Re: [PATCH v2 2/2] selftest/mm: ksm_functional_tests: extend test case for ksm fork/exec

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

 



On 25.03.24 03:24, Jinjiang Tu wrote:

在 2024/3/22 19:43, David Hildenbrand 写道:
On 22.03.24 07:09, Jinjiang Tu wrote:
This extends test_prctl_fork() and test_prctl_fork_exec() to make sure
that deduplication really happens, instead of only test the
MMF_VM_MERGE_ANY flag is set.

Signed-off-by: Jinjiang Tu <tujinjiang@xxxxxxxxxx>
---
   .../selftests/mm/ksm_functional_tests.c       | 79 +++++++++++++++++--
   1 file changed, 73 insertions(+), 6 deletions(-)

diff --git a/tools/testing/selftests/mm/ksm_functional_tests.c
b/tools/testing/selftests/mm/ksm_functional_tests.c
index d615767e396b..01999aab2e37 100644
--- a/tools/testing/selftests/mm/ksm_functional_tests.c
+++ b/tools/testing/selftests/mm/ksm_functional_tests.c
@@ -146,6 +146,54 @@ static int ksm_unmerge(void)
       return 0;
   }
   +static int child_test_merge(void)
+{
+    const unsigned int size = 2 * MiB;
+    char *map;
+    int ret = -1;
+
+    /* Stabilize accounting by disabling KSM completely. */
+    if (ksm_unmerge()) {
+        ksft_print_msg("Disabling (unmerging) KSM failed\n");
+        return ret;
+    }
+
+    if (get_my_merging_pages() > 0) {
+        ksft_print_msg("Still pages merged\n");
+        return ret;
+    }
+
+    map = mmap(NULL, size, PROT_READ|PROT_WRITE,
+           MAP_PRIVATE|MAP_ANON, -1, 0);
+    if (map == MAP_FAILED) {
+        ksft_print_msg("mmap() failed\n");
+        return ret;
+    }
+
+    /* Don't use THP. Ignore if THP are not around on a kernel. */
+    if (madvise(map, size, MADV_NOHUGEPAGE) && errno != EINVAL) {
+        ksft_print_msg("MADV_NOHUGEPAGE failed\n");
+        goto unmap;
+    }
+
+    memset(map, 0x1c, size);
+
+    if (ksm_merge()) {
+        ksft_print_msg("Running KSM failed\n");
+        goto unmap;
+    }
+
+    if (get_my_merging_pages() <= 0) {
+        ksft_print_msg("Fail to merge\n");
+        goto unmap;
+    }

Looks like all you want is use mmap_and_merge_range(), but neither
setting the prctl nor madvise().

Two alternatives:

1) switching from "bool use_prctl" to an enum like

enum ksm_merge_mode {
     KSM_MERGE_PRCTL
     KSM_MERGE_MADVISE,
     KSM_MERGE_NONE, /* PRCTL already set */
};

Then, you can simply use mmap_and_merge_range(0x1c, 2 * MiB,
PROT_READ|PROT_WRITE, KSM_MERGE_NONE);
I have considered this before. But, mmap_and_merge_range() calls
ksft_test_result_fail() when error occurs, ksft_test_result_fail()
prints prefixed with ksft_fail count. When mmap_and_merge_range() is
called in the child process, the ksft_fail isn't consisent with the
parent process due to the global variable ksft_fail is CoWed. As a
result, ksft_print_msg() is intended to be called in child process.

Maybe, We could introduce a macro ksm_print() to control which function
is called according to ksm_merge_mode :

As an alternative, convert all ksft_test_result_fail() in there into ksft_print_msg(), and in the callers of mmap_and_merge_range(), do something like

map = mmap_and_merge_range() ...
if (map == MAP_FAILED) {
	ksft_test_result_fail("Merging memory failed");
	return;
}


--
Cheers,

David / dhildenb





[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux