syscalls/madvise09: Fix for disabled swap accounting

RFC PATCH New Cyril Hrubis 2 participants 4 messages May 2, 2017 – May 5, 2017

lore

Share & tools

lore

Cyril Hrubis

author

2017-05-02 13:33 UTC #0 lore raw

RFC PATCH 1/1 syscalls/madvise09: Fix for disabled swap accounting

The test had to be fixed for systems with swap accounting disabled.

First there are no "memsw" prefixed tuning knobs in memcg directories if swap accounting is disabled.

Then we had to limit the hungry child since otherwise it will be killed only after all system swap was filled. Hence we exit the memory hungry child once the process fills 100MB of swap. That seems to be enough to put the cgroup into memory pressure while it limits the test runtime to a few seconds.

Signed-off-by Cyril Hrubis <[email protected]>

---
 testcases/kernel/syscalls/madvise/madvise09.c | 31 ++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)

testcases/kernel/syscalls/madvise/madvise09.c +28 -3

index 45a14a089..38a101fb6 100644
@@ -67,6 +67,8 @@ static char memsw_limit_in_bytes_path[PATH_MAX];
 static size_t page_size;
 static int sleep_between_faults;
 
+static int swap_accounting_enabled;
+
 #define PAGES 32
 #define TOUCHED_PAGE1 0
 #define TOUCHED_PAGE2 10

@@ -84,6 +86,17 @@ static void memory_pressure_child(void)
 			ptr[i * page_size] = i % 100;
 			usleep(sleep_between_faults);
 		}
+
+		/* If swap accounting is disabled exit after process swapped out 100MB */
+		if (!swap_accounting_enabled) {
+			int swapped;
+
+			SAFE_FILE_LINES_SCANF("/proc/self/status", "VmSwap: %d", &swapped);
+
+			if (swapped > 100 * 1024)
+				exit(0);
+		}
+
 	}
 
 	abort();

@@ -187,9 +200,14 @@ static void child(void)
 	tst_res(TINFO, "Setting memory limits to %u %u", usage, 2 * usage);
 
 	SAFE_FILE_SCANF(limit_in_bytes_path, "%u", &old_limit);
-	SAFE_FILE_SCANF(memsw_limit_in_bytes_path, "%u", &old_memsw_limit);
+
+	if (swap_accounting_enabled)
+		SAFE_FILE_SCANF(memsw_limit_in_bytes_path, "%u", &old_memsw_limit);
+
 	SAFE_FILE_PRINTF(limit_in_bytes_path, "%u", usage);
-	SAFE_FILE_PRINTF(memsw_limit_in_bytes_path, "%u", 2 * usage);
+
+	if (swap_accounting_enabled)
+		SAFE_FILE_PRINTF(memsw_limit_in_bytes_path, "%u", 2 * usage);
 
 	do {
 		sleep_between_faults++;

@@ -245,7 +263,9 @@ static void child(void)
 	else
 		tst_res(TPASS, "All pages have expected content");
 
-	SAFE_FILE_PRINTF(memsw_limit_in_bytes_path, "%u", old_memsw_limit);
+	if (swap_accounting_enabled)
+		SAFE_FILE_PRINTF(memsw_limit_in_bytes_path, "%u", old_memsw_limit);
+
 	SAFE_FILE_PRINTF(limit_in_bytes_path, "%u", old_limit);
 
 	SAFE_MUNMAP(ptr, PAGES);

@@ -298,6 +318,11 @@ static void setup(void)
 			"' not present, CONFIG_MEMCG missing?");
 	}
 
+	if (!access(MEMCG_PATH "memory.memsw.limit_in_bytes", F_OK))
+		swap_accounting_enabled = 1;
+	else
+		tst_res(TINFO, "Swap accounting is disabled");
+
 	SAFE_FILE_LINES_SCANF("/proc/meminfo", "SwapTotal: %ld", &swap_total);
 	if (swap_total <= 0)
 		tst_brk(TCONF, "MADV_FREE does not work without swap");

Signature · 1 line

2.12.2

Jan Stancek

2017-05-03 08:14 UTC #1 lore raw

RFC PATCH 1/1 syscalls/madvise09: Fix for disabled swap accounting

----- Original Message -----

Quoted 10 lines

The test had to be fixed for systems with swap accounting disabled.
First there are no "memsw" prefixed tuning knobs in memcg directories if swap accounting is disabled.
Then we had to limit the hungry child since otherwise it will be killed only after all system swap was filled. Hence we exit the memory hungry child once the process fills 100MB of swap. That seems to be enough to put the cgroup into memory pressure while it limits the test runtime to a few seconds.

Looks reasonable to me, only 2 alternatives I can think of is turning off swap for duration of test, or exit with TCONF if swap accounting is disabled.

Regards,
Jan

Cyril Hrubis

author

2017-05-03 08:18 UTC #2 lore raw

RFC PATCH 1/1 syscalls/madvise09: Fix for disabled swap accounting

Hi!

Quoted 11 lines

Quoted 10 lines
The test had to be fixed for systems with swap accounting disabled.
First there are no "memsw" prefixed tuning knobs in memcg directories if swap accounting is disabled.
Then we had to limit the hungry child since otherwise it will be killed only after all system swap was filled. Hence we exit the memory hungry child once the process fills 100MB of swap. That seems to be enough to put the cgroup into memory pressure while it limits the test runtime to a few seconds.

Quoted 3 lines

Looks reasonable to me, only 2 alternatives I can think of is turning off swap for duration of test, or exit with TCONF if swap accounting is disabled.

We cannot turn off the swap since MADV_FREE does not work without it. So it's either TCONF or limiting the swap usage in the hungry child as this patch does...

Signature · 2 lines

Cyril Hrubis
[email protected]

Cyril Hrubis

author

2017-05-05 09:09 UTC #3 lore raw

RFC PATCH 1/1 syscalls/madvise09: Fix for disabled swap accounting

Hi!

Quoted 11 lines

Quoted 11 lines
Quoted 10 lines
The test had to be fixed for systems with swap accounting disabled.
First there are no "memsw" prefixed tuning knobs in memcg directories if swap accounting is disabled.
Then we had to limit the hungry child since otherwise it will be killed only after all system swap was filled. Hence we exit the memory hungry child once the process fills 100MB of swap. That seems to be enough to put the cgroup into memory pressure while it limits the test runtime to a few seconds.

Quoted 3 lines

Quoted 3 lines
Looks reasonable to me, only 2 alternatives I can think of is turning off swap for duration of test, or exit with TCONF if swap accounting is disabled.

Pushed with your ack.

Signature · 2 lines

Cyril Hrubis
[email protected]