Quelle  run_vmtests.sh   Sprache: Shell

#!/bin/bash
# SPDX-License-Identifier: GPL-2.0
# Please run as root

# Kselftest framework requirement - SKIP code is 4.
ksft_skip=4

count_total=0
count_pass=0
count_fail=0
count_skip=0
exitcode=0

usage() {
 cat <<EOF
usage: ${BASH_SOURCE[0]:-$0} [ options ]

  -a: run all tests, including extra ones (other than destructive ones)
  -t: specify specific categories to tests to run
  -h: display this message
  -n: disable TAP output
  -d: run destructive tests

The default behavior is to run required tests only.  If -a is specified,
will run all tests.

Alternatively, specific groups tests can be run by passing a string
to the -t argument containing one or more of the following categories
separated by spaces:
- mmap
 tests for mmap(2)
- gup_test
 tests for gup
- userfaultfd
 tests for  userfaultfd(2)
- compaction
 a test for the patch "Allow compaction of unevictable pages"
- mlock
 tests for mlock(2)
- mremap
 tests for mremap(2)
- hugevm
 tests for very large virtual address space
- vmalloc
 vmalloc smoke tests
- hmm
 hmm smoke tests
- madv_guard
 test madvise(2) MADV_GUARD_INSTALL and MADV_GUARD_REMOVE options
- madv_populate
 test memadvise(2) MADV_POPULATE_{READ,WRITE} options
- memfd_secret
 test memfd_secret(2)
- process_mrelease
 test process_mrelease(2)
- ksm
 ksm tests that do not require >=2 NUMA nodes
- ksm_numa
 ksm tests that require >=2 NUMA nodes
- pkey
 memory protection key tests
- soft_dirty
 test soft dirty page bit semantics
- pagemap
 test pagemap_scan IOCTL
- pfnmap
 tests for VM_PFNMAP handling
- process_madv
 test for process_madv
- cow
 test copy-on-write semantics
- thp
 test transparent huge pages
- hugetlb
 test hugetlbfs huge pages
- migration
 invoke move_pages(2) to exercise the migration entry code
 paths in the kernel
- mkdirty
 test handling of code that might set PTE/PMD dirty in
 read-only VMAs
- mdwe
 test prctl(PR_SET_MDWE, ...)
- page_frag
 test handling of page fragment allocation and freeing
- vma_merge
 test VMA merge cases behave as expected

example: ./run_vmtests.sh -t "hmm mmap ksm"
EOF
 exit 0
}

RUN_ALL=false
RUN_DESTRUCTIVE=false
TAP_PREFIX="# "

while getopts "aht:n" OPT; do
 case ${OPT} in
  "a") RUN_ALL=true ;;
  "h") usage ;;
  "t") VM_SELFTEST_ITEMS=${OPTARG} ;;
  "n") TAP_PREFIX= ;;
  "d") RUN_DESTRUCTIVE=true ;;
 esac
done
shift $((OPTIND -1))

# default behavior: run all tests
VM_SELFTEST_ITEMS=${VM_SELFTEST_ITEMS:-default}

test_selected() {
 if [ "$VM_SELFTEST_ITEMS" == "default" ]; then
  # If no VM_SELFTEST_ITEMS are specified, run all tests
  return 0
 fi
 # If test selected argument is one of the test items
 if [[ " ${VM_SELFTEST_ITEMS[*]} " =~ " ${1} " ]]; then
         return 0
 else
         return 1
 fi
}

run_gup_matrix() {
    # -t: thp=on, -T: thp=off, -H: hugetlb=on
    local hugetlb_mb=$(( needmem_KB / 1024 ))

    for huge in -t -T "-H -m $hugetlb_mb"; do
        # -u: gup-fast, -U: gup-basic, -a: pin-fast, -b: pin-basic, -L: pin-longterm
        for test_cmd in -u -U -a -b -L; do
            # -w: write=1, -W: write=0
            for write in -w -W; do
                # -S: shared
                for share in -S " "; do
                    # -n: How many pages to fetch together?  512 is special
                    # because it's default thp size (or 2M on x86), 123 to
                    # just test partial gup when hit a huge in whatever form
                    for num in "-n 1" "-n 512" "-n 123"; do
                        CATEGORY="gup_test" run_test ./gup_test \
                                $huge $test_cmd $write $share $num
                    done
                done
            done
        done
    done
}

# get huge pagesize and freepages from /proc/meminfo
while read -r name size unit; do
 if [ "$name" = "HugePages_Free:" ]; then
  freepgs="$size"
 fi
 if [ "$name" = "Hugepagesize:" ]; then
  hpgsize_KB="$size"
 fi
done < /proc/meminfo

# Simple hugetlbfs tests have a hardcoded minimum requirement of
# huge pages totaling 256MB (262144KB) in size.  The userfaultfd
# hugetlb test requires a minimum of 2 * nr_cpus huge pages.  Take
# both of these requirements into account and attempt to increase
# number of huge pages available.
nr_cpus=$(nproc)
uffd_min_KB=$((hpgsize_KB * nr_cpus * 2))
hugetlb_min_KB=$((256 * 1024))
if [[ $uffd_min_KB -gt $hugetlb_min_KB ]]; then
 needmem_KB=$uffd_min_KB
else
 needmem_KB=$hugetlb_min_KB
fi

# set proper nr_hugepages
if [ -n "$freepgs" ] && [ -n "$hpgsize_KB" ]; then
 nr_hugepgs=$(cat /proc/sys/vm/nr_hugepages)
 needpgs=$((needmem_KB / hpgsize_KB))
 tries=2
 while [ "$tries" -gt 0 ] && [ "$freepgs" -lt "$needpgs" ]; do
  lackpgs=$((needpgs - freepgs))
  echo 3 > /proc/sys/vm/drop_caches
  if ! echo $((lackpgs + nr_hugepgs)) > /proc/sys/vm/nr_hugepages; then
   echo "Please run this test as root"
   exit $ksft_skip
  fi
  while read -r name size unit; do
   if [ "$name" = "HugePages_Free:" ]; then
    freepgs=$size
   fi
  done < /proc/meminfo
  tries=$((tries - 1))
 done
 if [ "$freepgs" -lt "$needpgs" ]; then
  printf "Not enough huge pages available (%d < %d)\n" \
         "$freepgs" "$needpgs"
 fi
 HAVE_HUGEPAGES=1
else
 echo "no hugetlbfs support in kernel?"
 HAVE_HUGEPAGES=0
fi

# filter 64bit architectures
ARCH64STR="arm64 mips64 parisc64 ppc64 ppc64le riscv64 s390x sparc64 x86_64"
if [ -z "$ARCH" ]; then
 ARCH=$(uname -m 2>/dev/null | sed -e 's/aarch64.*/arm64/')
fi
VADDR64=0
echo "$ARCH64STR" | grep "$ARCH" &>/dev/null && VADDR64=1

tap_prefix() {
 sed -e "s/^/${TAP_PREFIX}/"
}

tap_output() {
 if [[ ! -z "$TAP_PREFIX" ]]; then
  read str
  echo $str
 fi
}

pretty_name() {
 echo "$*" | sed -e 's/^\(bash \)\?\.\///'
}

# Usage: run_test [test binary] [arbitrary test arguments...]
run_test() {
 if test_selected ${CATEGORY}; then
  local skip=0

  # On memory constrainted systems some tests can fail to allocate hugepages.
  # perform some cleanup before the test for a higher success rate.
  if [ ${CATEGORY} == "thp" -o ${CATEGORY} == "hugetlb" ]; then
   if [ "${HAVE_HUGEPAGES}" = "1" ]; then
    echo 3 > /proc/sys/vm/drop_caches
    sleep 2
    echo 1 > /proc/sys/vm/compact_memory
    sleep 2
   else
    echo "hugepages not supported" | tap_prefix
    skip=1
   fi
  fi

  local test=$(pretty_name "$*")
  local title="running $*"
  local sep=$(echo -n "$title" | tr "[:graph:][:space:]" -)
  printf "%s\n%s\n%s\n" "$sep" "$title" "$sep" | tap_prefix

  if [ "${skip}" != "1" ]; then
   ("$@" 2>&1) | tap_prefix
   local ret=${PIPESTATUS[0]}
  else
   local ret=$ksft_skip
  fi
  count_total=$(( count_total + 1 ))
  if [ $ret -eq 0 ]; then
   count_pass=$(( count_pass + 1 ))
   echo "[PASS]" | tap_prefix
   echo "ok ${count_total} ${test}" | tap_output
  elif [ $ret -eq $ksft_skip ]; then
   count_skip=$(( count_skip + 1 ))
   echo "[SKIP]" | tap_prefix
   echo "ok ${count_total} ${test} # SKIP" | tap_output
   exitcode=$ksft_skip
  else
   count_fail=$(( count_fail + 1 ))
   echo "[FAIL]" | tap_prefix
   echo "not ok ${count_total} ${test} # exit=$ret" | tap_output
   exitcode=1
  fi
 fi # test_selected
}

echo "TAP version 13" | tap_output

CATEGORY="hugetlb" run_test ./hugepage-mmap

shmmax=$(cat /proc/sys/kernel/shmmax)
shmall=$(cat /proc/sys/kernel/shmall)
echo 268435456 > /proc/sys/kernel/shmmax
echo 4194304 > /proc/sys/kernel/shmall
CATEGORY="hugetlb" run_test ./hugepage-shm
echo "$shmmax" > /proc/sys/kernel/shmmax
echo "$shmall" > /proc/sys/kernel/shmall

CATEGORY="hugetlb" run_test ./map_hugetlb
CATEGORY="hugetlb" run_test ./hugepage-mremap
CATEGORY="hugetlb" run_test ./hugepage-vmemmap
CATEGORY="hugetlb" run_test ./hugetlb-madvise
CATEGORY="hugetlb" run_test ./hugetlb_dio

if [ "${HAVE_HUGEPAGES}" = "1" ]; then
 nr_hugepages_tmp=$(cat /proc/sys/vm/nr_hugepages)
 # For this test, we need one and just one huge page
 echo 1 > /proc/sys/vm/nr_hugepages
 CATEGORY="hugetlb" run_test ./hugetlb_fault_after_madv
 CATEGORY="hugetlb" run_test ./hugetlb_madv_vs_map
 # Restore the previous number of huge pages, since further tests rely on it
 echo "$nr_hugepages_tmp" > /proc/sys/vm/nr_hugepages
fi

if test_selected "hugetlb"; then
 echo "NOTE: These hugetlb tests provide minimal coverage. Use"   | tap_prefix
 echo " https://github.com/libhugetlbfs/libhugetlbfs.git for" | tap_prefix
 echo " hugetlb regression testing."     | tap_prefix
fi

CATEGORY="mmap" run_test ./map_fixed_noreplace

if $RUN_ALL; then
    run_gup_matrix
else
    # get_user_pages_fast() benchmark
    CATEGORY="gup_test" run_test ./gup_test -u
    # pin_user_pages_fast() benchmark
    CATEGORY="gup_test" run_test ./gup_test -a
fi
# Dump pages 0, 19, and 4096, using pin_user_pages:
CATEGORY="gup_test" run_test ./gup_test -ct -F 0x1 0 19 0x1000
CATEGORY="gup_test" run_test ./gup_longterm

CATEGORY="userfaultfd" run_test ./uffd-unit-tests
uffd_stress_bin=./uffd-stress
CATEGORY="userfaultfd" run_test ${uffd_stress_bin} anon 20 16
# Hugetlb tests require source and destination huge pages. Pass in half
# the size of the free pages we have, which is used for *each*.
# uffd-stress expects a region expressed in MiB, so we adjust
# half_ufd_size_MB accordingly.
half_ufd_size_MB=$(((freepgs * hpgsize_KB) / 1024 / 2))
CATEGORY="userfaultfd" run_test ${uffd_stress_bin} hugetlb "$half_ufd_size_MB" 32
CATEGORY="userfaultfd" run_test ${uffd_stress_bin} hugetlb-private "$half_ufd_size_MB" 32
CATEGORY="userfaultfd" run_test ${uffd_stress_bin} shmem 20 16
CATEGORY="userfaultfd" run_test ${uffd_stress_bin} shmem-private 20 16
# uffd-wp-mremap requires at least one page of each size.
have_all_size_hugepgs=true
declare -A nr_size_hugepgs
for f in /sys/kernel/mm/hugepages/**/nr_hugepages; do
 old=$(cat $f)
 nr_size_hugepgs["$f"]="$old"
 if [ "$old" == 0 ]; then
  echo 1 > "$f"
 fi
 if [ $(cat "$f") == 0 ]; then
  have_all_size_hugepgs=false
  break
 fi
done
if $have_all_size_hugepgs; then
 CATEGORY="userfaultfd" run_test ./uffd-wp-mremap
else
 echo "# SKIP ./uffd-wp-mremap"
fi

#cleanup
for f in "${!nr_size_hugepgs[@]}"; do
 echo "${nr_size_hugepgs["$f"]}" > "$f"
done
echo "$nr_hugepgs" > /proc/sys/vm/nr_hugepages

CATEGORY="compaction" run_test ./compaction_test

if command -v sudo &> /dev/null && sudo -u nobody ls ./on-fault-limit >/dev/null;
then
 CATEGORY="mlock" run_test sudo -u nobody ./on-fault-limit
else
 echo "# SKIP ./on-fault-limit"
fi

CATEGORY="mmap" run_test ./map_populate

CATEGORY="mlock" run_test ./mlock-random-test

CATEGORY="mlock" run_test ./mlock2-tests

CATEGORY="process_mrelease" run_test ./mrelease_test

CATEGORY="mremap" run_test ./mremap_test

CATEGORY="hugetlb" run_test ./thuge-gen
CATEGORY="hugetlb" run_test ./charge_reserved_hugetlb.sh -cgroup-v2
CATEGORY="hugetlb" run_test ./hugetlb_reparenting_test.sh -cgroup-v2
if $RUN_DESTRUCTIVE; then
nr_hugepages_tmp=$(cat /proc/sys/vm/nr_hugepages)
enable_soft_offline=$(cat /proc/sys/vm/enable_soft_offline)
echo 8 > /proc/sys/vm/nr_hugepages
CATEGORY="hugetlb" run_test ./hugetlb-soft-offline
echo "$nr_hugepages_tmp" > /proc/sys/vm/nr_hugepages
echo "$enable_soft_offline" > /proc/sys/vm/enable_soft_offline
CATEGORY="hugetlb" run_test ./hugetlb-read-hwpoison
fi

if [ $VADDR64 -ne 0 ]; then

 # set overcommit_policy as OVERCOMMIT_ALWAYS so that kernel
 # allows high virtual address allocation requests independent
 # of platform's physical memory.

 if [ -x ./virtual_address_range ]; then
  prev_policy=$(cat /proc/sys/vm/overcommit_memory)
  echo 1 > /proc/sys/vm/overcommit_memory
  CATEGORY="hugevm" run_test ./virtual_address_range
  echo $prev_policy > /proc/sys/vm/overcommit_memory
 fi

 # va high address boundary switch test
 ARCH_ARM64="arm64"
 prev_nr_hugepages=$(cat /proc/sys/vm/nr_hugepages)
 if [ "$ARCH" == "$ARCH_ARM64" ]; then
  echo 6 > /proc/sys/vm/nr_hugepages
 fi
 CATEGORY="hugevm" run_test bash ./va_high_addr_switch.sh
 if [ "$ARCH" == "$ARCH_ARM64" ]; then
  echo $prev_nr_hugepages > /proc/sys/vm/nr_hugepages
 fi
fi # VADDR64

# vmalloc stability smoke test
CATEGORY="vmalloc" run_test bash ./test_vmalloc.sh smoke

CATEGORY="mremap" run_test ./mremap_dontunmap

CATEGORY="hmm" run_test bash ./test_hmm.sh smoke

# MADV_GUARD_INSTALL and MADV_GUARD_REMOVE tests
CATEGORY="madv_guard" run_test ./guard-regions

# MADV_POPULATE_READ and MADV_POPULATE_WRITE tests
CATEGORY="madv_populate" run_test ./madv_populate

# PROCESS_MADV test
CATEGORY="process_madv" run_test ./process_madv

CATEGORY="vma_merge" run_test ./merge

if [ -x ./memfd_secret ]
then
if [ -f /proc/sys/kernel/yama/ptrace_scope ]; then
 (echo 0 > /proc/sys/kernel/yama/ptrace_scope 2>&1) | tap_prefix
fi
CATEGORY="memfd_secret" run_test ./memfd_secret
fi

# KSM KSM_MERGE_TIME_HUGE_PAGES test with size of 100
if [ "${HAVE_HUGEPAGES}" = "1" ]; then
 CATEGORY="ksm" run_test ./ksm_tests -H -s 100
fi
# 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
CATEGORY="ksm" run_test ./ksm_tests -M -p 10
# KSM unmerge test
CATEGORY="ksm" run_test ./ksm_tests -U
# KSM test with 10 zero pages and use_zero_pages = 0
CATEGORY="ksm" run_test ./ksm_tests -Z -p 10 -z 0
# KSM test with 10 zero pages and use_zero_pages = 1
CATEGORY="ksm" run_test ./ksm_tests -Z -p 10 -z 1
# KSM test with 2 NUMA nodes and merge_across_nodes = 1
CATEGORY="ksm_numa" run_test ./ksm_tests -N -m 1
# KSM test with 2 NUMA nodes and merge_across_nodes = 0
CATEGORY="ksm_numa" run_test ./ksm_tests -N -m 0

CATEGORY="ksm" run_test ./ksm_functional_tests

# protection_keys tests
nr_hugepgs=$(cat /proc/sys/vm/nr_hugepages)
if [ -x ./protection_keys_32 ]
then
 CATEGORY="pkey" run_test ./protection_keys_32
fi

if [ -x ./protection_keys_64 ]
then
 CATEGORY="pkey" run_test ./protection_keys_64
fi
echo "$nr_hugepgs" > /proc/sys/vm/nr_hugepages

if [ -x ./soft-dirty ]
then
 CATEGORY="soft_dirty" run_test ./soft-dirty
fi

CATEGORY="pagemap" run_test ./pagemap_ioctl

CATEGORY="pfnmap" run_test ./pfnmap

# COW tests
CATEGORY="cow" run_test ./cow

CATEGORY="thp" run_test ./khugepaged

CATEGORY="thp" run_test ./khugepaged -s 2

CATEGORY="thp" run_test ./khugepaged all:shmem

CATEGORY="thp" run_test ./khugepaged -s 4 all:shmem

CATEGORY="thp" run_test ./transhuge-stress -d 20

# Try to create XFS if not provided
if [ -z "${SPLIT_HUGE_PAGE_TEST_XFS_PATH}" ]; then
    if [ "${HAVE_HUGEPAGES}" = "1" ]; then
 if test_selected "thp"; then
     if grep xfs /proc/filesystems &>/dev/null; then
  XFS_IMG=$(mktemp /tmp/xfs_img_XXXXXX)
  SPLIT_HUGE_PAGE_TEST_XFS_PATH=$(mktemp -d /tmp/xfs_dir_XXXXXX)
  truncate -s 314572800 ${XFS_IMG}
  mkfs.xfs -q ${XFS_IMG}
  mount -o loop ${XFS_IMG} ${SPLIT_HUGE_PAGE_TEST_XFS_PATH}
  MOUNTED_XFS=1
     fi
 fi
    fi
fi

CATEGORY="thp" run_test ./split_huge_page_test ${SPLIT_HUGE_PAGE_TEST_XFS_PATH}

if [ -n "${MOUNTED_XFS}" ]; then
    umount ${SPLIT_HUGE_PAGE_TEST_XFS_PATH}
    rmdir ${SPLIT_HUGE_PAGE_TEST_XFS_PATH}
    rm -f ${XFS_IMG}
fi

CATEGORY="migration" run_test ./migration

CATEGORY="mkdirty" run_test ./mkdirty

CATEGORY="mdwe" run_test ./mdwe_test

CATEGORY="page_frag" run_test ./test_page_frag.sh smoke

CATEGORY="page_frag" run_test ./test_page_frag.sh aligned

CATEGORY="page_frag" run_test ./test_page_frag.sh nonaligned

echo "SUMMARY: PASS=${count_pass} SKIP=${count_skip} FAIL=${count_fail}" | tap_prefix
echo "1..${count_total}" | tap_output

exit $exitcode