In diagnosing some issues with the partner test in LTP I noticed that because this fills the device with zeros, it doesn't really activate the compressed allocator path as it is designed to do. This is a "lite" version of the LTP patch that simply perturbs the zero writes to avoid having all pages hit the same-page detection, and adds a sync so that we're looking at a more quiescent system to get the final stats. While we're here, this rewords and expands some of the mm_stat flag docs to be a bit more explicit about what's going on. Link: https://lore.kernel.org/ltp/20230808035641.364676-2-iwienand@xxxxxxxxxx/T/#u Signed-off-by: Ian Wienand <iwienand@xxxxxxxxxx> --- Documentation/admin-guide/blockdev/zram.rst | 22 ++++++++++++++------- tools/testing/selftests/zram/zram01.sh | 18 +++++++++++++++-- 2 files changed, 31 insertions(+), 9 deletions(-) diff --git a/Documentation/admin-guide/blockdev/zram.rst b/Documentation/admin-guide/blockdev/zram.rst index e4551579cb12..a39a01870f40 100644 --- a/Documentation/admin-guide/blockdev/zram.rst +++ b/Documentation/admin-guide/blockdev/zram.rst @@ -253,17 +253,25 @@ line of text and contains the following stats separated by whitespace: orig_data_size uncompressed size of data stored in this disk. Unit: bytes compr_data_size compressed size of data stored in this disk - mem_used_total the amount of memory allocated for this disk. This - includes allocator fragmentation and metadata overhead, - allocated for this disk. So, allocator space efficiency - can be calculated using compr_data_size and this statistic. - Unit: bytes + mem_used_total the amount of memory currently used by the compressed + memory allocator to hold compressed data. This + includes allocator fragmentation and metadata + overhead. The device space efficiency can be + calculated as a ratio of the compr_data_size / + mem_used_total. Note this value may be zero; + particularly if all pages are filled with identical + data (see same_pages). mem_limit the maximum amount of memory ZRAM can use to store the compressed data mem_used_max the maximum amount of memory zram has consumed to store the data - same_pages the number of same element filled pages written to this disk. - No memory is allocated for such pages. + same_pages pages identified as being filled exclusivley with an + identicial unsigned-long value are recorded + specially by zram and thus are not stored via the + compression allocator. This avoids fragmentation + and metadata overheads for common cases such as + zeroed or poision data. same_pages is the current + number of pages kept in this de-duplicated form. pages_compacted the number of pages freed during compaction huge_pages the number of incompressible pages huge_pages_since the number of incompressible pages since zram set up diff --git a/tools/testing/selftests/zram/zram01.sh b/tools/testing/selftests/zram/zram01.sh index 8f4affe34f3e..122625d744c2 100755 --- a/tools/testing/selftests/zram/zram01.sh +++ b/tools/testing/selftests/zram/zram01.sh @@ -33,16 +33,30 @@ zram_algs="lzo" zram_fill_fs() { + local page_size=$(getconf PAGE_SIZE) for i in $(seq $dev_start $dev_end); do echo "fill zram$i..." local b=0 while [ true ]; do - dd conv=notrunc if=/dev/zero of=zram${i}/file \ + # If we fill with all zeros, every page hits + # the same-page detection and never makes it + # to compressed backing. Filling the first 1K + # of the page with (likely lowly compressible) + # random data ensures we hit the compression + # paths, but the highly compressible rest of + # the page also ensures we get a sufficiently + # high ratio to assert on below. + local input_file='/dev/zero' + if [ $(( (b * 1024) % page_size )) -eq 0 ]; then + input_file='/dev/urandom' + fi + dd conv=notrunc if=${input_file} of=zram${i}/file \ oflag=append count=1 bs=1024 status=none \ > /dev/null 2>&1 || break b=$(($b + 1)) done - echo "zram$i can be filled with '$b' KB" + echo "zram$i was filled with '$b' KB" + sync local mem_used_total=`awk '{print $3}' "/sys/block/zram$i/mm_stat"` local v=$((100 * 1024 * $b / $mem_used_total)) -- 2.41.0
