How are percpu pointers implemented in the Linux kernel? -

on multiprocessor, each core can have own variables. thought different variables in different addresses, although in same process , have same name.

but wondering, how kernel implement this? dispense piece of memory deposit percpu pointers, , every time redirects pointer address shift or something?

normal global variables not per cpu. automatic variables on stack, , different cpus use different stack, naturally separate variables.

i guess you're referring linux's per-cpu variable infrastructure.
of magic here (asm-generic/percpu.h):

extern unsigned long __per_cpu_offset[nr_cpus];  #define per_cpu_offset(x) (__per_cpu_offset[x])  /* separate out type, (int[3], foo) works. */ #define define_per_cpu(type, name) \     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name  /* var in discarded region: offset particular copy want */ #define per_cpu(var, cpu) (*reloc_hide(&per_cpu__##var, __per_cpu_offset[cpu])) #define __get_cpu_var(var) per_cpu(var, smp_processor_id()) 

the macro reloc_hide(ptr, offset) advances ptr given offset in bytes (regardless of pointer type).

what do?

1. when defining define_per_cpu(int, x), integer __per_cpu_x created in special .data.percpu section.
2. when kernel loaded, section loaded multiple times - once per cpu (this part of magic isn't in code above).
3. the __per_cpu_offset array filled distances between copies. supposing 1000 bytes of per cpu data used, __per_cpu_offset[n] contain 1000*n.
4. the symbol per_cpu__x relocated, during load, cpu 0's per_cpu__x.
5. __get_cpu_var(x), when running on cpu 3, translate *reloc_hide(&per_cpu__x, __per_cpu_offset[3]). starts cpu 0's x, adds offset between cpu 0's data , cpu 3's, , dereferences resulting pointer.