From: Sarthak Sharma <sarthak.sharma@arm.com>
To: Andrew Morton <akpm@linux-foundation.org>,
David Hildenbrand <david@kernel.org>
Cc: Jason Gunthorpe <jgg@ziepe.ca>,
John Hubbard <jhubbard@nvidia.com>, Peter Xu <peterx@redhat.com>,
Lorenzo Stoakes <ljs@kernel.org>,
"Liam R . Howlett" <Liam.Howlett@oracle.com>,
Vlastimil Babka <vbabka@kernel.org>,
Mike Rapoport <rppt@kernel.org>,
Suren Baghdasaryan <surenb@google.com>,
Michal Hocko <mhocko@suse.com>, Shuah Khan <shuah@kernel.org>,
linux-mm@kvack.org, linux-kselftest@vger.kernel.org,
linux-kernel@vger.kernel.org,
Sarthak Sharma <sarthak.sharma@arm.com>
Subject: [PATCH 1/2] tools/mm: add a standalone GUP microbenchmark
Date: Fri, 15 May 2026 14:18:39 +0530 [thread overview]
Message-ID: <20260515084840.174652-2-sarthak.sharma@arm.com> (raw)
In-Reply-To: <20260515084840.174652-1-sarthak.sharma@arm.com>
Add a command-line tool for benchmarking get_user_pages fast-path
(GUP_FAST), pin_user_pages fast-path (PIN_FAST), and pin_user_pages
longterm (PIN_LONGTERM) via the CONFIG_GUP_TEST debugfs interface.
When invoked without arguments, gup_bench runs the same matrix of
configurations as run_gup_matrix() in run_vmtests.sh: all three GUP
commands across read/write, private/shared mappings, and a range of
page counts, with THP on/off for regular mappings and hugetlb for huge
page mappings.
This tool is a mix of reused and new logic. The mapping/setup path comes
from selftests/mm/gup_test.c, while the default benchmark matrix matches
run_gup_matrix() in run_vmtests.sh. The standalone CLI and tools/mm
integration are added here so tools/mm does not depend on kselftest.
Add gup_bench to BUILD_TARGETS and INSTALL_TARGETS in tools/mm/Makefile,
and ignore the resulting binary in tools/mm/.gitignore. While here, also
add the missing thp_swap_allocator_test entry to .gitignore.
Add tools/mm/gup_bench.c to the GUP entry in MAINTAINERS.
Suggested-by: David Hildenbrand (Arm) <david@kernel.org>
Signed-off-by: Sarthak Sharma <sarthak.sharma@arm.com>
---
MAINTAINERS | 1 +
tools/mm/.gitignore | 2 +
tools/mm/Makefile | 6 +-
tools/mm/gup_bench.c | 491 +++++++++++++++++++++++++++++++++++++++++++
4 files changed, 497 insertions(+), 3 deletions(-)
create mode 100644 tools/mm/gup_bench.c
diff --git a/MAINTAINERS b/MAINTAINERS
index 98d0a7a1c689..c91165b9280e 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -16830,6 +16830,7 @@ T: git git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
F: mm/gup.c
F: mm/gup_test.c
F: mm/gup_test.h
+F: tools/mm/gup_bench.c
F: tools/testing/selftests/mm/gup_longterm.c
F: tools/testing/selftests/mm/gup_test.c
diff --git a/tools/mm/.gitignore b/tools/mm/.gitignore
index 922879f93fc8..154d740be02e 100644
--- a/tools/mm/.gitignore
+++ b/tools/mm/.gitignore
@@ -2,3 +2,5 @@
slabinfo
page-types
page_owner_sort
+thp_swap_allocator_test
+gup_bench
diff --git a/tools/mm/Makefile b/tools/mm/Makefile
index f5725b5c23aa..8e4db797a17a 100644
--- a/tools/mm/Makefile
+++ b/tools/mm/Makefile
@@ -3,13 +3,13 @@
#
include ../scripts/Makefile.include
-BUILD_TARGETS=page-types slabinfo page_owner_sort thp_swap_allocator_test
+BUILD_TARGETS=page-types slabinfo page_owner_sort thp_swap_allocator_test gup_bench
INSTALL_TARGETS = $(BUILD_TARGETS) thpmaps
LIB_DIR = ../lib/api
LIBS = $(LIB_DIR)/libapi.a
-CFLAGS += -Wall -Wextra -I../lib/ -pthread
+CFLAGS += -Wall -Wextra -I../lib/ -I../.. -pthread
LDFLAGS += $(LIBS) -pthread
all: $(BUILD_TARGETS)
@@ -23,7 +23,7 @@ $(LIBS):
$(CC) $(CFLAGS) -o $@ $< $(LDFLAGS)
clean:
- $(RM) page-types slabinfo page_owner_sort thp_swap_allocator_test
+ $(RM) page-types slabinfo page_owner_sort thp_swap_allocator_test gup_bench
make -C $(LIB_DIR) clean
sbindir ?= /usr/sbin
diff --git a/tools/mm/gup_bench.c b/tools/mm/gup_bench.c
new file mode 100644
index 000000000000..2806ee0d7453
--- /dev/null
+++ b/tools/mm/gup_bench.c
@@ -0,0 +1,491 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Microbenchmark for get_user_pages (GUP) kernel interfaces.
+ *
+ * Exercises GUP_FAST_BENCHMARK, PIN_FAST_BENCHMARK, and
+ * PIN_LONGTERM_BENCHMARK via the CONFIG_GUP_TEST debugfs interface.
+ *
+ * Example use:
+ * # Run the full matrix (all commands, access modes, page counts):
+ * ./gup_bench
+ *
+ * # Single run: pin_user_pages_fast, 512 pages, write access, hugetlb:
+ * ./gup_bench -a -n 512 -w -H
+ *
+ * Requires CONFIG_GUP_TEST=y and debugfs mounted at /sys/kernel/debug.
+ * Must be run as root.
+ */
+
+#define __SANE_USERSPACE_TYPES__ // Use ll64
+#include <fcntl.h>
+#include <errno.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <dirent.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <pthread.h>
+#include <assert.h>
+#include <stdbool.h>
+#include <stdatomic.h>
+#include <limits.h>
+#include <mm/gup_test.h>
+#include <string.h>
+
+#define MB (1UL << 20)
+
+#ifndef ARRAY_SIZE
+#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
+#endif
+
+/* Just the flags we need, copied from the kernel internals. */
+#define FOLL_WRITE 0x01 /* check pte is writable */
+
+#define GUP_TEST_FILE "/sys/kernel/debug/gup_test"
+
+/*
+ * Local HugeTLB setup helpers for gup_bench.
+ *
+ * These helpers were copied from tools/testing/selftests/mm/ and adjusted to
+ * remove the ksft formatting. Keep this copy local so tools/mm does not
+ * depend on ksft output behavior.
+ */
+
+static unsigned int psize(void)
+{
+ static unsigned int __page_size;
+
+ if (!__page_size)
+ __page_size = sysconf(_SC_PAGESIZE);
+ return __page_size;
+}
+
+static unsigned long default_huge_page_size(void)
+{
+ FILE *f = fopen("/proc/meminfo", "r");
+ unsigned long hpage_size = 0;
+ char buf[256];
+
+ if (!f)
+ return 0;
+ while (fgets(buf, sizeof(buf), f)) {
+ if (sscanf(buf, "Hugepagesize: %lu kB", &hpage_size) == 1)
+ break;
+ }
+ fclose(f);
+ hpage_size <<= 10;
+ return hpage_size;
+}
+
+static void hugetlb_sysfs_path(char *buf, size_t buflen,
+ unsigned long size, const char *attr)
+{
+ snprintf(buf, buflen, "/sys/kernel/mm/hugepages/hugepages-%lukB/%s",
+ size / 1024, attr);
+}
+
+static unsigned long hugetlb_read_num(const char *path)
+{
+ char buf[32];
+ FILE *f = fopen(path, "r");
+ unsigned long val = 0;
+
+ if (!f)
+ return 0;
+ if (fgets(buf, sizeof(buf), f))
+ val = strtoul(buf, NULL, 10);
+ fclose(f);
+ return val;
+}
+
+static void hugetlb_write_num(const char *path, unsigned long num)
+{
+ FILE *f = fopen(path, "w");
+
+ if (!f)
+ return;
+ fprintf(f, "%lu\n", num);
+ fclose(f);
+}
+
+static unsigned long hugetlb_nr_pages(unsigned long size)
+{
+ char path[PATH_MAX];
+
+ hugetlb_sysfs_path(path, sizeof(path), size, "nr_hugepages");
+ return hugetlb_read_num(path);
+}
+
+static void hugetlb_set_nr_pages(unsigned long size, unsigned long nr)
+{
+ char path[PATH_MAX];
+
+ hugetlb_sysfs_path(path, sizeof(path), size, "nr_hugepages");
+ hugetlb_write_num(path, nr);
+}
+
+static unsigned long hugetlb_free_pages(unsigned long size)
+{
+ char path[PATH_MAX];
+
+ hugetlb_sysfs_path(path, sizeof(path), size, "free_hugepages");
+ return hugetlb_read_num(path);
+}
+
+/* Saved pool size to restore on exit */
+static unsigned long hugetlb_saved_nr;
+static unsigned long hugetlb_saved_size;
+
+static void hugetlb_restore_atexit(void)
+{
+ if (hugetlb_saved_size)
+ hugetlb_set_nr_pages(hugetlb_saved_size, hugetlb_saved_nr);
+}
+
+static bool __hugetlb_setup(unsigned long size, unsigned long nr)
+{
+ unsigned long free = hugetlb_free_pages(size);
+ unsigned long total = hugetlb_nr_pages(size);
+
+ if (free >= nr)
+ return true;
+
+ hugetlb_set_nr_pages(size, total + (nr - free));
+
+ return hugetlb_free_pages(size) >= nr;
+}
+
+static bool hugetlb_setup_default(unsigned long nr)
+{
+ unsigned long hsize = default_huge_page_size();
+
+ if (!hsize)
+ return false;
+
+ /* Save current pool so we can restore it on exit (only on first call) */
+ if (!hugetlb_saved_size) {
+ hugetlb_saved_size = hsize;
+ hugetlb_saved_nr = hugetlb_nr_pages(hsize);
+ atexit(hugetlb_restore_atexit);
+ }
+
+ return __hugetlb_setup(hsize, nr);
+}
+
+static unsigned long cmd;
+static const char *bench_label;
+static int gup_fd, repeats = 1;
+static unsigned long size = 128 * MB;
+static atomic_int bench_error;
+/* Serialize prints */
+static pthread_mutex_t print_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+static const unsigned long bench_cmds[] = {
+ GUP_FAST_BENCHMARK,
+ PIN_FAST_BENCHMARK,
+ PIN_LONGTERM_BENCHMARK,
+};
+static const int bench_thp_modes[] = { 1, 0 }; /* on, off */
+static const int bench_nr_pages_list[] = { 1, 512, 123, -1 };
+
+static const char *cmd_to_str(unsigned long cmd)
+{
+ switch (cmd) {
+ case GUP_FAST_BENCHMARK:
+ return "GUP_FAST_BENCHMARK";
+ case PIN_FAST_BENCHMARK:
+ return "PIN_FAST_BENCHMARK";
+ case PIN_LONGTERM_BENCHMARK:
+ return "PIN_LONGTERM_BENCHMARK";
+ }
+ return "Unknown command";
+}
+
+struct bench_run {
+ unsigned long cmd;
+ int thp; /* -1: default, 0: off, 1: on */
+ bool hugetlb;
+ bool write;
+ bool shared;
+ int nr_pages; /* -1 means all pages (size / psize()) */
+ unsigned long size;
+ char *file;
+ int nthreads;
+ unsigned int gup_flags;
+};
+
+void *gup_thread(void *data)
+{
+ struct gup_test gup = *(struct gup_test *)data;
+ int i, status;
+
+ for (i = 0; i < repeats; i++) {
+ gup.size = size;
+ status = ioctl(gup_fd, cmd, &gup);
+ if (status) {
+ bench_error = 1;
+ break;
+ }
+
+ pthread_mutex_lock(&print_mutex);
+ printf("%s time: get:%lld put:%lld us",
+ bench_label, gup.get_delta_usec,
+ gup.put_delta_usec);
+ if (gup.size != size)
+ printf(", truncated (size: %lld)", gup.size);
+ printf("\n");
+ pthread_mutex_unlock(&print_mutex);
+ }
+
+ return NULL;
+}
+
+static int run_bench(struct bench_run *run)
+{
+ struct gup_test gup = { 0 };
+ int zero_fd, i, ret, started_threads = 0;
+ int flags = MAP_PRIVATE;
+ pthread_t *tid;
+ char label[128];
+ char *p;
+
+ /* Set globals consumed by gup_thread */
+ cmd = run->cmd;
+ size = run->size;
+ bench_error = 0;
+
+ if (run->hugetlb) {
+ unsigned long hp_size = default_huge_page_size();
+
+ if (!hp_size) {
+ fprintf(stderr, "Could not determine huge page size\n");
+ return 1;
+ }
+ size = (size + hp_size - 1) & ~(hp_size - 1);
+ if (!hugetlb_setup_default(size / hp_size)) {
+ fprintf(stderr, "Not enough huge pages\n");
+ return 1;
+ }
+ flags |= (MAP_HUGETLB | MAP_ANONYMOUS);
+ }
+
+ if (run->shared) {
+ flags &= ~MAP_PRIVATE;
+ flags |= MAP_SHARED;
+ }
+
+ gup.nr_pages_per_call = run->nr_pages < 0 ? size / psize() :
+ (unsigned long)run->nr_pages;
+
+ gup.gup_flags = run->gup_flags;
+ if (run->write)
+ gup.gup_flags |= FOLL_WRITE;
+
+ snprintf(label, sizeof(label), "%s (nr_pages=%-4u %s %s %s %s)",
+ cmd_to_str(run->cmd),
+ gup.nr_pages_per_call,
+ run->write ? "write" : "read",
+ run->shared ? "shared" : "private",
+ run->hugetlb ? "hugetlb=on" : "hugetlb=off",
+ run->hugetlb ? "thp=off" :
+ (run->thp == 1 ? "thp=on" :
+ (run->thp == 0 ? "thp=off" : "thp=default")));
+ bench_label = label;
+
+ zero_fd = open(run->file, O_RDWR);
+ if (zero_fd < 0) {
+ fprintf(stderr, "Unable to open %s: %s\n", run->file, strerror(errno));
+ return 1;
+ }
+
+ p = mmap(NULL, size, PROT_READ | PROT_WRITE, flags, zero_fd, 0);
+ close(zero_fd);
+ if (p == MAP_FAILED) {
+ fprintf(stderr, "mmap: %s\n", strerror(errno));
+ return 1;
+ }
+ gup.addr = (unsigned long)p;
+
+ if (run->thp == 1)
+ madvise(p, size, MADV_HUGEPAGE);
+ else if (run->thp == 0)
+ madvise(p, size, MADV_NOHUGEPAGE);
+
+ /* Fault them in here, from user space. */
+ for (; (unsigned long)p < gup.addr + size; p += psize())
+ p[0] = 0;
+
+ tid = malloc(sizeof(pthread_t) * run->nthreads);
+ assert(tid);
+ for (i = 0; i < run->nthreads; i++) {
+ ret = pthread_create(&tid[i], NULL, gup_thread, &gup);
+ if (ret) {
+ fprintf(stderr, "pthread_create failed: %s\n", strerror(ret));
+ bench_error = 1;
+ break;
+ }
+ started_threads++;
+ }
+ for (i = 0; i < started_threads; i++) {
+ ret = pthread_join(tid[i], NULL);
+ if (ret) {
+ fprintf(stderr, "pthread_join failed: %s\n", strerror(ret));
+ bench_error = 1;
+ }
+ }
+
+ free(tid);
+ munmap((void *)gup.addr, size);
+
+ return bench_error ? 1 : 0;
+}
+
+static int run_matrix(void)
+{
+ unsigned int c, t, w, s, n;
+ int ret = 0;
+
+ for (c = 0; c < ARRAY_SIZE(bench_cmds); c++) {
+ for (w = 0; w <= 1; w++) {
+ for (s = 0; s <= 1; s++) {
+ for (t = 0; t < ARRAY_SIZE(bench_thp_modes); t++) {
+ for (n = 0; n < ARRAY_SIZE(bench_nr_pages_list); n++) {
+ struct bench_run run = {
+ .cmd = bench_cmds[c],
+ .thp = bench_thp_modes[t],
+ .hugetlb = false,
+ .write = w,
+ .shared = s,
+ .nr_pages = bench_nr_pages_list[n],
+ .size = 128 * MB,
+ .file = "/dev/zero",
+ .nthreads = 1,
+ };
+ ret |= run_bench(&run);
+ }
+ }
+ /* hugetlb: 256M to match run_gup_matrix() in run_vmtests.sh */
+ for (n = 0; n < ARRAY_SIZE(bench_nr_pages_list); n++) {
+ struct bench_run run = {
+ .cmd = bench_cmds[c],
+ .thp = -1,
+ .hugetlb = true,
+ .write = w,
+ .shared = s,
+ .nr_pages = bench_nr_pages_list[n],
+ .size = 256 * MB,
+ .file = "/dev/zero",
+ .nthreads = 1,
+ };
+ ret |= run_bench(&run);
+ }
+ }
+ }
+ }
+ return ret;
+}
+
+int main(int argc, char **argv)
+{
+ struct bench_run run = {
+ .cmd = GUP_FAST_BENCHMARK,
+ .thp = -1,
+ .hugetlb = false,
+ .write = true,
+ .shared = false,
+ .nr_pages = 1,
+ .size = 128 * MB,
+ .file = "/dev/zero",
+ .nthreads = 1,
+ };
+ int opt, result;
+
+ while ((opt = getopt(argc, argv, "m:r:n:F:f:aj:tTLuwWSH")) != -1) {
+ switch (opt) {
+
+ /* Command selection */
+ case 'u':
+ run.cmd = GUP_FAST_BENCHMARK;
+ break;
+ case 'a':
+ run.cmd = PIN_FAST_BENCHMARK;
+ break;
+ case 'L':
+ run.cmd = PIN_LONGTERM_BENCHMARK;
+ break;
+
+ /* Memory type */
+ case 'H':
+ run.hugetlb = true;
+ break;
+ case 't':
+ run.thp = 1;
+ break;
+ case 'T':
+ run.thp = 0;
+ break;
+
+ /* Access mode */
+ case 'w':
+ run.write = true;
+ break;
+ case 'W':
+ run.write = false;
+ break;
+ case 'S':
+ run.shared = true;
+ break;
+
+ /* Mapping */
+ case 'f':
+ run.file = optarg;
+ break;
+
+ /* Sizing and iteration */
+ case 'm':
+ run.size = atoi(optarg) * MB;
+ break;
+ case 'n':
+ run.nr_pages = atoi(optarg);
+ break;
+ case 'r':
+ repeats = atoi(optarg);
+ break;
+ case 'j':
+ run.nthreads = atoi(optarg);
+ break;
+
+ /* Advanced */
+ case 'F':
+ /* strtol, so you can pass flags in hex form */
+ run.gup_flags = strtol(optarg, 0, 0);
+ break;
+
+ default:
+ fprintf(stderr, "Wrong argument\n");
+ exit(1);
+ }
+ }
+
+ gup_fd = open(GUP_TEST_FILE, O_RDWR);
+ if (gup_fd == -1) {
+ if (errno == EACCES) {
+ fprintf(stderr, "Please run as root\n");
+ } else if (errno == ENOENT) {
+ if (opendir("/sys/kernel/debug") == NULL)
+ fprintf(stderr, "Mount debugfs at /sys/kernel/debug\n");
+ else
+ fprintf(stderr, "Check CONFIG_GUP_TEST in kernel config\n");
+ } else {
+ fprintf(stderr, "Failed to open %s: %s\n", GUP_TEST_FILE,
+ strerror(errno));
+ }
+ exit(1);
+ }
+
+ result = (argc == 1) ? run_matrix() : run_bench(&run);
+ close(gup_fd);
+ return result;
+}
--
2.39.5
next prev parent reply other threads:[~2026-05-15 8:49 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-05-15 8:48 [PATCH 0/2] selftests/mm: separate GUP benchmarking from functional testing Sarthak Sharma
2026-05-15 8:48 ` Sarthak Sharma [this message]
2026-05-15 8:48 ` [PATCH 2/2] selftests/mm: rewrite gup_test as a standalone harness-based selftest Sarthak Sharma
2026-05-15 20:33 ` John Hubbard
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260515084840.174652-2-sarthak.sharma@arm.com \
--to=sarthak.sharma@arm.com \
--cc=Liam.Howlett@oracle.com \
--cc=akpm@linux-foundation.org \
--cc=david@kernel.org \
--cc=jgg@ziepe.ca \
--cc=jhubbard@nvidia.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-kselftest@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=ljs@kernel.org \
--cc=mhocko@suse.com \
--cc=peterx@redhat.com \
--cc=rppt@kernel.org \
--cc=shuah@kernel.org \
--cc=surenb@google.com \
--cc=vbabka@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox