]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
selftests/mm: fix hugepages cleanup too early
authorChunyu Hu <chuhu@redhat.com>
Mon, 8 Sep 2025 12:47:37 +0000 (20:47 +0800)
committerAndrew Morton <akpm@linux-foundation.org>
Fri, 12 Sep 2025 00:25:59 +0000 (17:25 -0700)
Patch series "Fix va_high_addr_switch.sh test failure", v2.

These three patches fix a va_high_addr_switch.sh test failure on x86_64.

This patch (of 3):

The nr_hugepgs variable is used to keep the original nr_hugepages at the
hugepage setup step at test beginning.  After userfaultfd test, a cleaup
is executed, both /sys/kernel/mm/hugepages/hugepages-*/nr_hugepages and
/proc/sys//vm/nr_hugepages are reset to 'original' value before
userfaultfd test starts.

Issue here is the value used to restore /proc/sys/vm/nr_hugepages is
nr_hugepgs which is the initial value before the vm_runtests.sh runs, not
the value before userfaultfd test starts.  'va_high_addr_swith.sh' tests
runs after that will possibly see no hugepages available for test, and got
EINVAL when mmap(HUGETLB), making the result invalid.

And before pkey tests, nr_hugepgs is changed to be used as a temp variable
to save nr_hugepages before pkey test, and restore it after pkey tests
finish.  The original nr_hugepages value is not tracked anymore, so no way
to restore it after all tests finish.

Add a new variable orig_nr_hugepgs to save the original nr_hugepages, and
and restore it to nr_hugepages after all tests finish.  And change to use
the nr_hugepgs variable to save the /proc/sys/vm/nr_hugeages after
hugepage setup, it's also the value before userfaultfd test starts, and
the correct value to be restored after userfaultfd finishes.  The
va_high_addr_switch.sh broken will be resolved.

Link: https://lkml.kernel.org/r/20250908124740.2946005-1-chuhu@redhat.com
Link: https://lkml.kernel.org/r/20250908124740.2946005-2-chuhu@redhat.com
Signed-off-by: Chunyu Hu <chuhu@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
tools/testing/selftests/mm/run_vmtests.sh

index 9e88cc25b9df2d550fba2acc970d603e24799132..8115fc4526ed9d2a80eba1dcb5341d2ed549e350 100755 (executable)
@@ -174,13 +174,13 @@ fi
 
 # set proper nr_hugepages
 if [ -n "$freepgs" ] && [ -n "$hpgsize_KB" ]; then
-       nr_hugepgs=$(cat /proc/sys/vm/nr_hugepages)
+       orig_nr_hugepgs=$(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
+               if ! echo $((lackpgs + orig_nr_hugepgs)) > /proc/sys/vm/nr_hugepages; then
                        echo "Please run this test as root"
                        exit $ksft_skip
                fi
@@ -191,6 +191,7 @@ if [ -n "$freepgs" ] && [ -n "$hpgsize_KB" ]; then
                done < /proc/meminfo
                tries=$((tries - 1))
        done
+       nr_hugepgs=$(cat /proc/sys/vm/nr_hugepages)
        if [ "$freepgs" -lt "$needpgs" ]; then
                printf "Not enough huge pages available (%d < %d)\n" \
                       "$freepgs" "$needpgs"
@@ -540,6 +541,10 @@ CATEGORY="page_frag" run_test ./test_page_frag.sh nonaligned
 
 CATEGORY="rmap" run_test ./rmap
 
+if [ "${HAVE_HUGEPAGES}" = 1 ]; then
+       echo "$orig_nr_hugepgs" > /proc/sys/vm/nr_hugepages
+fi
+
 echo "SUMMARY: PASS=${count_pass} SKIP=${count_skip} FAIL=${count_fail}" | tap_prefix
 echo "1..${count_total}" | tap_output