Issue
Apart from clean way to write code, is there any other benefit (specially from performance point of view) to use kzalloc()
/ kmem_cache_zalloc()
as compare to kmalloc()
/ kmem_cache_alloc()
+ memset()
?
I tried to measure their performance, but results are almost same.
Kernel version: 4.18.0-372.19.1.el8_6.x86_64
Below code is creating 2 million slab objects each of 4096 bytes:
#define DATA_NUMBER 2000000
char *data[DATA_NUMBER];
/* kthread */
static int zero_slab(void *arg)
{
unsigned int counter, i;
struct kmem_cache *mp_cache;
unsigned long before, after;
mp_cache = kmem_cache_create("MP_CACHE", PAGE_SIZE, 0, 0, NULL);
if (!mp_cache) {
printk(KERN_INFO "Failed to create slab.\n");
return -ENOMEM;
}
printk(KERN_INFO "Slab MP_CACHE is created.\n");
before = ktime_get();
for (counter = 0; counter < DATA_NUMBER; counter++) {
// data[counter] = kmem_cache_zalloc(mp_cache, GFP_KERNEL);
data[counter] = kmem_cache_alloc(mp_cache, GFP_KERNEL);
if (!data[counter])
goto err_out;
memset(data[counter], 0, PAGE_SIZE);
}
after = ktime_get() - before;
printk(KERN_INFO "Time taken in ns (%lu)\n", after);
err_out:
printk(KERN_INFO "Total objects : %u\n", counter);
for (i = 0; i < counter; ++i)
kmem_cache_free(mp_cache, data[i]);
kmem_cache_destroy(mp_cache);
return 0;
}
Solution
Apart from clean way to write code, is there any other benefit
Nope. That's it. kzalloc
is simply a wrapper around kmalloc
that passes __GFP_ZERO
. That flag tells kmalloc
to do memset
for you before returning. There is no performance difference between kmalloc(x, GFP_KERNEL) + memset()
and kmalloc(x, GFP_KERNEL|__GFP_ZERO)
.
Answered By - Marco Bonelli Answer Checked By - Mildred Charles (WPSolving Admin)