Hi Breno, On 05/10/2023 17:39, Breno Leitao wrote: > Create a selftest that exercises the race between page faults and > madvise(MADV_DONTNEED) in the same huge page. Do it by running two > threads that touches the huge page and madvise(MADV_DONTNEED) at the same > time. > > In case of a SIGBUS coming at pagefault, the test should fail, since we > hit the bug. > > The test doesn't have a signal handler, and if it fails, it fails like > the following > > ---------------------------------- > running ./hugetlb_fault_after_madv > ---------------------------------- > ./run_vmtests.sh: line 186: 595563 Bus error (core dumped) "$@" > [FAIL] > > This selftest goes together with the fix of the bug[1] itself. > > [1] https://lore.kernel.org/all/20231001005659.2185316-1-riel@xxxxxxxxxxx/#r > > Signed-off-by: Breno Leitao <leitao@xxxxxxxxxx> > --- > tools/testing/selftests/mm/Makefile | 1 + > .../selftests/mm/hugetlb_fault_after_madv.c | 73 +++++++++++++++++++ > tools/testing/selftests/mm/run_vmtests.sh | 4 + > 3 files changed, 78 insertions(+) > create mode 100644 tools/testing/selftests/mm/hugetlb_fault_after_madv.c > > diff --git a/tools/testing/selftests/mm/Makefile b/tools/testing/selftests/mm/Makefile > index 6a9fc5693145..e71ec9910c62 100644 > --- a/tools/testing/selftests/mm/Makefile > +++ b/tools/testing/selftests/mm/Makefile > @@ -68,6 +68,7 @@ TEST_GEN_FILES += split_huge_page_test > TEST_GEN_FILES += ksm_tests > TEST_GEN_FILES += ksm_functional_tests > TEST_GEN_FILES += mdwe_test > +TEST_GEN_FILES += hugetlb_fault_after_madv > > ifneq ($(ARCH),arm64) > TEST_GEN_PROGS += soft-dirty > diff --git a/tools/testing/selftests/mm/hugetlb_fault_after_madv.c b/tools/testing/selftests/mm/hugetlb_fault_after_madv.c > new file mode 100644 > index 000000000000..73b81c632366 > --- /dev/null > +++ b/tools/testing/selftests/mm/hugetlb_fault_after_madv.c > @@ -0,0 +1,73 @@ > +// SPDX-License-Identifier: GPL-2.0 > +#include <pthread.h> > +#include <stdio.h> > +#include <stdlib.h> > +#include <sys/mman.h> > +#include <sys/types.h> > +#include <unistd.h> > + > +#include "vm_util.h" > +#include "../kselftest.h" > + > +#define MMAP_SIZE (1 << 21) > +#define INLOOP_ITER 100 > + > +char *huge_ptr; > + > +/* Touch the memory while it is being madvised() */ > +void *touch(void *unused) > +{ > + char *ptr = (char *)huge_ptr; > + > + for (int i = 0; i < INLOOP_ITER; i++) > + ptr[0] = '.'; > + > + return NULL; > +} > + > +void *madv(void *unused) > +{ > + usleep(rand() % 10); > + > + for (int i = 0; i < INLOOP_ITER; i++) > + madvise(huge_ptr, MMAP_SIZE, MADV_DONTNEED); > + > + return NULL; > +} > + > +int main(void) > +{ > + unsigned long free_hugepages; > + pthread_t thread1, thread2; > + /* > + * On kernel 6.4, we are able to reproduce the problem with ~1000 > + * interactions > + */ > + int max = 10000; > + > + srand(getpid()); > + > + free_hugepages = get_free_hugepages(); > + if (free_hugepages != 1) { > + ksft_exit_skip("This test needs one and only one page to execute. Got %lu\n", > + free_hugepages); > + } > + > + while (max--) { > + huge_ptr = mmap(NULL, MMAP_SIZE, PROT_READ | PROT_WRITE, > + MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, > + -1, 0); > + > + if ((unsigned long)huge_ptr == -1) > + ksft_exit_skip("Failed to allocated huge page\n"); > + > + pthread_create(&thread1, NULL, madv, NULL); > + pthread_create(&thread2, NULL, touch, NULL); > + > + pthread_join(thread1, NULL); > + pthread_join(thread2, NULL); > + munmap(huge_ptr, MMAP_SIZE); > + } > + > + return KSFT_PASS; > +} > diff --git a/tools/testing/selftests/mm/run_vmtests.sh b/tools/testing/selftests/mm/run_vmtests.sh > index 3e2bc818d566..9f53f7318a38 100755 > --- a/tools/testing/selftests/mm/run_vmtests.sh > +++ b/tools/testing/selftests/mm/run_vmtests.sh > @@ -221,6 +221,10 @@ CATEGORY="hugetlb" run_test ./hugepage-mremap > CATEGORY="hugetlb" run_test ./hugepage-vmemmap > CATEGORY="hugetlb" run_test ./hugetlb-madvise > > +# For this test, we need one and just one huge page > +echo 1 > /proc/sys/vm/nr_hugepages I've noticed that this change breaks some of the uffd-stress tests further down the file, because you have freed previously reserved hugepages that the test requires to run. I notice that the patch is already in mm-stable, so perhaps its possible to submit a patch that does a save and restore? Although I'm not sure if that might be tricky because the previous reservation is per-size and per-node (our CI does this on the kernel command line), and I suspect if you want just 1 huge page in the entire system you won't be able to get back to the previous state by just restoring this value? These are the failing tests for reference: # ------------------------------------ # running ./uffd-stress hugetlb 128 32 # ------------------------------------ # nr_pages: 64, nr_pages_per_cpu: 8 # ERROR: context init failed (errno=12, @uffd-stress.c:254) # [FAIL] # -------------------------------------------- # running ./uffd-stress hugetlb-private 128 32 # -------------------------------------------- # nr_pages: 64, nr_pages_per_cpu: 8 # ERROR: context init failed (errno=12, @uffd-stress.c:254) # [FAIL] Thanks, Ryan > +CATEGORY="hugetlb" run_test ./hugetlb_fault_after_madv > + > if test_selected "hugetlb"; then > echo "NOTE: These hugetlb tests provide minimal coverage. Use" > echo " https://github.com/libhugetlbfs/libhugetlbfs.git for"