Skip to contentSkip to conversation

API/cgroup: Add safe_cgroup_lines_scanf

PATCH In review Richard Palethorpe via ltp 3 participants 10 messages Jan 4, 2022 – Jan 5, 2022 Library code

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/5 API/cgroup: Add safe_cgroup_lines_scanf

Similar to file_lines_scanf. Allows us to read a particular key-value pair from a controller file. Can replace kselftest's cg_read_key_* when converting tests.

---

V3:
* Add permissions to open
* Add needs_root to avoid TBROK when creating the FSs

V2:
* Use all filesystems
* Add new EXP macro
* Switch to strtok_r
* Increase the margin of error for exfat and ext234
* Write to the file instead of reading to populate the page cache (allows it to work on tmpfs)

 include/tst_cgroup.h | 12 +++++++++++-
 lib/tst_cgroup.c     | 33 +++++++++++++++++++++++++++++++++
 2 files changed, 44 insertions(+), 1 deletion(-)
include/tst_cgroup.h +11 -1
index 632050e86..561216296 100644
@@ -188,9 +188,19 @@ void safe_cgroup_printf(const char *const file, const int lineno,
 void safe_cgroup_scanf(const char *file, const int lineno,
 		       const struct tst_cgroup_group *const cg,
 		       const char *const file_name,
-		       const char *fmt, ...)
+		       const char *const fmt, ...)
 		       __attribute__ ((format (scanf, 5, 6), nonnull));
 
+#define SAFE_CGROUP_LINES_SCANF(cg, file_name, fmt, ...)		\
+	safe_cgroup_lines_scanf(__FILE__, __LINE__,			\
+				(cg), (file_name), (fmt), __VA_ARGS__)
+
+void safe_cgroup_lines_scanf(const char *const file, const int lineno,
+			     const struct tst_cgroup_group *const cg,
+			     const char *const file_name,
+			     const char *const fmt, ...)
+			__attribute__ ((format (scanf, 5, 6), nonnull));
+
 #define SAFE_CGROUP_OCCURSIN(cg, file_name, needle)		\
 	safe_cgroup_occursin(__FILE__, __LINE__,		\
 			     (cg), (file_name), (needle))
lib/tst_cgroup.c +33 -0
index c08ff2f20..c78f28112 100644
@@ -1161,6 +1161,39 @@ void safe_cgroup_scanf(const char *const file, const int lineno,
 		 file_name, buf, fmt, ret, conv_cnt);
 }
 
+void safe_cgroup_lines_scanf(const char *const file, const int lineno,
+			     const struct tst_cgroup_group *const cg,
+			     const char *const file_name,
+			     const char *const fmt, ...)
+{
+	va_list va;
+	char buf[BUFSIZ];
+	ssize_t len = safe_cgroup_read(file, lineno,
+				       cg, file_name, buf, sizeof(buf));
+	const int conv_cnt = tst_count_scanf_conversions(fmt);
+	int ret = 0;
+	char *line, *buf_ptr;
+
+	if (len < 1)
+		return;
+
+	line = strtok_r(buf, "\n", &buf_ptr);
+	while (line && ret != conv_cnt) {
+		va_start(va, fmt);
+		ret = vsscanf(line, fmt, va);
+		va_end(va);
+
+		line = strtok_r(NULL, "\n", &buf_ptr);
+	}
+
+	if (conv_cnt == ret)
+		return;
+
+	tst_brk_(file, lineno, TBROK,
+		 "'%s': vsscanf('%s', '%s', ..): Less conversions than expected: %d != %d",
+		 file_name, buf, fmt, ret, conv_cnt);
+}
+
 int safe_cgroup_occursin(const char *const file, const int lineno,
 			 const struct tst_cgroup_group *const cg,
 			 const char *const file_name,
Signature · 5 lines
2.34.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#1 lore raw

PATCH v3 2/5 API/cgroup: Add memory.stat

---
 lib/tst_cgroup.c | 1 +
 1 file changed, 1 insertion(+)
lib/tst_cgroup.c +1 -0
index c78f28112..2ef599d9e 100644
@@ -153,6 +153,7 @@ static const struct cgroup_file cgroup_ctrl_files[] = {
 static const struct cgroup_file memory_ctrl_files[] = {
 	{ "memory.current", "memory.usage_in_bytes", CTRL_MEMORY },
 	{ "memory.max", "memory.limit_in_bytes", CTRL_MEMORY },
+	{ "memory.stat", "memory.stat", CTRL_MEMORY },
 	{ "memory.swappiness", "memory.swappiness", CTRL_MEMORY },
 	{ "memory.swap.current", "memory.memsw.usage_in_bytes", CTRL_MEMORY },
 	{ "memory.swap.max", "memory.memsw.limit_in_bytes", CTRL_MEMORY },
Signature · 5 lines
2.34.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#2 lore raw

PATCH v3 3/5 API/fs: Add exfat magic

In reply to Richard Palethorpe via ltp
---
 include/tst_fs.h  | 1 +
 lib/tst_fs_type.c | 2 ++
 2 files changed, 3 insertions(+)
include/tst_fs.h +1 -0
index 36d4b46f0..efcdff608 100644
@@ -30,6 +30,7 @@
 #define TST_EXOFS_MAGIC    0x5DF5
 #define TST_OVERLAYFS_MAGIC 0x794c7630
 #define TST_FUSE_MAGIC     0x65735546
+#define TST_EXFAT_MAGIC    0x2011BAB0UL
 
 enum {
 	TST_BYTES = 1,
lib/tst_fs_type.c +2 -0
index 8475f4c78..9de80224b 100644
@@ -86,6 +86,8 @@ const char *tst_fs_type_name(long f_type)
 		return "overlayfs";
 	case TST_FUSE_MAGIC:
 		return "fuse";
+	case TST_EXFAT_MAGIC:
+		return "exfat";
 	default:
 		return "unknown";
 	}
Signature · 5 lines
2.34.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#4 lore raw

PATCH v3 4/5 API: Add TST_EXP_EXPR macro

In reply to Richard Palethorpe via ltp

Simplifies evaluating an arbitrary expression or statement which is not a syscall.

---
 include/tst_test_macros.h | 3 +++
 1 file changed, 3 insertions(+)
include/tst_test_macros.h +3 -0
index 5dea1aabd..ec8c38523 100644
@@ -195,4 +195,7 @@ extern void *TST_RET_PTR;
 
 #define TST_EXP_FAIL2(SCALL, ERRNO, ...) TST_EXP_FAIL_(TST_RET >= 0, SCALL, #SCALL, ERRNO, ##__VA_ARGS__)
 
+#define TST_EXP_EXPR(EXPR, FMT, ...)						\
+	tst_res_(__FILE__, __LINE__, (EXPR) ? TPASS : TFAIL, "Expect: " FMT, ##__VA_ARGS__);
+
 #endif	/* TST_TEST_MACROS_H__ */
Signature · 5 lines
2.34.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#6 lore raw

PATCH v3 5/5 cgroup: Add memcontrol02

In reply to Richard Palethorpe via ltp

This test appears to compare the overall "current" counter with anon and file (page cache) memory counters. The file test assumes much more memory will be consumed by the page cache. This is certainly true on XFS and BTRFS where little or no memory outside the page cache is used. However on ext4, ntfs and especially exfat, we can see more memory being used.

This seems to be related to fs/buffer.c and buffer_head usage. exfat in particular allocates a lot of memory. This is possibly due to the buffer_head being allocated in cont_write_begin:

Children Self Samples bytes_alloc Parent symbol
........ ........ ............ .................. .................

    97.66%    97.66%        102401                 168  exfat_write_begin
            |
            ---0xf7eec549
               entry_SYSENTER_compat_after_hwframe
               do_fast_syscall_32
               __noinstr_text_start
               ksys_write
               vfs_write
               new_sync_write
               generic_file_write_iter
               __generic_file_write_iter
               generic_perform_write
               exfat_write_begin
               cont_write_begin
               __block_write_begin_int
               create_page_buffers
               create_empty_buffers
               alloc_page_buffers
               alloc_buffer_head
               kmem_cache_alloc
               kmem_cache_alloc

     0.20%     0.20%           205                 584  exfat_write_begin
     0.06%     0.06%            64                 168  exfat_fill_super
     0.05%     0.05%            49                 256  exfat_fill_super
     0.00%     0.00%             4                 584  exfat_fill_super
     0.00%     0.00%             1                 256  exfat_create
     0.00%     0.00%             1                1528  exfat_create
     0.00%     0.00%             1                1528  exfat_fill_super
     0.00%     0.00%             1                 312  exfat_fill_super

We can see (using slub_debug=U,buffer_head) that these buffer_head objects are not freed until the file is unlinked:

Outputting '/sys/kernel/debug/slab/buffer_head/alloc_traces' ... 102480 alloc_buffer_head+0x1b/0x90 age=0/268/2629529 pid=248-430 cpus=0,2

Outputting '/sys/kernel/debug/slab/buffer_head/free_traces' ... 102400 <not-available> age=4297543659 pid=0 cpus=0

     80 free_buffer_head+0x21/0x60 age=147/266217/2629538 pid=337-427 cpus=1-2

ext4 and ntfs also use some of the "buffer" code, but don't seem to allocate quite as much. Although ext4 begins to fail when slub debug is enabled due to the extra memory debugging uses.

In any case, it appears that the CGroup code is correct, so I have increased the error margin when exfat or ext234 is detected.

---
 runtest/controllers                           |   1 +
 testcases/kernel/controllers/memcg/.gitignore |   1 +
 .../kernel/controllers/memcg/memcontrol02.c   | 183 ++++++++++++++++++
 3 files changed, 185 insertions(+)
 create mode 100644 testcases/kernel/controllers/memcg/memcontrol02.c
runtest/controllers +1 -0
index 2b41a94d3..09e0107e4 100644
@@ -18,6 +18,7 @@ memcg_control		memcg_control_test.sh
 
 # kselftest ports
 memcontrol01 memcontrol01
+memcontrol02 memcontrol02
 
 cgroup_fj_function_debug cgroup_fj_function.sh debug
 cgroup_fj_function_cpuset cgroup_fj_function.sh cpuset
testcases/kernel/controllers/memcg/.gitignore +1 -0
index c3565f85c..f7de40d53 100644
@@ -6,3 +6,4 @@
 /regression/memcg_test_4
 /stress/memcg_process_stress
 memcontrol01
+memcontrol02
testcases/kernel/controllers/memcg/memcontrol02.c new file +183 -0

183 lines not on this page · raw message

Signature · 5 lines
2.34.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
Reviewed-by on v3 5/5
#7 lore raw

Re: [LTP] [PATCH v3 5/5] cgroup: Add memcontrol02

Hi!

Quoted 106 lines

This test appears to compare the overall "current" counter with anon and file (page cache) memory counters. The file test assumes much more memory will be consumed by the page cache. This is certainly true on XFS and BTRFS where little or no memory outside the page cache is used. However on ext4, ntfs and especially exfat, we can see more memory being used.

This seems to be related to fs/buffer.c and buffer_head usage. exfat in particular allocates a lot of memory. This is possibly due to the buffer_head being allocated in cont_write_begin:

Children Self Samples bytes_alloc Parent symbol
........ ........ ............ .................. .................

    97.66%    97.66%        102401                 168  exfat_write_begin
            |
            ---0xf7eec549
               entry_SYSENTER_compat_after_hwframe
               do_fast_syscall_32
               __noinstr_text_start
               ksys_write
               vfs_write
               new_sync_write
               generic_file_write_iter
               __generic_file_write_iter
               generic_perform_write
               exfat_write_begin
               cont_write_begin
               __block_write_begin_int
               create_page_buffers
               create_empty_buffers
               alloc_page_buffers
               alloc_buffer_head
               kmem_cache_alloc
               kmem_cache_alloc

     0.20%     0.20%           205                 584  exfat_write_begin
     0.06%     0.06%            64                 168  exfat_fill_super
     0.05%     0.05%            49                 256  exfat_fill_super
     0.00%     0.00%             4                 584  exfat_fill_super
     0.00%     0.00%             1                 256  exfat_create
     0.00%     0.00%             1                1528  exfat_create
     0.00%     0.00%             1                1528  exfat_fill_super
     0.00%     0.00%             1                 312  exfat_fill_super

We can see (using slub_debug=U,buffer_head) that these buffer_head objects are not freed until the file is unlinked:

Outputting '/sys/kernel/debug/slab/buffer_head/alloc_traces' ... 102480 alloc_buffer_head+0x1b/0x90 age=0/268/2629529 pid=248-430 cpus=0,2

Outputting '/sys/kernel/debug/slab/buffer_head/free_traces' ... 102400 <not-available> age=4297543659 pid=0 cpus=0

     80 free_buffer_head+0x21/0x60 age=147/266217/2629538 pid=337-427 cpus=1-2

ext4 and ntfs also use some of the "buffer" code, but don't seem to allocate quite as much. Although ext4 begins to fail when slub debug is enabled due to the extra memory debugging uses.

In any case, it appears that the CGroup code is correct, so I have increased the error margin when exfat or ext234 is detected.

---
 runtest/controllers                           |   1 +
 testcases/kernel/controllers/memcg/.gitignore |   1 +
 .../kernel/controllers/memcg/memcontrol02.c   | 183 ++++++++++++++++++
 3 files changed, 185 insertions(+)
 create mode 100644 testcases/kernel/controllers/memcg/memcontrol02.c
runtest/controllers +1 -0
index 2b41a94d3..09e0107e4 100644
@@ -18,6 +18,7 @@ memcg_control		memcg_control_test.sh
 
 # kselftest ports
 memcontrol01 memcontrol01
+memcontrol02 memcontrol02
 
 cgroup_fj_function_debug cgroup_fj_function.sh debug
 cgroup_fj_function_cpuset cgroup_fj_function.sh cpuset
testcases/kernel/controllers/memcg/.gitignore +1 -0
index c3565f85c..f7de40d53 100644
@@ -6,3 +6,4 @@
 /regression/memcg_test_4
 /stress/memcg_process_stress
 memcontrol01
+memcontrol02
testcases/kernel/controllers/memcg/memcontrol02.c new file +9 -0
new file mode 100644
index 000000000..f236e6cdc
@@ -0,0 +1,183 @@
+// SPDX-License-Identifier: GPL-2.0
+/*\
+ *
+ * [Description]
+ *
+ * Conversion of second kself test in cgroup/test_memcontrol.c.
+ *
+ * Original description:
+ * "This test creates a memory cgroup, allocates some anonymous memory

testcases/kernel/controllers/memcg/memcontrol02.c line 19

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +10 -0
+ * and some pagecache and check memory.current and some memory.stat
+ * values."
+ *
+ * Note that the V1 rss and cache counters were renamed to anon and
+ * file in V2. Besides error reporting, this test differs from the
+ * kselftest in the following ways:
+ *
+ * . It supports V1.
+ * . It writes instead of reads to fill the page cache. Because no
+ *   pages were allocated on tmpfs.

Shouldn't we actually run the test both for read/write and skip the read part of tmpfs?

Well I guess that the pages are put into the page cache the same way regardless if they came from userspace write or as a request for data to be read from the disk, so probably not I guess.

testcases/kernel/controllers/memcg/memcontrol02.c line 21

Quoted 2 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +2 -0
+ * . It runs on most filesystems available
+ * . On EXFAT and extN we change the margine of error between all and file
                                           ^
					   margin

testcases/kernel/controllers/memcg/memcontrol02.c line 27

Quoted 6 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +6 -0
+ *   memory to 50%. Because these allocate non-page-cache memory during writes.
+ */
+#define _GNU_SOURCE
+
+#include <stdlib.h>
+#include <stdio.h>

Do we really need stdio here?

Quoted 12 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +12 -0
+#include "tst_test.h"
+#include "tst_cgroup.h"
+
+#define TMPDIR "mntdir"
+#define MB(x) (x << 20)
+
+static size_t page_size;
+static const struct tst_cgroup_group *cg_test;
+static int is_v1_memcg;
+static struct tst_cgroup_group *cg_child;
+static int fd;
+static int file_to_all_error = 10;

testcases/kernel/controllers/memcg/memcontrol02.c line 50

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +10 -0
+
+/*
+ * Checks if two given values differ by less than err% of their sum.
+ */
+static inline int values_close(const ssize_t a,
+			       const ssize_t b,
+			       const ssize_t err)
+{
+	return labs(a - b) <= (a + b) / 100 * err;
+}

I guess that this crude integer version works only since we allocate reasonable large memory sizes.

Quoted 19 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +19 -0
+static void alloc_anon_50M_check(void)
+{
+	const ssize_t size = MB(50);
+	char *buf, *ptr;
+	ssize_t anon, current;
+	const char *const anon_key_fmt = is_v1_memcg ? "rss %zd" : "anon %zd";
+
+	buf = SAFE_MALLOC(size);
+	for (ptr = buf; ptr < buf + size; ptr += page_size)
+		*ptr = 0;
+
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zd", &current);
+	TST_EXP_EXPR(current >= size,
+		     "(memory.current=%zd) >= (size=%zd)", current, size);
+
+	SAFE_CGROUP_LINES_SCANF(cg_child, "memory.stat", anon_key_fmt, &anon);
+
+	TST_EXP_EXPR(anon > 0, "(memory.stat.anon=%zd) > 0", anon);
+	TST_EXP_EXPR(values_close(size, current, 3),

testcases/kernel/controllers/memcg/memcontrol02.c line 80

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +10 -0
+		     "(size=%zd) ~= (memory.stat.anon=%zd)", size, current);
+	TST_EXP_EXPR(values_close(anon, current, 3),
+		     "(memory.current=%zd) ~= (memory.stat.anon=%zd)",
+		     current, anon);
+}
+
+static void alloc_pagecache(const int fd, size_t size)
+{
+	char buf[BUFSIZ];
+	size_t i;

We may as well fill the buffer with something so that this function will not trigger static analyzers.

testcases/kernel/controllers/memcg/memcontrol02.c line 90

Quoted 9 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +9 -0
+	for (i = 0; i < size; i += sizeof(buf))
+		SAFE_WRITE(1, fd, buf, sizeof(buf));
+}
+
+static void alloc_pagecache_50M_check(void)
+{
+	const size_t size = MB(50);
+	size_t current, file;
+	const char *const file_key_fmt = is_v1_memcg ? "cache %zd" : "file %zd";

This may be a global initialized in the test setup.

Quoted 53 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +53 -0
+	TEST(open(TMPDIR"/tmpfile", O_RDWR | O_CREAT, 0600));
+
+	if (TST_RET < 0) {
+		if (TST_ERR == EOPNOTSUPP)
+			tst_brk(TCONF, "O_TMPFILE not supported by FS");
+
+		tst_brk(TBROK | TTERRNO,
+			"open(%s, O_TMPFILE | O_RDWR | O_EXCL", TMPDIR"/.");
+	}
+	fd = TST_RET;
+
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+	tst_res(TINFO, "Created temp file: memory.current=%zu", current);
+
+	alloc_pagecache(fd, size);
+
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+	TST_EXP_EXPR(current >= size,
+			 "(memory.current=%zu) >= (size=%zu)", current, size);
+
+	SAFE_CGROUP_LINES_SCANF(cg_child, "memory.stat", file_key_fmt, &file);
+	TST_EXP_EXPR(file > 0, "(memory.stat.file=%zd) > 0", file);
+
+	TST_EXP_EXPR(values_close(file, current, file_to_all_error),
+			 "(memory.current=%zd) ~= (memory.stat.file=%zd)",
+			 current, file);
+
+	SAFE_CLOSE(fd);
+}
+
+static void test_memcg_current(unsigned int n)
+{
+	size_t current;
+
+	cg_child = tst_cgroup_group_mk(cg_test, "child");
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+	TST_EXP_EXPR(current == 0, "(current=%zu) == 0", current);
+
+	if (!SAFE_FORK()) {
+		SAFE_CGROUP_PRINTF(cg_child, "cgroup.procs", "%d", getpid());
+
+		SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+		tst_res(TINFO, "Added proc to memcg: memory.current=%zu",
+			current);
+
+		if (!n)
+			alloc_anon_50M_check();
+		else
+			alloc_pagecache_50M_check();
+	} else {
+		tst_reap_children();
+		cg_child = tst_cgroup_group_rm(cg_child);
+	}

testcases/kernel/controllers/memcg/memcontrol02.c line 154

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +10 -0
+}
+
+static void setup(void)
+{
+	page_size = SAFE_SYSCONF(_SC_PAGESIZE);
+
+	tst_cgroup_require("memory", NULL);
+	cg_test = tst_cgroup_get_test_group();
+
+	is_v1_memcg = TST_CGROUP_VER(cg_test, "memory") == TST_CGROUP_V1;

I find this statement a bit confusing, maybe TST_CGGROUP_IS_V1() and TST_CGROUP_IS_V2() macros in the library would make it slightly better.

Quoted 18 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +18 -0
+	switch (tst_fs_type(TMPDIR)) {
+	case TST_EXFAT_MAGIC:
+	case TST_EXT234_MAGIC:
+		file_to_all_error = 50;
+		break;
+	}
+}
+
+static void cleanup(void)
+{
+	if (cg_child)
+		cg_child = tst_cgroup_group_rm(cg_child);
+
+	tst_cgroup_cleanup();
+}
+
+static struct tst_test test = {
+	.setup = setup,

testcases/kernel/controllers/memcg/memcontrol02.c line 183

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +10 -0
+	.cleanup = cleanup,
+	.tcnt = 2,
+	.test = test_memcg_current,
+	.mount_device = 1,
+	.dev_min_size = 256,
+	.mntpoint = TMPDIR,
+	.all_filesystems = 1,
+	.forks_child = 1,
+	.needs_root = 1,
+};

Generally minus typos and minor things it looks good, I guess that it can as well go in if you fix the typos as it is.

Signature · 5 lines
Cyril Hrubis
[email protected]

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#8 lore raw

Re: [LTP] [PATCH v3 5/5] cgroup: Add memcontrol02

Hello,

Quoted 18 lines · Cyril Hrubis
Quoted 9 lines

+ *
+ * [Description]
+ *
+ * Conversion of second kself test in cgroup/test_memcontrol.c.
+ *
+ * Original description:
+ * "This test creates a memory cgroup, allocates some anonymous memory
+ * and some pagecache and check memory.current and some memory.stat
+ * values."

Quoted 8 lines

+ *
+ * Note that the V1 rss and cache counters were renamed to anon and
+ * file in V2. Besides error reporting, this test differs from the
+ * kselftest in the following ways:
+ *
+ * . It supports V1.
+ * . It writes instead of reads to fill the page cache. Because no
+ * pages were allocated on tmpfs.

Quoted 6 lines

Shouldn't we actually run the test both for read/write and skip the read part of tmpfs?

Well I guess that the pages are put into the page cache the same way regardless if they came from userspace write or as a request for data to be read from the disk, so probably not I guess.

I reckon there are a lot of ways to fill the page cache from userland. mmap and madvise also come to mind. I don't know how many ways there are to get/allocate a page from the page cache internally. I guess it's possible to circumvent the accountancy.

I think for now just writing is good enough. This doesn't appear to be the only test which measures the page cache. So I think we should look at what the other tests do first.

Quoted 5 lines
Quoted 3 lines

+ * . It runs on most filesystems available
+ * . On EXFAT and extN we change the margine of error between all and file

                                           ^
					   margin
Quoted 8 lines

testcases/kernel/controllers/memcg/memcontrol02.c line 27

Quoted 6 lines

+ * memory to 50%. Because these allocate non-page-cache memory during writes.
+ */

testcases/kernel/controllers/memcg/memcontrol02.c new file +4 -0
+#define _GNU_SOURCE
+
+#include <stdlib.h>
+#include <stdio.h>

Do we really need stdio here?

Sorry, nope.

Quoted 18 lines
Quoted 18 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +17 -0
+#include "tst_test.h"
+#include "tst_cgroup.h"
+
+#define TMPDIR "mntdir"
+#define MB(x) (x << 20)
+
+static size_t page_size;
+static const struct tst_cgroup_group *cg_test;
+static int is_v1_memcg;
+static struct tst_cgroup_group *cg_child;
+static int fd;
+static int file_to_all_error = 10;
+
+/*
+ * Checks if two given values differ by less than err% of their sum.
+ */
+static inline int values_close(const ssize_t a,
Quoted 8 lines

testcases/kernel/controllers/memcg/memcontrol02.c line 50

Quoted 5 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +5 -0
+			       const ssize_t b,
+			       const ssize_t err)
+{
+	return labs(a - b) <= (a + b) / 100 * err;
+}

I guess that this crude integer version works only since we allocate reasonable large memory sizes.

I suppose I could multiply both sides of the inequality by 100... perhaps we could even use the FPU, I hear it's quite usable in userland. :-p

Quoted 25 lines
Quoted 25 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +24 -0
+static void alloc_anon_50M_check(void)
+{
+	const ssize_t size = MB(50);
+	char *buf, *ptr;
+	ssize_t anon, current;
+	const char *const anon_key_fmt = is_v1_memcg ? "rss %zd" : "anon %zd";
+
+	buf = SAFE_MALLOC(size);
+	for (ptr = buf; ptr < buf + size; ptr += page_size)
+		*ptr = 0;
+
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zd", &current);
+	TST_EXP_EXPR(current >= size,
+		     "(memory.current=%zd) >= (size=%zd)", current, size);
+
+	SAFE_CGROUP_LINES_SCANF(cg_child, "memory.stat", anon_key_fmt, &anon);
+
+	TST_EXP_EXPR(anon > 0, "(memory.stat.anon=%zd) > 0", anon);
+	TST_EXP_EXPR(values_close(size, current, 3),
+		     "(size=%zd) ~= (memory.stat.anon=%zd)", size, current);
+	TST_EXP_EXPR(values_close(anon, current, 3),
+		     "(memory.current=%zd) ~= (memory.stat.anon=%zd)",
+		     current, anon);
+}
Quoted 8 lines

testcases/kernel/controllers/memcg/memcontrol02.c line 80

Quoted 5 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +5 -0
+
+static void alloc_pagecache(const int fd, size_t size)
+{
+	char buf[BUFSIZ];
+	size_t i;

We may as well fill the buffer with something so that this function will not trigger static analyzers.

+1

Quoted 12 lines

testcases/kernel/controllers/memcg/memcontrol02.c line 90

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +9 -0
+	for (i = 0; i < size; i += sizeof(buf))
+		SAFE_WRITE(1, fd, buf, sizeof(buf));
+}
+
+static void alloc_pagecache_50M_check(void)
+{
+	const size_t size = MB(50);
+	size_t current, file;
+	const char *const file_key_fmt = is_v1_memcg ? "cache %zd" : "file %zd";

This may be a global initialized in the test setup.

Is it not easier to read when set in the context it is used? Otherwise the key name is hidden in setup.

Quoted 60 lines
Quoted 60 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +59 -0
+	TEST(open(TMPDIR"/tmpfile", O_RDWR | O_CREAT, 0600));
+
+	if (TST_RET < 0) {
+		if (TST_ERR == EOPNOTSUPP)
+			tst_brk(TCONF, "O_TMPFILE not supported by FS");
+
+		tst_brk(TBROK | TTERRNO,
+			"open(%s, O_TMPFILE | O_RDWR | O_EXCL", TMPDIR"/.");
+	}
+	fd = TST_RET;
+
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+	tst_res(TINFO, "Created temp file: memory.current=%zu", current);
+
+	alloc_pagecache(fd, size);
+
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+	TST_EXP_EXPR(current >= size,
+			 "(memory.current=%zu) >= (size=%zu)", current, size);
+
+	SAFE_CGROUP_LINES_SCANF(cg_child, "memory.stat", file_key_fmt, &file);
+	TST_EXP_EXPR(file > 0, "(memory.stat.file=%zd) > 0", file);
+
+	TST_EXP_EXPR(values_close(file, current, file_to_all_error),
+			 "(memory.current=%zd) ~= (memory.stat.file=%zd)",
+			 current, file);
+
+	SAFE_CLOSE(fd);
+}
+
+static void test_memcg_current(unsigned int n)
+{
+	size_t current;
+
+	cg_child = tst_cgroup_group_mk(cg_test, "child");
+	SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+	TST_EXP_EXPR(current == 0, "(current=%zu) == 0", current);
+
+	if (!SAFE_FORK()) {
+		SAFE_CGROUP_PRINTF(cg_child, "cgroup.procs", "%d", getpid());
+
+		SAFE_CGROUP_SCANF(cg_child, "memory.current", "%zu", &current);
+		tst_res(TINFO, "Added proc to memcg: memory.current=%zu",
+			current);
+
+		if (!n)
+			alloc_anon_50M_check();
+		else
+			alloc_pagecache_50M_check();
+	} else {
+		tst_reap_children();
+		cg_child = tst_cgroup_group_rm(cg_child);
+	}
+}
+
+static void setup(void)
+{
+	page_size = SAFE_SYSCONF(_SC_PAGESIZE);
+
Quoted 8 lines

testcases/kernel/controllers/memcg/memcontrol02.c line 154

Quoted 4 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +4 -0
+	tst_cgroup_require("memory", NULL);
+	cg_test = tst_cgroup_get_test_group();
+
+	is_v1_memcg = TST_CGROUP_VER(cg_test, "memory") == TST_CGROUP_V1;

I find this statement a bit confusing, maybe TST_CGGROUP_IS_V1() and TST_CGROUP_IS_V2() macros in the library would make it slightly better.

Yeah, I suppose this is proving to be tiresome.

Quoted 34 lines
Quoted 19 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +18 -0
+	switch (tst_fs_type(TMPDIR)) {
+	case TST_EXFAT_MAGIC:
+	case TST_EXT234_MAGIC:
+		file_to_all_error = 50;
+		break;
+	}
+}
+
+static void cleanup(void)
+{
+	if (cg_child)
+		cg_child = tst_cgroup_group_rm(cg_child);
+
+	tst_cgroup_cleanup();
+}
+
+static struct tst_test test = {
+	.setup = setup,

testcases/kernel/controllers/memcg/memcontrol02.c line 183

Quoted 10 lines
testcases/kernel/controllers/memcg/memcontrol02.c new file +10 -0
+	.cleanup = cleanup,
+	.tcnt = 2,
+	.test = test_memcg_current,
+	.mount_device = 1,
+	.dev_min_size = 256,
+	.mntpoint = TMPDIR,
+	.all_filesystems = 1,
+	.forks_child = 1,
+	.needs_root = 1,
+};

Generally minus typos and minor things it looks good, I guess that it can as well go in if you fix the typos as it is.

Signature · 5 lines
Thank you,
Richard.

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#9 lore raw

Re: [LTP] [PATCH v3 5/5] cgroup: Add memcontrol02

Hello,

Quoted 52 lines · Richard Palethorpe

Hello,

Quoted 18 lines · Cyril Hrubis
Quoted 9 lines

+ *
+ * [Description]
+ *
+ * Conversion of second kself test in cgroup/test_memcontrol.c.
+ *
+ * Original description:
+ * "This test creates a memory cgroup, allocates some anonymous memory
+ * and some pagecache and check memory.current and some memory.stat
+ * values."

Quoted 8 lines

+ *
+ * Note that the V1 rss and cache counters were renamed to anon and
+ * file in V2. Besides error reporting, this test differs from the
+ * kselftest in the following ways:
+ *
+ * . It supports V1.
+ * . It writes instead of reads to fill the page cache. Because no
+ * pages were allocated on tmpfs.

Quoted 6 lines

Shouldn't we actually run the test both for read/write and skip the read part of tmpfs?

Well I guess that the pages are put into the page cache the same way regardless if they came from userspace write or as a request for data to be read from the disk, so probably not I guess.

I reckon there are a lot of ways to fill the page cache from userland. mmap and madvise also come to mind. I don't know how many ways there are to get/allocate a page from the page cache internally. I guess it's possible to circumvent the accountancy.

I think for now just writing is good enough. This doesn't appear to be the only test which measures the page cache. So I think we should look at what the other tests do first.

Quoted 5 lines
Quoted 3 lines

+ * . It runs on most filesystems available
+ * . On EXFAT and extN we change the margine of error between all and file

                                           ^
					   margin
Quoted 8 lines

testcases/kernel/controllers/memcg/memcontrol02.c line 27

Quoted 6 lines

+ * memory to 50%. Because these allocate non-page-cache memory during writes.
+ */

testcases/kernel/controllers/memcg/memcontrol02.c new file +4 -0
+#define _GNU_SOURCE
+
+#include <stdlib.h>
+#include <stdio.h>

Do we really need stdio here?

Quoted 2 lines

Sorry, nope.

Actually it is needed otherwise BUFSIZ is undefined...

Signature · 5 lines
Thank you,
Richard.

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp