zram: skip exfat on systems with large page size
PATCH 1/1 zram: skip exfat on systems with large page size
exfat requires cluster size >= sector size, but defaults to 4KB. On systems with large page sizes (e.g., 64KB on POWER), zram uses page size as sector size, causing mkfs.exfat to fail.
Skip exfat filesystem creation when page size > 4096 to avoid test failure.
--- .../kernel/device-drivers/zram/zram01.sh | 91 +++++++++++++++++++ 1 file changed, 91 insertions(+)
testcases/kernel/device-drivers/zram/zram01.sh +91 -0
@@ -11,6 +11,10 @@ TST_TESTFUNC="do_test" TST_NEEDS_CMDS="awk bc dd" TST_SETUP="setup" +# Track devices that were skipped due to sector size incompatibility +# Format: space-separated list of device numbers (e.g., "2 5" if zram2 and zram5 were skipped) +skipped_devices="" + check_space_for_fs() { local fs="$1"
@@ -72,15 +76,59 @@ setup() zram_load } +check_fs_sector_support() +{ + local fs="$1" + local dev="$2" + local page_size + local sector_size + + # Get the system page size - zram uses this as sector size + page_size=$(getconf PAGE_SIZE 2>/dev/null) + + if [ -z "$page_size" ] || [ "$page_size" -lt 4096 ]; then + # Default to 4KB if we can't determine page size + page_size=4096 + fi + + tst_res TINFO "check_fs_sector_support: $fs on zram$dev system_page_size=$page_size" + + # CHECK: exfat requires cluster size >= sector size, but defaults to 4KB + # On systems with large page sizes (e.g., 64KB on POWER), sector size > 4KB + # This causes exfat filesystem creation to fail, so we skip it with TCONF + if [ "$fs" = "exfat" ] && [ "$page_size" -gt 4096 ]; then + tst_res TINFO "check_fs_sector_support: $fs on zram$dev has page_size=$page_size (> 4096), skipping" + return 1 # Return failure (skip this filesystem) + fi + + return 0 # All other filesystems pass this check +} + zram_makefs() { local i=$dev_start local fs + local sector_size + # Loop through each filesystem assigned to each zram device for fs in $zram_filesystems; do + tst_res TINFO "zram_makefs: checking device zram$i for filesystem $fs" + + # Check if this filesystem is compatible with the device's sector size + if ! check_fs_sector_support "$fs" "$i"; then + # Filesystem not supported on this device - mark device as skipped + sector_size=$(cat /sys/block/zram${i}/queue/logical_block_size) + tst_res TCONF "mkfs.$fs does not support sector size $sector_size" + skipped_devices="$skipped_devices $i" # Add to skipped list + i=$(($i + 1)) + continue # Skip to next filesystem + fi + + # Filesystem is supported - attempt to create it tst_res TINFO "make $fs filesystem on /dev/zram$i" mkfs.$fs /dev/zram$i > err.log 2>&1 if [ $? -ne 0 ]; then + # Filesystem creation failed cat err.log tst_res TFAIL "Failed to make $fs on /dev/zram$i" tst_brk TBROK "Can't continue with mounting the FS"
@@ -95,8 +143,27 @@ zram_makefs() zram_mount() { local i + local skip + # Attempt to mount all zram devices for i in $(seq $dev_start $dev_end); do + skip=0 + + # Check if this device was marked as skipped during filesystem creation + for dev in $skipped_devices; do + if [ "$dev" = "$i" ]; then + skip=1 + break + fi + done + + # Skip mounting devices that don't have filesystems + if [ $skip -eq 1 ]; then + tst_res TINFO "skipping mount of /dev/zram$i (filesystem not created)" + continue + fi + + # Mount the filesystem that was created in zram_makefs tst_res TINFO "mount /dev/zram$i" mkdir zram$i ROD mount /dev/zram$i zram$i
@@ -130,8 +197,27 @@ zram_fill_fs() { local mem_used_total local b i r v + local skip + # Fill each mounted filesystem with data to test compression for i in $(seq $dev_start $dev_end); do + skip=0 + + # Check if this device was marked as skipped (has no filesystem) + for dev in $skipped_devices; do + if [ "$dev" = "$i" ]; then + skip=1 + break + fi + done + + # Skip filling devices that don't have filesystems + if [ $skip -eq 1 ]; then + tst_res TINFO "skipping fill of zram$i (filesystem not created)" + continue + fi + + # Fill this zram device with zeros to test compression ratio tst_res TINFO "filling zram$i (it can take long time)" b=0 while true; do
@@ -140,12 +226,15 @@ zram_fill_fs() >/dev/null 2>err.txt || break b=$(($b + 1)) done + + # Verify that at least some data was written if [ $b -eq 0 ]; then [ -s err.txt ] && tst_res TWARN "dd error: $(cat err.txt)" tst_brk TBROK "cannot fill zram $i" fi tst_res TPASS "zram$i was filled with '$b' KB" + # Check if compression statistics are available (requires mm_stat) if [ ! -f "/sys/block/zram$i/mm_stat" ]; then if [ $i -eq 0 ]; then tst_res TCONF "zram compression ratio test requires zram mm_stat sysfs file"
@@ -154,6 +243,7 @@ zram_fill_fs() continue fi + # Read and verify the compression ratio TST_RETRY_FN_EXP_BACKOFF "check_read_mem_used_total /sys/block/zram$i/mm_stat" 0 10 mem_used_total=$(read_mem_used_total /sys/block/zram$i/mm_stat) tst_res TINFO "mem_used_total: $mem_used_total"
@@ -161,6 +251,7 @@ zram_fill_fs() v=$((100 * 1024 * $b / $mem_used_total)) r=$(echo "scale=2; $v / 100 " | bc) + # Fail if compression ratio is less than 1:1 (data grew instead of compressed) if [ "$v" -lt 100 ]; then tst_res TFAIL "compression ratio: $r:1" break
Signature · 5 lines
2.39.1 -- Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 5 lines
exfat requires cluster size >= sector size, but defaults to 4KB. On systems with large page sizes (e.g., 64KB on POWER), zram uses page size as sector size, causing mkfs.exfat to fail.
Skip exfat filesystem creation when page size > 4096 to avoid test failure.
Or, can we specify the sector size with 'mkfs.$fs -c $page_size' when detecting a large pagesize system?
Signature · 6 lines
Regards, Li Wang -- Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 2 lines · Li Wang
Or, can we specify the sector size with 'mkfs.$fs -c $page_size' when detecting a large pagesize system?
Thanks for the review.
Using mkfs.exfat -c $page_size would technically work, but I avoided it because:
1. zram01 tests zram behaviour, not filesystem configuration -- using a non-default cluster size feels out of scope.
2. The cluster size must be valid for the disk size (only 25MB here), so large cluster sizes could hit additional constraints.
Skipping with TCONF honestly reflects that default exfat tooling does not support this hardware configuration, which is useful signal.
Happy to send a v2 with the -c approach if you prefer.
Thanks,
Soma
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
Thanks for the explanation.
I still think we should go with the -c approach rather than skipping. A couple of points:
On 1, I don't think this changes the scope of the test. zram01 already invokes mkfs.$fs as a means to exercise zram, adjusting a parameter so that mkfs actually succeeds is just making the test work, not turning it into a filesystem configuration test.
On 2, 25MB with a 64KB cluster size gives roughly 390 clusters, which is well within exfat's limits. Could you point out a concrete constraint that would actually fail here? If there isn't one, I'd prefer not to use a hypothetical concern as justification for skipping.
On the TCONF, reporting TCONF here would make it look like exfat on zram is fundamentally broken on 64K-page systems, when in reality it works with a one-line change. That is a false negative we should avoid in a test suite.
Please send a v2 with the -c approach.
Quoted 22 lines · Soma Das
Quoted 2 lines · Li Wang
Or, can we specify the sector size with 'mkfs.$fs -c $page_size' when detecting a large pagesize system?
Thanks for the review.
Using mkfs.exfat -c $page_size would technically work, but I avoided it because:
1. zram01 tests zram behaviour, not filesystem configuration -- using a non-default cluster size feels out of scope.
2. The cluster size must be valid for the disk size (only 25MB here), so large cluster sizes could hit additional constraints.Skipping with TCONF honestly reflects that default exfat tooling does not support this hardware configuration, which is useful signal.
Happy to send a v2 with the -c approach if you prefer.
Thanks,
Soma
Signature · 6 lines
Regards, Li Wang -- Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 1 line · Li Wang
I still think we should go with the -c approach rather than skipping.
Thanks for the detailed explanation.
Understood — I will send a v2 with the -c $page_size approach.
Thanks,
Soma
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp