Re: [LinuxBIOS] Is there sched_clock equivalent in LB?

[Lucas C. Villa Real]

On 4/27/06, Eric Poulsen  wrote:
> Daniel Parthey wrote:
> > Eric Poulsen schrieb:
> > > I was going to hack LB a bit, and add a timestamp, just like you see in
> > > the newer Linux kernels, to look for potential bottlenecks specific to
> > > my MLB. The kernel printk uses sched_clock() to get the current time.
> > > Is there an equivalent call in LB?  I've looked through the code base
> > > for files / code that has *clock* or *time* or something similar, but
> > > without much luck.
> >
> > I suppose MLB means VIA EPIA ML Board?
>
> I had meant Main Logic Board, but I guess it works both ways.
> > On Intel compatible Pentium (or higher) boards like VIA EPIA, you could
> > read the Time Stamp Counter (TSC) CPU register which tells you the number
> > of cycles from the last machine reset. Use the "cpuid" assembler command
> > as a barrier (to prevent out-of-order command execution) in combination
> > with the "rdtsc" assembler command to read the time time stamp counter.
> > The linux kernel sourcecode contains examples for this.
>
> My Intel assembly is 15 years rusty -- I'd have better luck doing this
> on a PIC.  My co-worker hacks Intel assembly -- I'll get him to help me.

I'm not yet started at LB development, but this sample code can
probably do the job for you, given that you're using GCC:

#define do_rdtsc(ticks) \
     __asm__ volatile (".byte 0x0f, 0x31" : "=A" (ticks));

uint64_t current_ticks;
   do_rdtsc(current_ticks);

This instruction returns a 64-bit value in EDX:EAX. If you don't need
to use 64 bits, getting the lower 32 bits will require a change on
"=A" by "=a" and on uint64_t by uint32_t.

Hope that helps,