When running mm selftest to verify mm patches, 'compaction_test' case failed on an x86 server with 1TB memory. And the root cause is that it has too much free memory than what the test supports.
The test case tries to allocate 100000 huge pages, which is about 200 GB for that x86 server, and when it succeeds, it expects it's large than 1/3 of 80% of the free memory in system. This logic only works for platform with 750 GB ( 200 / (1/3) / 80% ) or less free memory, and may raise false alarm for others.
Fix it by changing the fixed page number to self-adjustable number according to the real number of free memory.
Fixes: bd67d5c15cc19 ("Test compaction of mlocked memory") Signed-off-by: Feng Tang feng.tang@linux.alibaba.com --- tools/testing/selftests/mm/compaction_test.c | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-)
diff --git a/tools/testing/selftests/mm/compaction_test.c b/tools/testing/selftests/mm/compaction_test.c index 2c3a0eb6b22d..9bc4591c7b16 100644 --- a/tools/testing/selftests/mm/compaction_test.c +++ b/tools/testing/selftests/mm/compaction_test.c @@ -90,6 +90,8 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, int compaction_index = 0; char nr_hugepages[20] = {0}; char init_nr_hugepages[24] = {0}; + char target_nr_hugepages[24] = {0}; + int slen;
snprintf(init_nr_hugepages, sizeof(init_nr_hugepages), "%lu", initial_nr_hugepages); @@ -106,11 +108,18 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, goto out; }
- /* Request a large number of huge pages. The Kernel will allocate - as much as it can */ - if (write(fd, "100000", (6*sizeof(char))) != (6*sizeof(char))) { - ksft_print_msg("Failed to write 100000 to /proc/sys/vm/nr_hugepages: %s\n", - strerror(errno)); + /* + * Request huge pages for about half of the free memory. The Kernel + * will allocate as much as it can, and we expect it will get at least 1/3 + */ + nr_hugepages_ul = mem_free / hugepage_size / 2; + snprintf(target_nr_hugepages, sizeof(target_nr_hugepages), + "%lu", nr_hugepages_ul); + + slen = strlen(target_nr_hugepages); + if (write(fd, target_nr_hugepages, slen) != slen) { + ksft_print_msg("Failed to write %lu to /proc/sys/vm/nr_hugepages: %s\n", + nr_hugepages_ul, strerror(errno)); goto close_fd; }
On 23/04/25 4:06 pm, Feng Tang wrote:
When running mm selftest to verify mm patches, 'compaction_test' case failed on an x86 server with 1TB memory. And the root cause is that it has too much free memory than what the test supports.
The test case tries to allocate 100000 huge pages, which is about 200 GB for that x86 server, and when it succeeds, it expects it's large than 1/3 of 80% of the free memory in system. This logic only works for platform with 750 GB ( 200 / (1/3) / 80% ) or less free memory, and may raise false alarm for others.
Fix it by changing the fixed page number to self-adjustable number according to the real number of free memory.
Fixes: bd67d5c15cc19 ("Test compaction of mlocked memory") Signed-off-by: Feng Tang feng.tang@linux.alibaba.com
Not sure if fixes tag is needed.
Acked-by: Dev Jain dev.jain@arm.com
tools/testing/selftests/mm/compaction_test.c | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-)
diff --git a/tools/testing/selftests/mm/compaction_test.c b/tools/testing/selftests/mm/compaction_test.c index 2c3a0eb6b22d..9bc4591c7b16 100644 --- a/tools/testing/selftests/mm/compaction_test.c +++ b/tools/testing/selftests/mm/compaction_test.c @@ -90,6 +90,8 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, int compaction_index = 0; char nr_hugepages[20] = {0}; char init_nr_hugepages[24] = {0};
- char target_nr_hugepages[24] = {0};
- int slen;
snprintf(init_nr_hugepages, sizeof(init_nr_hugepages), "%lu", initial_nr_hugepages); @@ -106,11 +108,18 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, goto out; }
- /* Request a large number of huge pages. The Kernel will allocate
as much as it can */- if (write(fd, "100000", (6*sizeof(char))) != (6*sizeof(char))) {
ksft_print_msg("Failed to write 100000 to /proc/sys/vm/nr_hugepages: %s\n",strerror(errno));
- /*
* Request huge pages for about half of the free memory. The Kernel* will allocate as much as it can, and we expect it will get at least 1/3*/- nr_hugepages_ul = mem_free / hugepage_size / 2;
- snprintf(target_nr_hugepages, sizeof(target_nr_hugepages),
"%lu", nr_hugepages_ul);- slen = strlen(target_nr_hugepages);
- if (write(fd, target_nr_hugepages, slen) != slen) {
ksft_print_msg("Failed to write %lu to /proc/sys/vm/nr_hugepages: %s\n",nr_hugepages_ul, strerror(errno));
On Wed, Apr 23, 2025 at 04:33:19PM +0530, Dev Jain wrote:
On 23/04/25 4:06 pm, Feng Tang wrote:
When running mm selftest to verify mm patches, 'compaction_test' case failed on an x86 server with 1TB memory. And the root cause is that it has too much free memory than what the test supports.
The test case tries to allocate 100000 huge pages, which is about 200 GB for that x86 server, and when it succeeds, it expects it's large than 1/3 of 80% of the free memory in system. This logic only works for platform with 750 GB ( 200 / (1/3) / 80% ) or less free memory, and may raise false alarm for others.
Fix it by changing the fixed page number to self-adjustable number according to the real number of free memory.
Fixes: bd67d5c15cc19 ("Test compaction of mlocked memory") Signed-off-by: Feng Tang feng.tang@linux.alibaba.com
Not sure if fixes tag is needed.
Yep, I was not very sure either :). And I'm fine with dropping the tag.
Acked-by: Dev Jain dev.jain@arm.com
Many thanks for the review!
- Feng
tools/testing/selftests/mm/compaction_test.c | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-)
diff --git a/tools/testing/selftests/mm/compaction_test.c b/tools/testing/selftests/mm/compaction_test.c index 2c3a0eb6b22d..9bc4591c7b16 100644 --- a/tools/testing/selftests/mm/compaction_test.c +++ b/tools/testing/selftests/mm/compaction_test.c @@ -90,6 +90,8 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, int compaction_index = 0; char nr_hugepages[20] = {0}; char init_nr_hugepages[24] = {0};
- char target_nr_hugepages[24] = {0};
- int slen; snprintf(init_nr_hugepages, sizeof(init_nr_hugepages), "%lu", initial_nr_hugepages);
@@ -106,11 +108,18 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, goto out; }
- /* Request a large number of huge pages. The Kernel will allocate
as much as it can */- if (write(fd, "100000", (6*sizeof(char))) != (6*sizeof(char))) {
ksft_print_msg("Failed to write 100000 to /proc/sys/vm/nr_hugepages: %s\n",strerror(errno));
- /*
* Request huge pages for about half of the free memory. The Kernel* will allocate as much as it can, and we expect it will get at least 1/3*/- nr_hugepages_ul = mem_free / hugepage_size / 2;
- snprintf(target_nr_hugepages, sizeof(target_nr_hugepages),
"%lu", nr_hugepages_ul);- slen = strlen(target_nr_hugepages);
- if (write(fd, target_nr_hugepages, slen) != slen) {
ksft_print_msg("Failed to write %lu to /proc/sys/vm/nr_hugepages: %s\n",nr_hugepages_ul, strerror(errno));
On 2025/4/23 18:36, Feng Tang wrote:
When running mm selftest to verify mm patches, 'compaction_test' case failed on an x86 server with 1TB memory. And the root cause is that it has too much free memory than what the test supports.
The test case tries to allocate 100000 huge pages, which is about 200 GB for that x86 server, and when it succeeds, it expects it's large than 1/3 of 80% of the free memory in system. This logic only works for platform with 750 GB ( 200 / (1/3) / 80% ) or less free memory, and may raise false alarm for others.
Fix it by changing the fixed page number to self-adjustable number according to the real number of free memory.
Fixes: bd67d5c15cc19 ("Test compaction of mlocked memory") Signed-off-by: Feng Tang feng.tang@linux.alibaba.com
Make sense to me. I tested it and passed on my platform. So: Reviewed-by: Baolin Wang baolin.wang@linux.alibaba.com Tested-by: Baolin Wang baolin.wang@inux.alibaba.com
tools/testing/selftests/mm/compaction_test.c | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-)
diff --git a/tools/testing/selftests/mm/compaction_test.c b/tools/testing/selftests/mm/compaction_test.c index 2c3a0eb6b22d..9bc4591c7b16 100644 --- a/tools/testing/selftests/mm/compaction_test.c +++ b/tools/testing/selftests/mm/compaction_test.c @@ -90,6 +90,8 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, int compaction_index = 0; char nr_hugepages[20] = {0}; char init_nr_hugepages[24] = {0};
- char target_nr_hugepages[24] = {0};
- int slen;
snprintf(init_nr_hugepages, sizeof(init_nr_hugepages), "%lu", initial_nr_hugepages); @@ -106,11 +108,18 @@ int check_compaction(unsigned long mem_free, unsigned long hugepage_size, goto out; }
- /* Request a large number of huge pages. The Kernel will allocate
as much as it can */- if (write(fd, "100000", (6*sizeof(char))) != (6*sizeof(char))) {
ksft_print_msg("Failed to write 100000 to /proc/sys/vm/nr_hugepages: %s\n",strerror(errno));
- /*
* Request huge pages for about half of the free memory. The Kernel* will allocate as much as it can, and we expect it will get at least 1/3*/- nr_hugepages_ul = mem_free / hugepage_size / 2;
- snprintf(target_nr_hugepages, sizeof(target_nr_hugepages),
"%lu", nr_hugepages_ul);- slen = strlen(target_nr_hugepages);
- if (write(fd, target_nr_hugepages, slen) != slen) {
ksft_print_msg("Failed to write %lu to /proc/sys/vm/nr_hugepages: %s\n",nr_hugepages_ul, strerror(errno));
linux-kselftest-mirror@lists.linaro.org
-
Baolin Wang -
Dev Jain -
Feng Tang