On Mon, Sep 07, 2026 at 09:15:08AM +0100, Yeoreum Yun wrote: > HPAGE_SIZE and HPAGE_SHIFT macro is written based on the 4KB PAGE_SIZE. > When this macro is used in some test, test result would be strange in > the system where PAGE_SIZE is more then 4KB. > > Here is the example with transhuge-stress test with 16KB PAGE_SIZE: > > transhuge-stress: allocate 61073 transhuge pages, using 122146 MiB virtual > memory and 1908 MiB of ram > 3.292 s/loop, 0.054 ms/page, 37106.002 MiB/s 2566 succeed, 58507 > failed, 2566 different pages > 0.591 s/loop, 0.010 ms/page, 206850.792 MiB/s 0 succeed, 61073 > failed, 0 different pages > 0.527 s/loop, 0.009 ms/page, 231895.107 MiB/s 0 succeed, 61073 > failed, 0 different pages > 0.527 s/loop, 0.009 ms/page, 231839.704 MiB/s 0 succeed, 61073 > failed, 0 different pages > 0.528 s/loop, 0.009 ms/page, 231544.782 MiB/s 0 succeed, 61073 > failed, 0 different pages > 0.528 s/loop, 0.009 ms/page, 231462.074 MiB/s 0 succeed, 61073 > failed, 0 different pages > 0.527 s/loop, 0.009 ms/page, 231770.300 MiB/s 0 succeed, 61073 > failed, 0 different pages > ... > ok 1 Completed > > Remove the HPAGE_SIZE and HPAGE_SHIFT macros and introduce pmd_page_shift() > helper to get the HPAGE_SHIFT properly. For HPAGE_SIZE, use pre-existing > helper, read_pmd_pagesize(). > > Also, run the KSM_MERGE_TIME_HUGE_PAGES test with a size of 512 MiB, > which is the least common multiple of the PMD sizes for 4 KiB, 16 KiB, > and 64 KiB base pages. Since allocate_transhuge() allocates mappings in > PMD-sized units, the test may fail with the previous size of 100 MiB, > which is not a multiple of the PMD size when the base page size is > 16 KiB or 64 KiB. > > After this patch, output of transhuge-stress: > > transhuge-stress: allocate 3817 transhuge pages, using 122146 MiB virtual > memory and 119 MiB of ram > 2.558 s/loop, 0.670 ms/page, 47755.759 MiB/s 2585 succeed, 1232 > failed, 2585 different pages > 2.640 s/loop, 0.692 ms/page, 46268.432 MiB/s 2585 succeed, 1232 > failed, 2585 different pages > 2.635 s/loop, 0.690 ms/page, 46360.298 MiB/s 2585 succeed, 1232 > failed, 2585 different pages > 2.782 s/loop, 0.729 ms/page, 43899.795 MiB/s 2616 succeed, 1201 > failed, 2616 different pages > 2.692 s/loop, 0.705 ms/page, 45380.876 MiB/s 2627 succeed, 1190 > failed, 2627 different pages > 2.612 s/loop, 0.684 ms/page, 46765.812 MiB/s 2628 succeed, 1189 > failed, 2628 different pages > 2.683 s/loop, 0.703 ms/page, 45520.990 MiB/s 2630 succeed, 1187 > failed, 2630 different pages > 2.727 s/loop, 0.714 ms/page, 44789.321 MiB/s 2631 succeed, 1186 > failed, 2631 different pages > ... > ok 1 Completed > > Suggested-by: Lorenzo Stoakes (ARM) <[email protected]> > Signed-off-by: Yeoreum Yun <[email protected]>
Tested and works locally :) (weird stall in data below is because it was being run in a VM that was given too much memory :P): $ sudo ./transhuge-stress -d 20 TAP version 13 1..1 transhuge-stress: allocate 251 transhuge pages, using 8041 MiB virtual memory and 7 MiB of ram # 2.317 s/loop, 9.230 ms/page, 3467.040 MiB/s 251 succeed, 0 failed, 233 different pages # 0.175 s/loop, 0.696 ms/page, 46008.352 MiB/s 251 succeed, 0 failed, 235 different pages # 0.166 s/loop, 0.661 ms/page, 48416.410 MiB/s 251 succeed, 0 failed, 232 different pages # 3.367 s/loop, 13.416 ms/page, 2385.195 MiB/s 251 succeed, 0 failed, 235 different pages # 10.092 s/loop, 40.206 ms/page, 795.906 MiB/s 251 succeed, 0 failed, 231 different pages # 0.190 s/loop, 0.757 ms/page, 42252.947 MiB/s 251 succeed, 0 failed, 230 different pages # 0.170 s/loop, 0.678 ms/page, 47177.691 MiB/s 251 succeed, 0 failed, 231 different pages # 0.259 s/loop, 1.032 ms/page, 30995.582 MiB/s 251 succeed, 0 failed, 230 different pages # 1.416 s/loop, 5.642 ms/page, 5671.301 MiB/s 251 succeed, 0 failed, 230 different pages # 0.170 s/loop, 0.677 ms/page, 47252.908 MiB/s 251 succeed, 0 failed, 229 different pages # 0.166 s/loop, 0.662 ms/page, 48339.908 MiB/s 251 succeed, 0 failed, 228 different pages # 0.165 s/loop, 0.659 ms/page, 48572.554 MiB/s 251 succeed, 0 failed, 231 different pages # 0.164 s/loop, 0.655 ms/page, 48841.927 MiB/s 251 succeed, 0 failed, 228 different pages # 0.216 s/loop, 0.861 ms/page, 37184.310 MiB/s 251 succeed, 0 failed, 189 different pages # 0.175 s/loop, 0.698 ms/page, 45864.590 MiB/s 251 succeed, 0 failed, 233 different pages ok 1 Completed # Totals: pass:1 fail:0 xfail:0 xpass:0 skip:0 error:0 Thanks for this! Tested-by: Lorenzo Stoakes (ARM) <[email protected]> Reviewed-by: Lorenzo Stoakes (ARM) <[email protected]> > --- > Chnage in v2: > - rename hpage_* to pmd_page_* > - Link to v1: > https://lore.kernel.org/all/[email protected]/#t > > --- > tools/testing/selftests/mm/guard-regions.c | 11 ++++-- > tools/testing/selftests/mm/ksm_tests.c | 16 +++++---- > tools/testing/selftests/mm/run_vmtests.sh | 4 +-- > tools/testing/selftests/mm/transhuge-stress.c | 25 +++++++------ > tools/testing/selftests/mm/vm_util.c | 36 ++++++++++++++++--- > tools/testing/selftests/mm/vm_util.h | 9 ++--- > 6 files changed, 71 insertions(+), 30 deletions(-) > > diff --git a/tools/testing/selftests/mm/guard-regions.c > b/tools/testing/selftests/mm/guard-regions.c > index 5c8ec3ca75d7..e57cc7e39e41 100644 > --- a/tools/testing/selftests/mm/guard-regions.c > +++ b/tools/testing/selftests/mm/guard-regions.c > @@ -2195,14 +2195,21 @@ TEST_F(guard_regions, pagemap_scan) > TEST_F(guard_regions, collapse) > { > const unsigned long page_size = self->page_size; > - const unsigned long size = 2 * HPAGE_SIZE; > - const unsigned long num_pages = size / page_size; > + const uint64_t hpage_size = read_pmd_pagesize(); > + unsigned long size; > + unsigned long num_pages; > char *ptr; > int i; > > if (!thp_available()) > SKIP(return, "Transparent Hugepages not available\n"); > > + if (!hpage_size) > + ksft_exit_fail_msg("reading hpage_size failed\n"); > + > + size = 2 * hpage_size; > + num_pages = size / page_size; > + > /* Need file to be correct size for tests for non-anon. */ > if (variant->backing != ANON_BACKED) > ASSERT_EQ(ftruncate(self->fd, size), 0); > diff --git a/tools/testing/selftests/mm/ksm_tests.c > b/tools/testing/selftests/mm/ksm_tests.c > index 5fd7792a0d47..ef27ce467b8c 100644 > --- a/tools/testing/selftests/mm/ksm_tests.c > +++ b/tools/testing/selftests/mm/ksm_tests.c > @@ -521,19 +521,23 @@ static int ksm_merge_hugepages_time(int merge_type, int > mapping, int prot, > struct timespec start_time, end_time; > unsigned long scan_time_ns; > int pagemap_fd, n_normal_pages, n_huge_pages; > + const uint64_t hpage_size = read_pmd_pagesize(); > > if (!thp_is_enabled()) { > ksft_print_msg("Transparent Hugepages not available\n"); > return KSFT_SKIP; > } > > + if (!hpage_size) > + ksft_exit_fail_msg("reading hpage_size failed\n"); > + > map_size *= MB; > size_t len = map_size; > > - len -= len % HPAGE_SIZE; > - map_ptr_orig = mmap(NULL, len + HPAGE_SIZE, PROT_READ | PROT_WRITE, > + len -= len % hpage_size; > + map_ptr_orig = mmap(NULL, len + hpage_size, PROT_READ | PROT_WRITE, > MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE, -1, 0); > - map_ptr = map_ptr_orig + HPAGE_SIZE - (uintptr_t)map_ptr_orig % > HPAGE_SIZE; > + map_ptr = map_ptr_orig + hpage_size - (uintptr_t)map_ptr_orig % > hpage_size; > > if (map_ptr_orig == MAP_FAILED) > err(2, "initial mmap"); > @@ -547,7 +551,7 @@ static int ksm_merge_hugepages_time(int merge_type, int > mapping, int prot, > > n_normal_pages = 0; > n_huge_pages = 0; > - for (void *p = map_ptr; p < map_ptr + len; p += HPAGE_SIZE) { > + for (void *p = map_ptr; p < map_ptr + len; p += hpage_size) { > if (allocate_transhuge(p, pagemap_fd) < 0) > n_normal_pages++; > else > @@ -578,11 +582,11 @@ static int ksm_merge_hugepages_time(int merge_type, int > mapping, int prot, > ksft_print_msg("Average speed: %.3f MiB/s\n", (map_size / MB) / > ((double)scan_time_ns / > NSEC_PER_SEC)); > > - munmap(map_ptr_orig, len + HPAGE_SIZE); > + munmap(map_ptr_orig, len + hpage_size); > return KSFT_PASS; > > err_out: > - munmap(map_ptr_orig, len + HPAGE_SIZE); > + munmap(map_ptr_orig, len + hpage_size); > return KSFT_FAIL; > } > > diff --git a/tools/testing/selftests/mm/run_vmtests.sh > b/tools/testing/selftests/mm/run_vmtests.sh > index d09f9f6a384e..24aa1de27c25 100755 > --- a/tools/testing/selftests/mm/run_vmtests.sh > +++ b/tools/testing/selftests/mm/run_vmtests.sh > @@ -360,8 +360,8 @@ fi > CATEGORY="memfd_secret" run_test ./memfd_secret > fi > > -# KSM KSM_MERGE_TIME_HUGE_PAGES test with size of 100 > -CATEGORY="ksm" run_test ./ksm_tests -H -s 100 > +# KSM KSM_MERGE_TIME_HUGE_PAGES test with size of 512 > +CATEGORY="ksm" run_test ./ksm_tests -H -s 512 > # KSM KSM_MERGE_TIME test with size of 100 > CATEGORY="ksm" run_test ./ksm_tests -P -s 100 > # KSM MADV_MERGEABLE test with 10 identical pages > diff --git a/tools/testing/selftests/mm/transhuge-stress.c > b/tools/testing/selftests/mm/transhuge-stress.c > index 8eb0c5630e7e..9cd9790224db 100644 > --- a/tools/testing/selftests/mm/transhuge-stress.c > +++ b/tools/testing/selftests/mm/transhuge-stress.c > @@ -35,12 +35,17 @@ int main(int argc, char **argv) > size_t map_len; > int pagemap_fd; > int duration = 0; > + const uint64_t hpage_size = read_pmd_pagesize(); > + const uint64_t hpage_shift = pmd_page_shift(); > > ksft_print_header(); > > if (!thp_is_enabled()) > ksft_exit_skip("Transparent Hugepages not available\n"); > > + if (!hpage_size || !hpage_shift) > + ksft_exit_fail_msg("reading hpage_size failed\n"); > + > ram = sysconf(_SC_PHYS_PAGES); > if (ram > SIZE_MAX / psize() / 4) > ram = SIZE_MAX / 4; > @@ -70,23 +75,23 @@ int main(int argc, char **argv) > } > > warnx("allocate %zd transhuge pages, using %zd MiB virtual memory" > - " and %zd MiB of ram", len >> HPAGE_SHIFT, len >> 20, > - ram >> (20 + HPAGE_SHIFT - pshift() - 1)); > + " and %zd MiB of ram", len >> hpage_shift, len >> 20, > + ram >> (20 + hpage_shift - pshift() - 1)); > > pagemap_fd = open("/proc/self/pagemap", O_RDONLY); > if (pagemap_fd < 0) > ksft_exit_fail_msg("open pagemap\n"); > > - len -= len % HPAGE_SIZE; > - ptr = mmap(NULL, len + HPAGE_SIZE, PROT_RW, mmap_flags, backing_fd, 0); > + len -= len % hpage_size; > + ptr = mmap(NULL, len + hpage_size, PROT_RW, mmap_flags, backing_fd, 0); > if (ptr == MAP_FAILED) > ksft_exit_fail_msg("initial mmap"); > - ptr += HPAGE_SIZE - (uintptr_t)ptr % HPAGE_SIZE; > + ptr += hpage_size - (uintptr_t)ptr % hpage_size; > > if (madvise(ptr, len, MADV_HUGEPAGE)) > ksft_exit_fail_msg("MADV_HUGEPAGE"); > > - map_len = ram >> (HPAGE_SHIFT - 1); > + map_len = ram >> (hpage_shift - 1); > map = malloc(map_len); > if (!map) > ksft_exit_fail_msg("map malloc\n"); > @@ -99,7 +104,7 @@ int main(int argc, char **argv) > memset(map, 0, map_len); > > clock_gettime(CLOCK_MONOTONIC, &a); > - for (p = ptr; p < ptr + len; p += HPAGE_SIZE) { > + for (p = ptr; p < ptr + len; p += hpage_size) { > int64_t pfn; > > pfn = allocate_transhuge(p, pagemap_fd); > @@ -107,7 +112,7 @@ int main(int argc, char **argv) > if (pfn < 0) { > nr_failed++; > } else { > - size_t idx = pfn >> (HPAGE_SHIFT - pshift()); > + size_t idx = pfn >> (hpage_shift - pshift()); > > nr_succeed++; > if (idx >= map_len) { > @@ -123,7 +128,7 @@ int main(int argc, char **argv) > } > > /* split transhuge page, keep last page */ > - if (madvise(p, HPAGE_SIZE - psize(), MADV_DONTNEED)) > + if (madvise(p, hpage_size - psize(), MADV_DONTNEED)) > ksft_exit_fail_msg("MADV_DONTNEED"); > } > clock_gettime(CLOCK_MONOTONIC, &b); > @@ -131,7 +136,7 @@ int main(int argc, char **argv) > > ksft_print_msg("%.3f s/loop, %.3f ms/page, %10.3f MiB/s\t" > "%4d succeed, %4d failed, %4d different pages\n", > - s, s * 1000 / (len >> HPAGE_SHIFT), len / s / (1 > << 20), > + s, s * 1000 / (len >> hpage_shift), len / s / (1 > << 20), > nr_succeed, nr_failed, nr_pages); > > if (duration > 0 && b.tv_sec - start.tv_sec >= duration) { > diff --git a/tools/testing/selftests/mm/vm_util.c > b/tools/testing/selftests/mm/vm_util.c > index 4821a3563036..f75bf0bacd18 100644 > --- a/tools/testing/selftests/mm/vm_util.c > +++ b/tools/testing/selftests/mm/vm_util.c > @@ -21,6 +21,8 @@ > > unsigned int __page_size; > unsigned int __page_shift; > +uint64_t __pmd_page_size; > +uint64_t __pmd_page_shift; > > uint64_t pagemap_get_entry(int fd, char *start) > { > @@ -161,6 +163,9 @@ uint64_t read_pmd_pagesize(void) > char buf[20]; > ssize_t num_read; > > + if (__pmd_page_size) > + return __pmd_page_size; > + > fd = open(PMD_SIZE_FILE_PATH, O_RDONLY); > if (fd == -1) > return 0; > @@ -173,7 +178,25 @@ uint64_t read_pmd_pagesize(void) > buf[num_read] = '\0'; > close(fd); > > - return strtoul(buf, NULL, 10); > + __pmd_page_size = strtoul(buf, NULL, 10); > + > + return __pmd_page_size; > +} > + > +uint64_t pmd_page_shift(void) > +{ > + if (__pmd_page_shift) > + return __pmd_page_shift; > + > + if (!__pmd_page_size) > + __pmd_page_size = read_pmd_pagesize(); > + > + if (!__pmd_page_size) > + return 0; > + > + __pmd_page_shift = (ffsl(__pmd_page_size) - 1); > + > + return __pmd_page_shift; > } > > unsigned long rss_anon(void) > @@ -452,14 +475,19 @@ bool check_huge_shmem(void *addr, size_t len, int > nr_hpages, uint64_t hpage_size > int64_t allocate_transhuge(void *ptr, int pagemap_fd) > { > uint64_t ent[2]; > + const uint64_t hpage_size = read_pmd_pagesize(); > + const uint64_t hpage_shift = pmd_page_shift(); > + > + if (!hpage_size || !hpage_shift) > + return -1; > > /* drop pmd */ > - if (mmap(ptr, HPAGE_SIZE, PROT_READ | PROT_WRITE, > + if (mmap(ptr, hpage_size, PROT_READ | PROT_WRITE, > MAP_FIXED | MAP_ANONYMOUS | > MAP_NORESERVE | MAP_PRIVATE, -1, 0) != ptr) > ksft_exit_fail_msg("mmap transhuge\n"); > > - if (madvise(ptr, HPAGE_SIZE, MADV_HUGEPAGE)) > + if (madvise(ptr, hpage_size, MADV_HUGEPAGE)) > ksft_exit_fail_msg("MADV_HUGEPAGE\n"); > > /* allocate transparent huge page */ > @@ -471,7 +499,7 @@ int64_t allocate_transhuge(void *ptr, int pagemap_fd) > > if (PAGEMAP_PRESENT(ent[0]) && PAGEMAP_PRESENT(ent[1]) && > PAGEMAP_PFN(ent[0]) + 1 == PAGEMAP_PFN(ent[1]) && > - !(PAGEMAP_PFN(ent[0]) & ((1 << (HPAGE_SHIFT - pshift())) - 1))) > + !(PAGEMAP_PFN(ent[0]) & ((1 << (hpage_shift - pshift())) - 1))) > return PAGEMAP_PFN(ent[0]); > > return -1; > diff --git a/tools/testing/selftests/mm/vm_util.h > b/tools/testing/selftests/mm/vm_util.h > index 9a49af88702e..1d2608c1ad08 100644 > --- a/tools/testing/selftests/mm/vm_util.h > +++ b/tools/testing/selftests/mm/vm_util.h > @@ -31,6 +31,8 @@ > > extern unsigned int __page_size; > extern unsigned int __page_shift; > +extern uint64_t __pmd_page_size; > +extern uint64_t __pmd_page_shift; > > /* > * Represents an open fd and PROCMAP_QUERY state for binary (via ioctl) > @@ -89,6 +91,7 @@ unsigned long pagemap_get_pfn(int fd, char *start); > void clear_softdirty(void); > bool check_for_pattern(FILE *fp, const char *pattern, char *buf, size_t len); > uint64_t read_pmd_pagesize(void); > +uint64_t pmd_page_shift(void); > unsigned long rss_anon(void); > bool check_huge_anon(void *addr, size_t len, int nr_hpages, uint64_t > hpage_size); > bool check_huge_file(void *addr, size_t len, int nr_hpages, uint64_t > hpage_size); > @@ -157,12 +160,6 @@ int ksm_stop(void); > int get_hardware_corrupted_size(unsigned long *val); > int unpoison_memory(unsigned long pfn); > > -/* > - * On ppc64 this will only work with radix 2M hugepage size > - */ > -#define HPAGE_SHIFT 21 > -#define HPAGE_SIZE (1 << HPAGE_SHIFT) > - > #define PAGEMAP_PRESENT(ent) (((ent) & (1ull << 63)) != 0) > #define PAGEMAP_PFN(ent) ((ent) & ((1ull << 55) - 1)) > > -- > LEVI:{C3F47F37-75D8-414A-A8BA-3980EC8A46D7} > -- Cheers, Lorenzo
