2019 at 05:25:54PM +0000, Vincent Li X wrote:
> We realized on our machine, the timespec.time_t (long int) is of 4 bytes, 
> which would cause overflow when converting to internal tmv_t, if the second 
> value is bigger than 2pow(31)-1.
> https://en.wikipedia.org/wiki/Year_2038_problem
> 
> How do we solve this problem in general?

You will have to re-compile your 32 userland with a new gnu libc.

> Would below conversion to unsigned is a proper fix?

No, that doesn't help at all.

> diff --git a/tmv.h b/tmv.h
> index cca4da7..2e34ff4 100644
> --- a/tmv.h
> +++ b/tmv.h
> @@ -139,7 +139,7 @@ static inline struct Timestamp 
> tmv_to_Timestamp(tmv_t x) static inline tmv_t timespec_to_tmv(struct 
> timespec ts) {
>         tmv_t t;
> -       t.ns = ts.tv_sec * NS_PER_SEC + ts.tv_nsec;
> +       t.ns = (unsigned long int)ts.tv_sec * NS_PER_SEC + ts.tv_nsec;
>         return t;
> }

Just casting doesn't magically add the missing bits!

Sorry,
Richard


_______________________________________________
Linuxptp-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linuxptp-devel

Reply via email to