2.5.61: x86_64 num_online_cpus() buglet?
Andi,
Kernel 2.5.61's include/asm-x86_64/smp.h contains:
extern unsigned long cpu_online_map;
...
extern inline unsigned int num_online_cpus(void)
{
return hweight32(cpu_online_map);
} and similarly for cpu_callout_map.
hweight32() truncates a 64-bit operand to 32-bits, so either
- the maps should be int rather than long, or
- x86_64 needs to define and use a new hweight64(), or
- CONFIG_NR_CPUS must not exceed 32 on x86_64.
Comments?
/Mikael
Quoted 11 lines · Mikael Pettersson
Andi,
Kernel 2.5.61's
include/asm-x86_64/smp.hcontains:extern unsigned long
cpu_online_map;
...
extern inline unsigned intnum_online_cpus(void){ return hweight32(cpu_online_map); }
Quoted 8 lines
and similarly for
cpu_callout_map.
hweight32()truncates a 64-bit operand to 32-bits, so either
- the maps should be int rather than long, or
-x86_64needs to define and use a newhweight64(), or
-CONFIG_NR_CPUSmust not exceed 32 onx86_64.Comments?
You're right - it should use hweight64. Thanks for the headup. I'll fix it.
Currently the x86-64 port is limited to 8 CPUs because the APIC drivers don't support cluster mode (yet) and can only talk to 8 local apics, so it isn't that harmful.
-Andi
Quoted 2 lines · Andi Kleen
You're right - it should use hweight64. Thanks for the headup. I'll fix it.
Can someone finally add a hweight64() to the generic code? I've been avoiding usage of hweight64() anywhere for this very reason.