|
The pcsample() function provides CPU-use statistics by profiling the amount of CPU time expended by a program.
For profiling dynamically-linked programs and 64-bit programs, it is superior to the profil(2)
function, which assumes that the entire program is contained in a small, contiguous segment of the address space, divides this segment into "bins", and on each clock tick increments the counter
in the bin where the program is currently executing. With shared libraries creating discontinuous program segments spread throughout the address space, and with 64-bit address spaces so large that the size
of "bins" would be measured in megabytes, the profil() function is of limited value.
The pcsample() function is passed an array samples containing nsamples pointer-sized elements. During program execution, the kernel
samples the program counter of the process, storing unadulterated values in the array on each clock tick. The kernel stops writing to the array when it is full, which occurs after nsamples
/ HZ seconds of process virtual time. The HZ value is obtained by invoking the call sysconf(_SC_CLK_TCK). See sysconf(3C).
The sampling can be stopped by a subsequent call to pcsample() with the nsamples argument set to 0. Like profil(), sampling continues
across a call to fork(2), but is disabled by a call to one of the exec family
of functions (see exec(2)). It is also disabled if an update of the samples[] array causes a memory fault.
|
