re: Adding clock_gettime1() to be shared with emulations ?

[matthew green <mrg%eterna.com.au@localhost>]

   On Thu, Apr 01, 2010 at 09:52:07AM +0200, Nicolas Joly wrote:
   [...]
   > The attached patch introduce a new function clock_getres1() function
   > which has most of the corresponding syscall logic except for the
   > copyout call.
   > 
   > This has the benefit to have all emulations can use it and are now
   > automatically synced with the native version (as a side effect, they
   > all gain CLOCK_MONOTONIC support i was looking for).
   > 
   > I'd like to commit it in a few days, unless someone objects ... And
   > then do mostly the same for clock_gettime().
   
   I was about to do the same for clock_gettime syscall, when i
   discovered the SoC project `Light weight precision user level time
   reading' which talks about optimising clock_gettime(2) :
   
     http://www.netbsd.org/contrib/soc-projects.html#fast-time
   
   While adding a clock_gettime1() function ensure that all the
   corresponding syscalls will be kept in sync; adding a function call
   here does not look like a good step towards optimisation.

IMO, these are orthogonal problems and you should make the
change to clock_gettime() you proposed.

we don't care about making emulated system calls faster more
than we care about them actually working...


.mrg.