How to execute a piece of kernel code on all CPUs?

I'm trying to make a kernel module to enable FOP compatibility mode for x87 FPU. This is done via setting bit 2 in IA32_MISC_ENABLE MSR. Here's the code:

#include <linux/init.h>
#include <linux/module.h>
#include <linux/kernel.h>
#include <asm/msr-index.h>
#include <asm/msr.h>

MODULE_LICENSE("GPL");
MODULE_AUTHOR("10110111");
MODULE_DESCRIPTION("Module to enable FOPcode compatibility mode");
MODULE_VERSION("0.1");

static int __init fopCompat_init(void)
{
   unsigned long long misc_enable=native_read_msr(MSR_IA32_MISC_ENABLE);
   printk(KERN_INFO "Before trying to set FOP_COMPAT, IA32_MISC_ENABLE=%llx,"
                    " i.e. FOP_COMPAT is %senabled\n"
                    ,misc_enable,misc_enable&MSR_IA32_MISC_ENABLE_X87_COMPAT?"":"NOT ");

   wrmsrl(MSR_IA32_MISC_ENABLE,misc_enable|MSR_IA32_MISC_ENABLE_X87_COMPAT);
   misc_enable=native_read_msr(MSR_IA32_MISC_ENABLE);

   printk(KERN_INFO "Tried to set FOP_COMPAT. Result: IA32_MISC_ENABLE=%llx,"
                    " i.e. FOP_COMPAT is now %senabled\n"
                    ,misc_enable,misc_enable&MSR_IA32_MISC_ENABLE_X87_COMPAT?"":"NOT ");
   return 0;
}

static void __exit fopCompat_exit(void)
{
   const unsigned long long misc_enable=native_read_msr(MSR_IA32_MISC_ENABLE);
   printk(KERN_INFO "Quitting FOP-compat with IA32_MISC_ENABLE=%llx\n",misc_enable);
   if(!(misc_enable & MSR_IA32_MISC_ENABLE_X87_COMPAT))
       printk(KERN_INFO "NOTE: seems some CPUs still have to be set up, "
                        "or compatibility mode will work inconsistently\n");
   printk(KERN_INFO "\n");
}

module_init(fopCompat_init);
module_exit(fopCompat_exit);

It seems to work, but on multiple insmod/rmmod cycles I sometimes get dmesg output that the compatibility mode wasn't still enabled, although it was immediately after doing wrmsr. After some thinking I realized that it's because the module code was executed on different logical CPUs (I have Core i7 with 4 cores*HT=8 logical CPUs), so I had a 1/8 chance of getting "enabled" print on doing rmmod. After repeating the cycle for about 20 times I got consistent "enabled" prints, and my userspace application happily works with it.

So now my question is: how can I make my code execute on all logical CPUs present on the system, so as to enable compatibility mode for all of them?

Solution

For execute code on every CPU use on_each_cpu function.

Signature:

int on_each_cpu(void (*func) (void *info), void *info, int wait)

Description:

Call a function on all processors.

If wait parameter is non-zero, it waits for function's completion on all CPUs.

Function func shouldn't sleep, but whole on_each_cpu() call shouldn't be done in atomic context.