Re: [RFC PATCH 0/7] mm: Get rid of vmalloc_sync_(un)mappings()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, May 11, 2020 at 08:52:04AM -0700, Matthew Wilcox wrote:
> On Mon, May 11, 2020 at 09:31:34AM +0200, Peter Zijlstra wrote:
> > On Sat, May 09, 2020 at 06:11:57PM -0700, Matthew Wilcox wrote:
> > > Iterating an XArray (whether the entire thing
> > > or with marks) is RCU-safe and faster than iterating a linked list,
> > > so this should solve the problem?
> > 
> > It can hardly be faster if you want all elements -- which is I think the
> > case here. We only call into this if we change an entry, and then we
> > need to propagate that change to all.
> 
> Of course it can be faster.  Iterating an array is faster than iterating
> a linked list because caches.  While an XArray is a segmented array
> (so slower than a plain array), it's plainly going to be faster than
> iterating a linked list.

Quantifying this:

$ ./array-vs-list 
walked sequential array in 0.002039s
walked sequential list in 0.002807s
walked sequential array in 0.002017s
walked shuffled list in 0.102367s
walked shuffled array in 0.012114s

Attached is the source code; above results on a Kaby Lake with
CFLAGS="-O2 -W -Wall -g".
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <unistd.h>

unsigned long count = 1000 * 1000;
unsigned int verbose;

struct object {
	struct object *next;
	struct object *prev;
	unsigned int value;
};

#define printv(level, fmt, ...) \
	if (level <= verbose) { printf(fmt, ##__VA_ARGS__); }

void check_total(unsigned long total)
{
	if (total * 2 != count * (count + 1))
		printf("Check your staging! (%lu %lu)\n", total, count);
}

void alloc_objs(struct object **array)
{
	unsigned long i;

	for (i = 0; i < count; i++) {
		struct object *obj = malloc(sizeof(*obj));

		obj->value = i + 1;
		/* Add to the array */
		array[i] = obj;
	}
}

void shuffle(struct object **array, unsigned long seed)
{
	unsigned long i;

	printv(1, "random seed %lu\n", seed);
	srand48(seed);

	/* Shuffle the array */
	for (i = 1; i < count; i++) {
		struct object *obj;
		unsigned long j = (unsigned int)mrand48() % (i + 1);

		if (i == j)
			continue;
		obj = array[j];
		array[j] = array[i];
		array[i] = obj;
	}
}

void create_list(struct object **array, struct object *list)
{
	unsigned long i;

	list->next = list;
	list->prev = list;

	for (i = 0; i < count; i++) {
		struct object *obj = array[i];
		/* Add to the tail of the list */
		obj->next = list;
		obj->prev = list->prev;
		list->prev->next = obj;
		list->prev = obj;
	}
}

void walk_list(struct object *list)
{
	unsigned long total = 0;
	struct object *obj;

	for (obj = list->next; obj != list; obj = obj->next) {
		total += obj->value;
	}

	check_total(total);
}

void walk_array(struct object **array)
{
	unsigned long total = 0;
	unsigned long i;

	for (i = 0; i < count; i++) {
		total += array[i]->value;
	}

	check_total(total);
}

/* time2 - time1 */
double diff_time(struct timespec *time1, struct timespec *time2)
{
	double result = time2->tv_nsec - time1->tv_nsec;

	return time2->tv_sec - time1->tv_sec + result / 1000 / 1000 / 1000;
}

int main(int argc, char **argv)
{
	int opt;
	unsigned long seed = time(NULL);
	struct object **array;
	struct object list;
	struct timespec time1, time2;

	while ((opt = getopt(argc, argv, "c:s:v")) != -1) {
		if (opt == 'c')
			count *= strtoul(optarg, NULL, 0);
		else if (opt == 's')
			seed = strtoul(optarg, NULL, 0);
		else if (opt == 'v')
			verbose++;
	}

	clock_gettime(CLOCK_MONOTONIC, &time1);
	array = calloc(count, sizeof(void *));
	alloc_objs(array);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printv(1, "allocated %lu items in %fs\n", count,
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	walk_array(array);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printf("walked sequential array in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	create_list(array, &list);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printv(1, "created list in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	walk_list(&list);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printf("walked sequential list in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	walk_array(array);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printf("walked sequential array in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	shuffle(array, seed);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printv(1, "shuffled array in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	create_list(array, &list);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printv(1, "created list in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	walk_list(&list);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printf("walked shuffled list in %fs\n",
			diff_time(&time1, &time2));

	clock_gettime(CLOCK_MONOTONIC, &time1);
	walk_array(array);
	clock_gettime(CLOCK_MONOTONIC, &time2);
	printf("walked shuffled array in %fs\n",
			diff_time(&time1, &time2));

	return 0;
}

[Index of Archives]     [Linux IBM ACPI]     [Linux Power Management]     [Linux Kernel]     [Linux Laptop]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Video 4 Linux]     [Device Mapper]     [Linux Resources]

  Powered by Linux