On Sun, Aug 1, 2021 at 11:43 AM Zhansaya Bagdauletkyzy zhansayabagdaulet@gmail.com wrote:

Add ksm_merge_time() function to determine speed and time needed for merging. The total spent time is shown in seconds while speed is in MB/s. User must specify the size of duplicated memory area (in MB) before running the test.

The test is run as follows: ./ksm_tests -P -s 100 The output: Total size: 100 MB Total time: 0.309561 s Average speed: 323 MB/s

Signed-off-by: Zhansaya Bagdauletkyzy zhansayabagdaulet@gmail.com

tools/testing/selftests/vm/ksm_tests.c | 76 ++++++++++++++++++++++++-- 1 file changed, 72 insertions(+), 4 deletions(-)

diff --git a/tools/testing/selftests/vm/ksm_tests.c b/tools/testing/selftests/vm/ksm_tests.c index cdeb4a028538..91c6ff496655 100644 --- a/tools/testing/selftests/vm/ksm_tests.c +++ b/tools/testing/selftests/vm/ksm_tests.c @@ -7,6 +7,7 @@ #include <numa.h>

#include "../kselftest.h" +#include "../../../../include/vdso/time64.h"

#define KSM_SYSFS_PATH "/sys/kernel/mm/ksm/" #define KSM_FP(s) (KSM_SYSFS_PATH s) @@ -15,6 +16,7 @@ #define KSM_PROT_STR_DEFAULT "rw" #define KSM_USE_ZERO_PAGES_DEFAULT false #define KSM_MERGE_ACROSS_NODES_DEFAULT true +#define MB_TO_B 1000000ul

We should calculate everything in true megabytes.

#define MB (1ul << 20)

Also, print in MiB/s.

struct ksm_sysfs { unsigned long max_page_sharing; @@ -30,7 +32,8 @@ enum ksm_test_name { CHECK_KSM_MERGE, CHECK_KSM_UNMERGE, CHECK_KSM_ZERO_PAGE_MERGE,

•   CHECK_KSM_NUMA_MERGE
  

•   CHECK_KSM_NUMA_MERGE,
  

•   KSM_MERGE_TIME
  

};

static int ksm_write_sysfs(const char *file_path, unsigned long val) @@ -86,13 +89,16 @@ static int str_to_prot(char *prot_str) static void print_help(void) { printf("usage: ksm_tests [-h] <test type> [-a prot] [-p page_count] [-l timeout]\n"

•          "[-z use_zero_pages] [-m merge_across_nodes]\n");
  

•          "[-z use_zero_pages] [-m merge_across_nodes] [-s size]\n");
  
    printf("Supported <test type>:\n"
           " -M (page merging)\n"
           " -Z (zero pages merging)\n"
           " -N (merging of pages in different NUMA nodes)\n"
  

•          " -U (page unmerging)\n\n");
  

•          " -U (page unmerging)\n"
  

•          " -P evaluate merging time and speed.\n"
  

•          "    For this test, the size of duplicated memory area (in MB)\n"
  

•          "    must be provided using -s option\n\n");
  
    printf(" -a: specify the access protections of pages.\n"
           "     <prot> must be of the form [rwx].\n"
  

@@ -105,6 +111,7 @@ static void print_help(void) " Default: %d\n", KSM_USE_ZERO_PAGES_DEFAULT); printf(" -m: change merge_across_nodes tunable\n" " Default: %d\n", KSM_MERGE_ACROSS_NODES_DEFAULT);

•   printf(" -s: the size of duplicated memory area (in MB)\n");
  
    exit(0);
  

} @@ -407,6 +414,49 @@ static int check_ksm_numa_merge(int mapping, int prot, int timeout, bool merge_a return KSFT_FAIL; }

+static int ksm_merge_time(int mapping, int prot, int timeout, size_t map_size) +{

•   void *map_ptr;
  

•   struct timespec start_time, end_time;
  

•   long scan_time_s, scan_time_ns;
  

Should be "unsigned long".

•   int avg_speed;
  

No needed, see below.

•   map_size *= MB_TO_B;
  

•   map_ptr = allocate_memory(NULL, prot, mapping, '*', map_size);
  

•   if (!map_ptr)
  

•           return KSFT_FAIL;
  

•   if (clock_gettime(CLOCK_MONOTONIC_RAW, &start_time)) {
  

•           perror("clock_gettime");
  

•           goto err_out;
  

•   }
  

•   if (ksm_merge_pages(map_ptr, map_size, start_time, timeout))
  

•           goto err_out;
  

•   if (clock_gettime(CLOCK_MONOTONIC_RAW, &end_time)) {
  

•           perror("clock_gettime");
  

•           goto err_out;
  

•   }
  

•   scan_time_ns = (end_time.tv_sec - start_time.tv_sec) * NSEC_PER_SEC +
  

•                  (end_time.tv_nsec - start_time.tv_nsec);
  

•   scan_time_s = scan_time_ns / NSEC_PER_SEC;
  

•   scan_time_ns %= NSEC_PER_SEC;
  

•   avg_speed = map_size / (scan_time_s * USEC_PER_SEC + scan_time_ns / NSEC_PER_USEC);
  

Please remove the above three lines, and replace with bellow suggestions:

•   printf("Total size:    %lu MB\n", map_size / MB_TO_B);
  

printf("Total size: %lu MiB\n", map_size / MB);

•   printf("Total time:    %ld.%06ld s\n", scan_time_s, scan_time_ns / NSEC_PER_USEC);
  

printf("Total time: %ld.%09ld s\n", scan_time_ns / NSEC_PER_SEC, scan_time_ns % NSEC_PER_SEC));

•   printf("Average speed: %d MB/s\n", avg_speed);
  

The following would give a much better precision: printf("Average speed: %.3f MiB/s\n", (map_size / MB) / ((double)scan_time_ns / NSEC_PER_SEC));

•   munmap(map_ptr, map_size);
  

•   return KSFT_PASS;
  

+err_out:

•   printf("Not OK\n");
  

•   munmap(map_ptr, map_size);
  

•   return KSFT_FAIL;
  

+}

int main(int argc, char *argv[]) { int ret, opt; @@ -418,8 +468,9 @@ int main(int argc, char *argv[]) int test_name = CHECK_KSM_MERGE; bool use_zero_pages = KSM_USE_ZERO_PAGES_DEFAULT; bool merge_across_nodes = KSM_MERGE_ACROSS_NODES_DEFAULT;

•   long size_MB = 0;
  
  

•   while ((opt = getopt(argc, argv, "ha:p:l:z:m:MUZN")) != -1) {
  

•   while ((opt = getopt(argc, argv, "ha:p:l:z:m:s:MUZNP")) != -1) {
            switch (opt) {
            case 'a':
                    prot = str_to_prot(optarg);
  

@@ -453,6 +504,12 @@ int main(int argc, char *argv[]) else merge_across_nodes = 1; break;

•           case 's':
  

•                   size_MB = atoi(optarg);
  

•                   if (size_MB <= 0) {
  

•                           printf("Size must be greater than 0\n");
  

•                           return KSFT_FAIL;
  

•                   }
            case 'M':
                    break;
            case 'U':
  

@@ -464,6 +521,9 @@ int main(int argc, char *argv[]) case 'N': test_name = CHECK_KSM_NUMA_MERGE; break;

•           case 'P':
  

•                   test_name = KSM_MERGE_TIME;
  

•                   break;
            default:
                    return KSFT_FAIL;
            }
  

@@ -505,6 +565,14 @@ int main(int argc, char *argv[]) ret = check_ksm_numa_merge(MAP_PRIVATE | MAP_ANONYMOUS, prot, ksm_scan_limit_sec, merge_across_nodes, page_size); break;

•   case KSM_MERGE_TIME:
  

•           if (size_MB == 0) {
  

•                   printf("Option '-s' is required.\n");
  

•                   return KSFT_FAIL;
  

•           }
  

•           ret = ksm_merge_time(MAP_PRIVATE | MAP_ANONYMOUS, prot, ksm_scan_limit_sec,
  

•                                size_MB);
  

•           break;
    }
  
    if (ksm_restore(&ksm_sysfs_old)) {
  

-- 2.25.1

On Sun, Aug 1, 2021 at 11:43 AM Zhansaya Bagdauletkyzy zhansayabagdaulet@gmail.com wrote:

Since merged pages are copied every time they need to be modified, the write access time is different between shared and non-shared pages. Add ksm_cow_time() function which evaluates latency of these COW breaks. First, duplicated pages are merged and then the time required to write to each of the pages is detected.

Hi Zhansaya,

This test would make more sense if we also had a baseline on how long it takes to modify unmerged pages. This way it would show the true cost of having a write on a previously merged page vs. if it was never merged.

Also, instead of having all pages merged together, I'd rather see a collection of merged pairs of pages, which are decoupled as one of the pages gets modified: i.e. Page1 - Page2: merged Page3 - Page4: merged ...

then, go through pages: Page2, Page4, ... and modify the first byte in each of them and measure the time it takes. Do the same when these pages are not merged, and print both numbers so the cost of KSM and COW can be calculated.

Thanks, Pasha

The test is run as follows: ./ksm_tests -C -p 5000 The output: Total COW time: 0.012612 s The number of pages: 5000 Average speed: 1623 MB/s

Signed-off-by: Zhansaya Bagdauletkyzy zhansayabagdaulet@gmail.com

tools/testing/selftests/vm/ksm_tests.c | 66 ++++++++++++++++++++++++-- 1 file changed, 63 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/vm/ksm_tests.c b/tools/testing/selftests/vm/ksm_tests.c index 91c6ff496655..3ffd899b2bd9 100644 --- a/tools/testing/selftests/vm/ksm_tests.c +++ b/tools/testing/selftests/vm/ksm_tests.c @@ -33,7 +33,8 @@ enum ksm_test_name { CHECK_KSM_UNMERGE, CHECK_KSM_ZERO_PAGE_MERGE, CHECK_KSM_NUMA_MERGE,

•   KSM_MERGE_TIME
  

•   KSM_MERGE_TIME,
  

•   KSM_COW_TIME
  

};

static int ksm_write_sysfs(const char *file_path, unsigned long val) @@ -98,7 +99,9 @@ static void print_help(void) " -U (page unmerging)\n" " -P evaluate merging time and speed.\n" " For this test, the size of duplicated memory area (in MB)\n"

•          "    must be provided using -s option\n\n");
  

•          "    must be provided using -s option\n"
  

•          " -C evaluate the time required to break COW of merged pages.\n"
  

•          "    The number of pages can be defined using -p option.\n\n");
  
    printf(" -a: specify the access protections of pages.\n"
           "     <prot> must be of the form [rwx].\n"
  

@@ -457,6 +460,56 @@ static int ksm_merge_time(int mapping, int prot, int timeout, size_t map_size) return KSFT_FAIL; }

+static int ksm_cow_time(int mapping, int prot, int timeout, size_t page_size, long page_count) +{

•   void *map_ptr;
  

•   struct timespec start_time, end_time;
  

•   long cow_time_s, cow_time_ns;
  

•   int avg_speed;
  

•   if (clock_gettime(CLOCK_MONOTONIC_RAW, &start_time)) {
  

•           perror("clock_gettime");
  

•           return KSFT_FAIL;
  

•   }
  

•   map_ptr = allocate_memory(NULL, prot, mapping, '*', page_size * page_count);
  

•   if (!map_ptr)
  

•           return KSFT_FAIL;
  

•   if (ksm_merge_pages(map_ptr, page_size * page_count, start_time, timeout))
  

•           goto err_out;
  

•   if (clock_gettime(CLOCK_MONOTONIC_RAW, &start_time)) {
  

•           perror("clock_gettime");
  

•           goto err_out;
  

•   }
  

•   for (size_t i = 0; i < page_count; i++)
  

•           memset(map_ptr + page_size * i, '-', 1);
  

•   if (clock_gettime(CLOCK_MONOTONIC_RAW, &end_time)) {
  

•           perror("clock_gettime");
  

•           goto err_out;
  

•   }
  

•   cow_time_ns = (end_time.tv_sec - start_time.tv_sec) * NSEC_PER_SEC +
  

•                  (end_time.tv_nsec - start_time.tv_nsec);
  

•   cow_time_s = cow_time_ns / NSEC_PER_SEC;
  

•   cow_time_ns %= NSEC_PER_SEC;
  

•   avg_speed = (page_size * page_count) / (cow_time_s * USEC_PER_SEC +
  

•                                           cow_time_ns / NSEC_PER_USEC);
  

•   printf("Total COW time:      %ld.%06ld s\n", cow_time_s, cow_time_ns / NSEC_PER_USEC);
  

•   printf("The number of pages: %ld\n", page_count);
  

•   printf("Average speed:       %d MB/s\n", avg_speed);
  

•   munmap(map_ptr, page_size * page_count);
  

•   return KSFT_PASS;
  

+err_out:

•   printf("Not OK\n");
  

•   munmap(map_ptr, page_size * page_count);
  

•   return KSFT_FAIL;
  

+}

int main(int argc, char *argv[]) { int ret, opt; @@ -470,7 +523,7 @@ int main(int argc, char *argv[]) bool merge_across_nodes = KSM_MERGE_ACROSS_NODES_DEFAULT; long size_MB = 0;

•   while ((opt = getopt(argc, argv, "ha:p:l:z:m:s:MUZNP")) != -1) {
  

•   while ((opt = getopt(argc, argv, "ha:p:l:z:m:s:MUZNPC")) != -1) {
            switch (opt) {
            case 'a':
                    prot = str_to_prot(optarg);
  

@@ -524,6 +577,9 @@ int main(int argc, char *argv[]) case 'P': test_name = KSM_MERGE_TIME; break;

•           case 'C':
  

•                   test_name = KSM_COW_TIME;
  

•                   break;
            default:
                    return KSFT_FAIL;
            }
  

@@ -573,6 +629,10 @@ int main(int argc, char *argv[]) ret = ksm_merge_time(MAP_PRIVATE | MAP_ANONYMOUS, prot, ksm_scan_limit_sec, size_MB); break;

•   case KSM_COW_TIME:
  

•           ret = ksm_cow_time(MAP_PRIVATE | MAP_ANONYMOUS, prot, ksm_scan_limit_sec,
  

•                              page_size, page_count);
  

•           break;
    }
  
    if (ksm_restore(&ksm_sysfs_old)) {
  

-- 2.25.1