[PATCH v2 0/2] selftests/mm: separate GUP microbenchmarking from functional testing

[PATCH v2 0/2] selftests/mm: separate GUP microbenchmarking from functional testing

[Sarthak Sharma]

gup_test.c currently serves two distinct purposes: microbenchmarking
(GUP_FAST_BENCHMARK, PIN_FAST_BENCHMARK, PIN_LONGTERM_BENCHMARK) and
functional correctness testing (GUP_BASIC_TEST, PIN_BASIC_TEST,
DUMP_USER_PAGES_TEST). Mixing these in a single binary means functional
tests cannot be run or reported individually, and run_vmtests.sh must
invoke the binary multiple times with different flag combinations to
cover all configurations. This patch series separates the two concerns:
tools/mm/gup_bench for benchmarking and tools/testing/selftests/mm/gup_test
for functional testing.

Patch 1 adds tools/mm/gup_bench.c, a standalone microbenchmark for
GUP_FAST, PIN_FAST and PIN_LONGTERM via the CONFIG_GUP_TEST debugfs
interface. It runs the same matrix of configurations as the old
run_gup_matrix() shell function (all three commands, read/write,
private/shared, four page counts, THP on/off, hugetlb), but as a
standalone C program under tools/mm with no dependency on kselftest.

Patch 2 rewrites gup_test.c as a kselftest harness-based selftest. It
covers all five GUP kernel functions (get_user_pages, get_user_pages_fast,
pin_user_pages, pin_user_pages_fast, pin_user_pages with FOLL_LONGTERM)
plus DUMP_USER_PAGES_TEST, across 12 mapping configurations (THP on,
THP off and hugetlb, each across private/shared and read/write variants)
and four batch sizes (1, 512, 123, all pages). Results are reported as
standard TAP output with no command-line arguments required.

---
These patches apply on top of mm/mm-new.

Changes in v2:
- Address v1 feedback from Sashiko
- Add fast and longterm GUP/PUP coverage
- Sweep nr_pages_per_call over 1, 512, 123, and all pages
- Call madvise(MADV_NOHUGEPAGE) in non-THP variants
- Use 256 MB for hugetlb fixtures
- Use hugetlb_restore_settings() in FIXTURE_TEARDOWN instead of atexit()
- Add TH_LOG to report nr_pages_per_call for each iteration
- Update Documentation/core-api/pin_user_pages.rst unit testing section

Sarthak Sharma (2):
  tools/mm: add a standalone GUP microbenchmark
  selftests/mm: rewrite gup_test as a standalone harness-based selftest

 Documentation/core-api/pin_user_pages.rst |  12 +-
 MAINTAINERS                               |   1 +
 tools/mm/.gitignore                       |   2 +
 tools/mm/Makefile                         |   6 +-
 tools/mm/gup_bench.c                      | 491 ++++++++++++++++++++
 tools/testing/selftests/mm/gup_test.c     | 536 +++++++++++++---------
 tools/testing/selftests/mm/run_vmtests.sh |  37 +-
 7 files changed, 822 insertions(+), 263 deletions(-)
 create mode 100644 tools/mm/gup_bench.c


base-commit: 2c3f468717231305523ddcd94d91c0d5e4a72419
-- 
2.39.5

[PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Sarthak Sharma]

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

[PATCH v2 2/2] selftests/mm: rewrite gup_test as a standalone harness-based selftest

[Sarthak Sharma]

Rewrite gup_test.c using kselftest_harness.h. The new test covers 12
mapping configurations: THP on, THP off and hugetlb, each across
private/shared and read/write variants. It runs seven test cases per
variant: get_user_pages, get_user_pages_fast, pin_user_pages,
pin_user_pages_fast, pin_user_pages_longterm, and DUMP_USER_PAGES_TEST
via both get and pin.

Each test case sweeps four nr_pages_per_call values: 1, 512, 123, and
all pages. This preserves the old run_gup_matrix() sweep: 12 mapping
combinations x 5 GUP/PUP operations x 4 batch sizes = 240 ioctl sweeps.
It also expands DUMP_USER_PAGES_TEST coverage from one standalone
invocation to 12 variants x 2 dump modes x 4 batch sizes = 96
additional sweeps, for 336 total ioctl sweeps and 84 TAP-reported cases.

On a Radxa Orion O6 board, ./gup_test completes in 5.07s on average
over 10 runs (range: 4.94s - 5.18s).

Update run_vmtests.sh: remove run_gup_matrix() and the multiple flagged
invocations of gup_test, replacing them with a single unconditional
invocation. Benchmark functionality is handled by tools/mm/gup_bench
introduced in the previous patch.

Update Documentation/core-api/pin_user_pages.rst to reflect the new
harness-based gup_test interface rather than command-line flag
invocations.

Suggested-by: David Hildenbrand (Arm) <david@kernel.org>
Signed-off-by: Sarthak Sharma <sarthak.sharma@arm.com>
---
 Documentation/core-api/pin_user_pages.rst |  12 +-
 tools/testing/selftests/mm/gup_test.c     | 536 +++++++++++++---------
 tools/testing/selftests/mm/run_vmtests.sh |  37 +-
 3 files changed, 325 insertions(+), 260 deletions(-)

diff --git a/Documentation/core-api/pin_user_pages.rst b/Documentation/core-api/pin_user_pages.rst
index c16ca163b55e..ea722adf22cc 100644
--- a/Documentation/core-api/pin_user_pages.rst
+++ b/Documentation/core-api/pin_user_pages.rst
@@ -230,10 +230,16 @@ This file::
 
  tools/testing/selftests/mm/gup_test.c
 
-has the following new calls to exercise the new pin*() wrapper functions:
+contains the following test cases to exercise pin_user_pages*():
 
-* PIN_FAST_BENCHMARK (./gup_test -a)
-* PIN_BASIC_TEST (./gup_test -b)
+* pin_user_pages via PIN_BASIC_TEST
+* pin_user_pages_fast via PIN_FAST_BENCHMARK
+* pin_user_pages_longterm via PIN_LONGTERM_BENCHMARK
+
+Run with::
+
+  make -C tools/testing/selftests/mm
+  ./tools/testing/selftests/mm/gup_test
 
 You can monitor how many total dma-pinned pages have been acquired and released
 since the system was booted, via two new /proc/vmstat entries: ::
diff --git a/tools/testing/selftests/mm/gup_test.c b/tools/testing/selftests/mm/gup_test.c
index 3f841a96f870..d60d48bb9126 100644
--- a/tools/testing/selftests/mm/gup_test.c
+++ b/tools/testing/selftests/mm/gup_test.c
@@ -9,267 +9,361 @@
 #include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
-#include <pthread.h>
-#include <assert.h>
 #include <mm/gup_test.h>
 #include "kselftest.h"
 #include "vm_util.h"
 #include "hugepage_settings.h"
+#include "kselftest_harness.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 */
 
+/* Page counts exercising single, THP-batch, partial, and full-mapping GUP. */
+static const int nr_pages_list[] = { 1, 512, 123, -1 };
+
 #define GUP_TEST_FILE "/sys/kernel/debug/gup_test"
 
-static unsigned long cmd = GUP_FAST_BENCHMARK;
-static int gup_fd, repeats = 1;
-static unsigned long size = 128 * MB;
-/* Serialize prints */
-static pthread_mutex_t print_mutex = PTHREAD_MUTEX_INITIALIZER;
+FIXTURE(gup_test) {
+	int gup_fd;
+	char *addr;
+	unsigned long size;
+};
+
+FIXTURE_VARIANT(gup_test) {
+	bool thp;
+	bool hugetlb;
+	bool write;
+	bool shared;
+};
+
+FIXTURE_VARIANT_ADD(gup_test, private_write)
+{
+	.thp = false,
+	.hugetlb = false,
+	.write = true,
+	.shared = false,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, private_readonly)
+{
+	.thp = false,
+	.hugetlb = false,
+	.write = false,
+	.shared = false,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, private_write_thp)
+{
+	.thp = true,
+	.hugetlb = false,
+	.write = true,
+	.shared = false,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, private_readonly_thp)
+{
+	.thp = true,
+	.hugetlb = false,
+	.write = false,
+	.shared = false,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, private_write_hugetlb)
+{
+	.thp = false,
+	.hugetlb = true,
+	.write = true,
+	.shared = false,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, private_readonly_hugetlb)
+{
+	.thp = false,
+	.hugetlb = true,
+	.write = false,
+	.shared = false,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, shared_write)
+{
+	.thp = false,
+	.hugetlb = false,
+	.write = true,
+	.shared = true,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, shared_readonly)
+{
+	.thp = false,
+	.hugetlb = false,
+	.write = false,
+	.shared = true,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, shared_write_thp)
+{
+	.thp = true,
+	.hugetlb = false,
+	.write = true,
+	.shared = true,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, shared_readonly_thp)
+{
+	.thp = true,
+	.hugetlb = false,
+	.write = false,
+	.shared = true,
+};
+
+FIXTURE_VARIANT_ADD(gup_test, shared_write_hugetlb)
+{
+	.thp = false,
+	.hugetlb = true,
+	.write = true,
+	.shared = true,
+};
 
-static char *cmd_to_str(unsigned long cmd)
+FIXTURE_VARIANT_ADD(gup_test, shared_readonly_hugetlb)
 {
-	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";
-	case GUP_BASIC_TEST:
-		return "GUP_BASIC_TEST";
-	case PIN_BASIC_TEST:
-		return "PIN_BASIC_TEST";
-	case DUMP_USER_PAGES_TEST:
-		return "DUMP_USER_PAGES_TEST";
+	.thp = false,
+	.hugetlb = true,
+	.write = false,
+	.shared = true,
+};
+
+FIXTURE_SETUP(gup_test) {
+	int mmap_flags = MAP_PRIVATE;
+	int zero_fd;
+	char *p;
+
+	self->size = variant->hugetlb ? 256 * MB : 128 * MB;
+
+	/* Check for hugetlb */
+	if (variant->hugetlb) {
+		unsigned long hp_size = default_huge_page_size();
+
+		if (!hp_size)
+			SKIP(return, "HugeTLB not available\n");
+
+		self->size = (self->size + hp_size - 1) & ~(hp_size - 1);
+		if (!hugetlb_setup_default(self->size / hp_size))
+			SKIP(return, "Not enough huge pages\n");
+
+		mmap_flags |= (MAP_HUGETLB | MAP_ANONYMOUS);
 	}
-	return "Unknown command";
+
+	/* zero_fd has to be >= 0. Already checked in main() */
+	zero_fd = open("/dev/zero", O_RDWR);
+	ASSERT_GE(zero_fd, 0);
+
+	/* gup_fd has to be >= 0. Already checked in main() */
+	self->gup_fd = open(GUP_TEST_FILE, O_RDWR);
+	ASSERT_GE(self->gup_fd, 0);
+
+	if (variant->shared)
+		mmap_flags = (mmap_flags & ~MAP_PRIVATE) | MAP_SHARED;
+
+	self->addr = mmap(NULL, self->size, PROT_READ | PROT_WRITE,
+					mmap_flags, zero_fd, 0);
+	close(zero_fd);
+	ASSERT_NE(self->addr, MAP_FAILED);
+
+	if (variant->thp)
+		madvise(self->addr, self->size, MADV_HUGEPAGE);
+	else
+		madvise(self->addr, self->size, MADV_NOHUGEPAGE);
+
+	for (p = self->addr; (unsigned long)p < (unsigned long)self->addr
+			+ self->size; p += psize())
+		p[0] = 0;
 }
 
-void *gup_thread(void *data)
-{
-	struct gup_test gup = *(struct gup_test *)data;
-	int i, status;
-
-	/* Only report timing information on the *_BENCHMARK commands: */
-	if ((cmd == PIN_FAST_BENCHMARK) || (cmd == GUP_FAST_BENCHMARK) ||
-	     (cmd == PIN_LONGTERM_BENCHMARK)) {
-		for (i = 0; i < repeats; i++) {
-			gup.size = size;
-			status = ioctl(gup_fd, cmd, &gup);
-			if (status)
-				break;
-
-			pthread_mutex_lock(&print_mutex);
-			ksft_print_msg("%s: Time: get:%lld put:%lld us",
-				       cmd_to_str(cmd), gup.get_delta_usec,
-				       gup.put_delta_usec);
-			if (gup.size != size)
-				ksft_print_msg(", truncated (size: %lld)", gup.size);
-			ksft_print_msg("\n");
-			pthread_mutex_unlock(&print_mutex);
-		}
-	} else {
-		gup.size = size;
-		status = ioctl(gup_fd, cmd, &gup);
-		if (status)
-			goto return_;
-
-		pthread_mutex_lock(&print_mutex);
-		ksft_print_msg("%s: done\n", cmd_to_str(cmd));
-		if (gup.size != size)
-			ksft_print_msg("Truncated (size: %lld)\n", gup.size);
-		pthread_mutex_unlock(&print_mutex);
+FIXTURE_TEARDOWN(gup_test) {
+	munmap(self->addr, self->size);
+	close(self->gup_fd);
+
+	if (variant->hugetlb)
+		hugetlb_restore_settings();
+}
+
+TEST_F(gup_test, get_user_pages) {
+	/* Tests the get_user_pages path */
+	int i;
+
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
+
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
+
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, GUP_BASIC_TEST, &gup), 0);
 	}
+}
+
+TEST_F(gup_test, pin_user_pages) {
+	/* Tests the pin_user_pages path */
+	int i;
+
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
+
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
 
-return_:
-	ksft_test_result(!status, "ioctl status %d\n", status);
-	return NULL;
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, PIN_BASIC_TEST, &gup), 0);
+	}
 }
 
-int main(int argc, char **argv)
-{
-	struct gup_test gup = { 0 };
-	int filed, i, opt, nr_pages = 1, thp = -1, write = 1, nthreads = 1, ret;
-	int flags = MAP_PRIVATE;
-	char *file = "/dev/zero";
-	bool hugetlb = false;
-	pthread_t *tid;
-	char *p;
+TEST_F(gup_test, dump_user_pages_with_get) {
+	/* Tests DUMP_USER_PAGES_TEST using get_user_pages */
+	int i;
 
-	while ((opt = getopt(argc, argv, "m:r:n:F:f:abcj:tTLUuwWSHpz")) != -1) {
-		switch (opt) {
-		case 'a':
-			cmd = PIN_FAST_BENCHMARK;
-			break;
-		case 'b':
-			cmd = PIN_BASIC_TEST;
-			break;
-		case 'L':
-			cmd = PIN_LONGTERM_BENCHMARK;
-			break;
-		case 'c':
-			cmd = DUMP_USER_PAGES_TEST;
-			/*
-			 * Dump page 0 (index 1). May be overridden later, by
-			 * user's non-option arguments.
-			 *
-			 * .which_pages is zero-based, so that zero can mean "do
-			 * nothing".
-			 */
-			gup.which_pages[0] = 1;
-			break;
-		case 'p':
-			/* works only with DUMP_USER_PAGES_TEST */
-			gup.test_flags |= GUP_TEST_FLAG_DUMP_PAGES_USE_PIN;
-			break;
-		case 'F':
-			/* strtol, so you can pass flags in hex form */
-			gup.gup_flags = strtol(optarg, 0, 0);
-			break;
-		case 'j':
-			nthreads = atoi(optarg);
-			break;
-		case 'm':
-			size = atoi(optarg) * MB;
-			break;
-		case 'r':
-			repeats = atoi(optarg);
-			break;
-		case 'n':
-			nr_pages = atoi(optarg);
-			if (nr_pages < 0)
-				nr_pages = size / psize();
-			break;
-		case 't':
-			thp = 1;
-			break;
-		case 'T':
-			thp = 0;
-			break;
-		case 'U':
-			cmd = GUP_BASIC_TEST;
-			break;
-		case 'u':
-			cmd = GUP_FAST_BENCHMARK;
-			break;
-		case 'w':
-			write = 1;
-			break;
-		case 'W':
-			write = 0;
-			break;
-		case 'f':
-			file = optarg;
-			break;
-		case 'S':
-			flags &= ~MAP_PRIVATE;
-			flags |= MAP_SHARED;
-			break;
-		case 'H':
-			flags |= (MAP_HUGETLB | MAP_ANONYMOUS);
-			hugetlb = true;
-			break;
-		default:
-			ksft_exit_fail_msg("Wrong argument\n");
-		}
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
+
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
+
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		gup.which_pages[0] = 1;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, DUMP_USER_PAGES_TEST, &gup), 0);
 	}
+}
 
-	if (optind < argc) {
-		int extra_arg_count = 0;
-		/*
-		 * For example:
-		 *
-		 *   ./gup_test -c 0 1 0x1001
-		 *
-		 * ...to dump pages 0, 1, and 4097
-		 */
-
-		while ((optind < argc) &&
-		       (extra_arg_count < GUP_TEST_MAX_PAGES_TO_DUMP)) {
-			/*
-			 * Do the 1-based indexing here, so that the user can
-			 * use normal 0-based indexing on the command line.
-			 */
-			long page_index = strtol(argv[optind], 0, 0) + 1;
-
-			gup.which_pages[extra_arg_count] = page_index;
-			extra_arg_count++;
-			optind++;
-		}
+TEST_F(gup_test, dump_user_pages_with_pin) {
+	/* Tests DUMP_USER_PAGES_TEST using pin_user_pages */
+	int i;
+
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
+
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
+
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		gup.which_pages[0] = 1;
+		gup.test_flags |= GUP_TEST_FLAG_DUMP_PAGES_USE_PIN;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, DUMP_USER_PAGES_TEST, &gup), 0);
 	}
+}
 
-	ksft_print_header();
+TEST_F(gup_test, get_user_pages_fast) {
+	/* Tests the lockless get_user_pages_fast() path */
+	int i;
 
-	if (hugetlb) {
-		unsigned long hp_size = default_huge_page_size();
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
 
-		if (!hp_size)
-			ksft_exit_skip("HugeTLB is unavailable\n");
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
 
-		size = (size + hp_size - 1) & ~(hp_size - 1);
-		if (!hugetlb_setup_default(size / hp_size))
-			ksft_exit_skip("Not enough huge pages\n");
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, GUP_FAST_BENCHMARK, &gup), 0);
 	}
+}
 
-	ksft_set_plan(nthreads);
+TEST_F(gup_test, pin_user_pages_fast) {
+	/* Tests the lockless pin_user_pages_fast() path */
+	int i;
 
-	filed = open(file, O_RDWR|O_CREAT, 0664);
-	if (filed < 0)
-		ksft_exit_fail_msg("Unable to open %s: %s\n", file, strerror(errno));
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
 
-	gup.nr_pages_per_call = nr_pages;
-	if (write)
-		gup.gup_flags |= FOLL_WRITE;
-
-	gup_fd = open(GUP_TEST_FILE, O_RDWR);
-	if (gup_fd == -1) {
-		switch (errno) {
-		case EACCES:
-			if (getuid())
-				ksft_print_msg("Please run this test as root\n");
-			break;
-		case ENOENT:
-			if (opendir("/sys/kernel/debug") == NULL)
-				ksft_print_msg("mount debugfs at /sys/kernel/debug\n");
-			ksft_print_msg("check if CONFIG_GUP_TEST is enabled in kernel config\n");
-			break;
-		default:
-			ksft_print_msg("failed to open %s: %s\n", GUP_TEST_FILE, strerror(errno));
-			break;
-		}
-		ksft_test_result_skip("Please run this test as root\n");
-		ksft_exit_pass();
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
+
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, PIN_FAST_BENCHMARK, &gup), 0);
 	}
+}
 
-	p = mmap(NULL, size, PROT_READ | PROT_WRITE, flags, filed, 0);
-	if (p == MAP_FAILED)
-		ksft_exit_fail_msg("mmap: %s\n", strerror(errno));
-	gup.addr = (unsigned long)p;
+TEST_F(gup_test, pin_user_pages_longterm) {
+	/* Tests pin_user_pages() with FOLL_LONGTERM */
+	int i;
 
-	if (thp == 1)
-		madvise(p, size, MADV_HUGEPAGE);
-	else if (thp == 0)
-		madvise(p, size, MADV_NOHUGEPAGE);
+	for (i = 0; i < (int)ARRAY_SIZE(nr_pages_list); i++) {
+		struct gup_test gup = { 0 };
 
-	/* Fault them in here, from user space. */
-	for (; (unsigned long)p < gup.addr + size; p += psize())
-		p[0] = 0;
+		gup.addr = (unsigned long)self->addr;
+		gup.size = self->size;
+		gup.nr_pages_per_call = nr_pages_list[i] < 0 ?
+			self->size / psize() : nr_pages_list[i];
 
-	tid = malloc(sizeof(pthread_t) * nthreads);
-	assert(tid);
-	for (i = 0; i < nthreads; i++) {
-		ret = pthread_create(&tid[i], NULL, gup_thread, &gup);
-		assert(ret == 0);
+		if (variant->write)
+			gup.gup_flags |= FOLL_WRITE;
+
+		TH_LOG("nr_pages_per_call=%u", gup.nr_pages_per_call);
+		ASSERT_EQ(ioctl(self->gup_fd, PIN_LONGTERM_BENCHMARK, &gup), 0);
 	}
-	for (i = 0; i < nthreads; i++) {
-		ret = pthread_join(tid[i], NULL);
-		assert(ret == 0);
+}
+
+int main(int argc, char **argv)
+{
+	int fd;
+	char *file = "/dev/zero";
+
+	fd = open(file, O_RDWR);
+	if (fd < 0) {
+		ksft_print_header();
+		ksft_exit_fail_msg("Unable to open %s: %s\n", file, strerror(errno));
 	}
+	close(fd);
 
-	free(tid);
+	fd = open(GUP_TEST_FILE, O_RDWR);
+	if (fd == -1) {
+		ksft_print_header();
+		if (errno == EACCES)
+			ksft_exit_skip("Please run this test as root\n");
+		if (errno == ENOENT) {
+			if (opendir("/sys/kernel/debug") == NULL)
+				ksft_exit_skip("Mount debugfs at /sys/kernel/debug\n");
+			else
+				ksft_exit_skip("Check CONFIG_GUP_TEST in kernel config\n");
+		}
+		ksft_exit_skip("failed to open %s: %s\n", GUP_TEST_FILE, strerror(errno));
+	}
+	close(fd);
 
-	ksft_exit_pass();
+	return test_harness_run(argc, argv);
 }
diff --git a/tools/testing/selftests/mm/run_vmtests.sh b/tools/testing/selftests/mm/run_vmtests.sh
index 043aa3ed2596..65a4ef0f3748 100755
--- a/tools/testing/selftests/mm/run_vmtests.sh
+++ b/tools/testing/selftests/mm/run_vmtests.sh
@@ -130,30 +130,6 @@ test_selected() {
 	fi
 }
 
-run_gup_matrix() {
-    # -t: thp=on, -T: thp=off, -H: hugetlb=on
-    local hugetlb_mb=256
-
-    for huge in -t -T "-H -m $hugetlb_mb"; do
-        # -u: gup-fast, -U: gup-basic, -a: pin-fast, -b: pin-basic, -L: pin-longterm
-        for test_cmd in -u -U -a -b -L; do
-            # -w: write=1, -W: write=0
-            for write in -w -W; do
-                # -S: shared
-                for share in -S " "; do
-                    # -n: How many pages to fetch together?  512 is special
-                    # because it's default thp size (or 2M on x86), 123 to
-                    # just test partial gup when hit a huge in whatever form
-                    for num in "-n 1" "-n 512" "-n 123" "-n -1"; do
-                        CATEGORY="gup_test" run_test ./gup_test \
-                                $huge $test_cmd $write $share $num
-                    done
-                done
-            done
-        done
-    done
-}
-
 # filter 64bit architectures
 ARCH64STR="arm64 mips64 parisc64 ppc64 ppc64le riscv64 s390x sparc64 x86_64"
 if [ -z "$ARCH" ]; then
@@ -239,18 +215,7 @@ fi
 
 CATEGORY="mmap" run_test ./map_fixed_noreplace
 
-if $RUN_ALL; then
-    run_gup_matrix
-else
-    # get_user_pages_fast() benchmark
-    CATEGORY="gup_test" run_test ./gup_test -u -n 1
-    CATEGORY="gup_test" run_test ./gup_test -u -n -1
-    # pin_user_pages_fast() benchmark
-    CATEGORY="gup_test" run_test ./gup_test -a -n 1
-    CATEGORY="gup_test" run_test ./gup_test -a -n -1
-fi
-# Dump pages 0, 19, and 4096, using pin_user_pages:
-CATEGORY="gup_test" run_test ./gup_test -ct -F 0x1 0 19 0x1000
+CATEGORY="gup_test" run_test ./gup_test
 CATEGORY="gup_test" run_test ./gup_longterm
 
 CATEGORY="userfaultfd" run_test ./uffd-unit-tests
-- 
2.39.5

Re: [PATCH v2 0/2] selftests/mm: separate GUP microbenchmarking from functional testing

[Andrew Morton]

On Tue, 19 May 2026 17:35:04 +0530 Sarthak Sharma <sarthak.sharma@arm.com> wrote:

> gup_test.c currently serves two distinct purposes: microbenchmarking
> (GUP_FAST_BENCHMARK, PIN_FAST_BENCHMARK, PIN_LONGTERM_BENCHMARK) and
> functional correctness testing (GUP_BASIC_TEST, PIN_BASIC_TEST,
> DUMP_USER_PAGES_TEST). Mixing these in a single binary means functional
> tests cannot be run or reported individually, and run_vmtests.sh must
> invoke the binary multiple times with different flag combinations to
> cover all configurations. This patch series separates the two concerns:
> tools/mm/gup_bench for benchmarking and tools/testing/selftests/mm/gup_test
> for functional testing.
> 
> Patch 1 adds tools/mm/gup_bench.c, a standalone microbenchmark for
> GUP_FAST, PIN_FAST and PIN_LONGTERM via the CONFIG_GUP_TEST debugfs
> interface. It runs the same matrix of configurations as the old
> run_gup_matrix() shell function (all three commands, read/write,
> private/shared, four page counts, THP on/off, hugetlb), but as a
> standalone C program under tools/mm with no dependency on kselftest.
> 
> Patch 2 rewrites gup_test.c as a kselftest harness-based selftest. It
> covers all five GUP kernel functions (get_user_pages, get_user_pages_fast,
> pin_user_pages, pin_user_pages_fast, pin_user_pages with FOLL_LONGTERM)
> plus DUMP_USER_PAGES_TEST, across 12 mapping configurations (THP on,
> THP off and hugetlb, each across private/shared and read/write variants)
> and four batch sizes (1, 512, 123, all pages). Results are reported as
> standard TAP output with no command-line arguments required.

Thanks.  AI review asked a few things which seem fairly minor to me,
but probably legitimate:
	https://sashiko.dev/#/patchset/20260519120506.184512-1-sarthak.sharma@arm.com

Re: [PATCH v2 0/2] selftests/mm: separate GUP microbenchmarking from functional testing

[Sarthak Sharma]

Hi Andrew!

On 5/19/26 11:50 PM, Andrew Morton wrote:
> On Tue, 19 May 2026 17:35:04 +0530 Sarthak Sharma <sarthak.sharma@arm.com> wrote:
> 
>> gup_test.c currently serves two distinct purposes: microbenchmarking
>> (GUP_FAST_BENCHMARK, PIN_FAST_BENCHMARK, PIN_LONGTERM_BENCHMARK) and
>> functional correctness testing (GUP_BASIC_TEST, PIN_BASIC_TEST,
>> DUMP_USER_PAGES_TEST). Mixing these in a single binary means functional
>> tests cannot be run or reported individually, and run_vmtests.sh must
>> invoke the binary multiple times with different flag combinations to
>> cover all configurations. This patch series separates the two concerns:
>> tools/mm/gup_bench for benchmarking and tools/testing/selftests/mm/gup_test
>> for functional testing.
>>
>> Patch 1 adds tools/mm/gup_bench.c, a standalone microbenchmark for
>> GUP_FAST, PIN_FAST and PIN_LONGTERM via the CONFIG_GUP_TEST debugfs
>> interface. It runs the same matrix of configurations as the old
>> run_gup_matrix() shell function (all three commands, read/write,
>> private/shared, four page counts, THP on/off, hugetlb), but as a
>> standalone C program under tools/mm with no dependency on kselftest.
>>
>> Patch 2 rewrites gup_test.c as a kselftest harness-based selftest. It
>> covers all five GUP kernel functions (get_user_pages, get_user_pages_fast,
>> pin_user_pages, pin_user_pages_fast, pin_user_pages with FOLL_LONGTERM)
>> plus DUMP_USER_PAGES_TEST, across 12 mapping configurations (THP on,
>> THP off and hugetlb, each across private/shared and read/write variants)
>> and four batch sizes (1, 512, 123, all pages). Results are reported as
>> standard TAP output with no command-line arguments required.
> 
> Thanks.  AI review asked a few things which seem fairly minor to me,
> but probably legitimate:
> 	https://sashiko.dev/#/patchset/20260519120506.184512-1-sarthak.sharma@arm.com

Thanks for pointing it out. I'll address the review comments and send a v3.

Re: [PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Mike Rapoport]

(added broonie)

Hi,

On Tue, May 19, 2026 at 05:35:05PM +0530, Sarthak Sharma wrote:
> 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

...
 
> +/*
> + * 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.
> + */

It looks like self tests of at least 5 subsystems beside mm use hugetlb:

$ git grep -l "Hugepagesize:" tools/testing/selftests/ | grep -v "selftests/mm"
tools/testing/selftests/arm64/mte/check_hugetlb_options.c
tools/testing/selftests/cgroup/test_hugetlb_memcg.c
tools/testing/selftests/kvm/lib/test_util.c
tools/testing/selftests/memfd/common.c
tools/testing/selftests/net/tcp_mmap.c

It seems that we need to better share the common code in
tools/testing/selftest.

And adding another copy of the hugetlb detection and setup code does not
seem like a great idea.

> +
> +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;
> +}

-- 
Sincerely yours,
Mike.

Re: [PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Dev Jain]

On 20/05/26 2:25 pm, Mike Rapoport wrote:
> (added broonie)
> 
> Hi,
> 
> On Tue, May 19, 2026 at 05:35:05PM +0530, Sarthak Sharma wrote:
>> 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
> 
> ...
>  
>> +/*
>> + * 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.
>> + */
> 
> It looks like self tests of at least 5 subsystems beside mm use hugetlb:
> 
> $ git grep -l "Hugepagesize:" tools/testing/selftests/ | grep -v "selftests/mm"
> tools/testing/selftests/arm64/mte/check_hugetlb_options.c
> tools/testing/selftests/cgroup/test_hugetlb_memcg.c
> tools/testing/selftests/kvm/lib/test_util.c
> tools/testing/selftests/memfd/common.c
> tools/testing/selftests/net/tcp_mmap.c
> 
> It seems that we need to better share the common code in
> tools/testing/selftest.
> 
> And adding another copy of the hugetlb detection and setup code does not
> seem like a great idea.


Does it sound too insane to just do some sort of #include "../testing/selftests/mm/..."
to use the common helpers?

> 
>> +
>> +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;
>> +}
> 

Re: [PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Sarthak Sharma]

Hi Mike!

On 5/20/26 2:25 PM, Mike Rapoport wrote:
> (added broonie)
> 
> Hi,
> 
> On Tue, May 19, 2026 at 05:35:05PM +0530, Sarthak Sharma wrote:
>> 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
> 
> ...
>  
>> +/*
>> + * 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.
>> + */
> 
> It looks like self tests of at least 5 subsystems beside mm use hugetlb:
> 
> $ git grep -l "Hugepagesize:" tools/testing/selftests/ | grep -v "selftests/mm"
> tools/testing/selftests/arm64/mte/check_hugetlb_options.c
> tools/testing/selftests/cgroup/test_hugetlb_memcg.c
> tools/testing/selftests/kvm/lib/test_util.c
> tools/testing/selftests/memfd/common.c
> tools/testing/selftests/net/tcp_mmap.c
> 
> It seems that we need to better share the common code in
> tools/testing/selftest.
> 
> And adding another copy of the hugetlb detection and setup code does not
> seem like a great idea.

Agreed, but that was the least disruptive approach I could think of.

I am thinking of doing this now: should I move the
hugepage_settings.[ch] to tools/lib/ and move the read_num(),
write_num(), read_file() and write_file() helpers to a separate file in
tools/lib/ itself without any ksft dependency? Then both
tools/testing/selftests/* and tools/mm/ could share the same code.

Please let me know if some different approach is preferred.

> 
>> +
>> +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;
>> +}
> 

Re: [PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 910 bytes --]

On Wed, May 20, 2026 at 03:45:53PM +0530, Sarthak Sharma wrote:
> On 5/20/26 2:25 PM, Mike Rapoport wrote:

> > It seems that we need to better share the common code in
> > tools/testing/selftest.

> > And adding another copy of the hugetlb detection and setup code does not
> > seem like a great idea.

> Agreed, but that was the least disruptive approach I could think of.

> I am thinking of doing this now: should I move the
> hugepage_settings.[ch] to tools/lib/ and move the read_num(),
> write_num(), read_file() and write_file() helpers to a separate file in
> tools/lib/ itself without any ksft dependency? Then both
> tools/testing/selftests/* and tools/mm/ could share the same code.

Using tools/lib sounds sensible to me - as well as the sharing it makes
it clear that it's a library used by multiple things so avoids the
issues we sometimes have with selftest directories referencing each
other.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Mike Rapoport]

On Wed, May 20, 2026 at 12:58:32PM +0100, Mark Brown wrote:
> On Wed, May 20, 2026 at 03:45:53PM +0530, Sarthak Sharma wrote:
> > On 5/20/26 2:25 PM, Mike Rapoport wrote:
> 
> > > It seems that we need to better share the common code in
> > > tools/testing/selftest.
> 
> > > And adding another copy of the hugetlb detection and setup code does not
> > > seem like a great idea.
> 
> > Agreed, but that was the least disruptive approach I could think of.
> 
> > I am thinking of doing this now: should I move the
> > hugepage_settings.[ch] to tools/lib/ and move the read_num(),
> > write_num(), read_file() and write_file() helpers to a separate file in

these might need some adjustments because they use ksft_(), but in general
it makes sense to me.

> > tools/lib/ itself without any ksft dependency? Then both
> > tools/testing/selftests/* and tools/mm/ could share the same code.
>
> Using tools/lib sounds sensible to me - as well as the sharing it makes
> it clear that it's a library used by multiple things so avoids the
> issues we sometimes have with selftest directories referencing each
> other.

I'd make it tools/lib/mm as most of the files tools/lib/*.c are stubs for
the kernel functions.

-- 
Sincerely yours,
Mike.

Re: [PATCH v2 1/2] tools/mm: add a standalone GUP microbenchmark

[Sarthak Sharma]

On 5/20/26 6:28 PM, Mike Rapoport wrote:
> On Wed, May 20, 2026 at 12:58:32PM +0100, Mark Brown wrote:
>> On Wed, May 20, 2026 at 03:45:53PM +0530, Sarthak Sharma wrote:
>>> On 5/20/26 2:25 PM, Mike Rapoport wrote:
>>
>>>> It seems that we need to better share the common code in
>>>> tools/testing/selftest.
>>
>>>> And adding another copy of the hugetlb detection and setup code does not
>>>> seem like a great idea.
>>
>>> Agreed, but that was the least disruptive approach I could think of.
>>
>>> I am thinking of doing this now: should I move the
>>> hugepage_settings.[ch] to tools/lib/ and move the read_num(),
>>> write_num(), read_file() and write_file() helpers to a separate file in
> 
> these might need some adjustments because they use ksft_(), but in general
> it makes sense to me.
> 
>>> tools/lib/ itself without any ksft dependency? Then both
>>> tools/testing/selftests/* and tools/mm/ could share the same code.
>>
>> Using tools/lib sounds sensible to me - as well as the sharing it makes
>> it clear that it's a library used by multiple things so avoids the
>> issues we sometimes have with selftest directories referencing each
>> other.
> 
> I'd make it tools/lib/mm as most of the files tools/lib/*.c are stubs for
> the kernel functions.
> 

Thanks Mark and Mike, I'll include these changes in v3.