ESwap

Basic Concepts

Enhanced Swap (ESwap) allows a custom partition to serve as a swap partition and uses a resident process zswapd to encrypt and swap the anonymous pages compressed by zram to the ESwap partition. In this way, a block of memory can be completely released to ensure the available memory (Memavailable) waterline. In addition to this reclaiming mechanism, the entire memory framework is enhanced to improve the reclaiming efficiency of anonymous pages and file pages and streamline the reclaiming ratio of these two types of pages to prevent refaults caused by excessive reclamation.

Configuring zram and ESwap

icon-note.gif NOTE
Enable ESwap before zram is enabled. If ESwap is not used, you can enable zram alone. If a device does not have the storage device for swap-out or have the corresponding storage partition created, you can enable zram to reclaim memory using zswapd.

Enabling ESwap

  1. Enable related configuration items and dependencies.

    To enable ESwap, you must enable the corresponding configuration items and dependencies during kernel compilation. The configuration items related to ESwap are as follows:

    CONFIG_HYPERHOLD=y                  // Enable Hyperhold
CONFIG_HYPERHOLD_DEBUG=y            // Enable Hyperhold debug
CONFIG_HYPERHOLD_ZSWAPD=y           // Enable the zswapd thread to reclaim Anon pages in background
CONFIG_HYPERHOLD_FILE_LRU=y         // Enable Hyperhold FILE LRU
CONFIG_HYPERHOLD_MEMCG=y            // Enable Memcg management in Hyperhold
CONFIG_ZRAM_GROUP=y                 // Enable Manage Zram objs with mem_cgroup
CONFIG_ZRAM_GROUP_DEBUG=y           // Enable Manage Zram objs with mem_cgroup Debug
CONFIG_ZLIST_DEBUG=y                // Enable Debug info for zram group list
CONFIG_ZRAM_GROUP_WRITEBACK=y       // Enable write back grouped zram objs to Hyperhold driver

    Enable the following dependencies:

    CONFIG_MEMCG=y       // Enable memory controller
CONFIG_SWAP=y        // Enable paging of anonymous memory (swap)
CONFIG_ZSMALLOC=y    // Enable memory allocator for compressed pages
CONFIG_ZRAM=y        // Enable compressed RAM block device support

  2. Create an ESwap device.

    You can use any block device as the ESwap device. In this example, create the hpdisk file and mount it to the loop6 device.

    // Run the dd command to create the hpdisk file for ESwap. In this example, the file size is 512 MB. Set the file size based on service requirements.
dd if=/dev/random of=/data/hpdisk bs=4096 count=131072
// Associate the hpdisk file with the ESwap device.
losetup /dev/block/loop6 hpdisk

  3. Configure ESwap.

    Bind the device created in step 2 as the ESwap device.

    echo /dev/block/loop6 > /proc/sys/kernel/hyperhold/device

    By default, ESwap encrypts the data swapped out. If the ESwap device created in step 2 supports inline encryption, you can disable the ESwap software encryption function.

    // Check whether hardware-based encryption is supported and enabled. If yes, disable software encryption. Otherwise, do not disable software encryption.
echo 0 > /proc/sys/kernel/hyperhold/soft_crypt

    icon-caution.gif CAUTION
    For security purposes, all swapped content must be encrypted. If the ESwap device created does not support inline encryption or the inline encryption macro is not enabled during compilation, ESwap cannot be enabled after software encryption is disabled.

  4. Enable ESwap.

    After ESwap is enabled, the preceding configuration cannot be modified.

    echo enable > /proc/sys/kernel/hyperhold/enable

Enabling zram

  1. Initialize zram.

    Configure the interaction between zram and ESwap and set the zram size.

    // Enable the swap-in and swap-out functions from zram to ESwap. Perform this step before setting the zram size.
echo readwrite > /sys/block/zram0/group
// Set the zram size. In this example, the zram size is set to 512 MB.
echo 512M > /sys/block/zram0/disksize

    icon-note.gif NOTE
    The parameters and functions of /sys/block/zram0/group are as follows:

    • disable: disables the function.
    • readonly: only records the cgroup information of the data but not swap it out.
    • readwrite: enables swap-in and swap-out from zram to ESwap.

  2. Enable zram.

    Use the zram device as the swap partition.

    mkswap /dev/block/zram0
swapon /dev/block/zram0

Disabling ESwap and zram

  1. Disable ESwap.

    echo disable > /proc/sys/kernel/hyperhold/enable
Or
echo force_disable > /proc/sys/kernel/hyperhold/enable

    icon-note.gif NOTE
    The difference of the two commands is as follows:

    • disable: If there is no data in the ESwap partition, disable ESwap. Otherwise, changes ESwap to readonly mode.
    • force_disable: If there is no data in the ESwap partition, disable ESwap. Otherwise, changes ESwap to readonly mode and disable ESwap until all data in the ESWAP partition is read.

  2. Disable zram and zram group.

    // If swapon is executed, run swapoff first.
swapoff /dev/block/zram0
echo 1 > /sys/block/zram0/reset

ESwap APIs

ESwap provides APIs to control swap-in and swap-out policies and record the current status. These APIs are located in the directory to which memcg is mounted, for example, /dev/memcg/.

Category API Description Reference Value
Control avail_buffers Sets the buffer range. 300 250 350 200
zswapd_single_memcg_param Sets memcg configuration. 300 40 0 0
zram_wm_ratio Sets the zram swap-out waterline. 0
Status zswapd_pressure_show Records the current buffer and refault. NA
stat Checks the real-time status of ESwap. NA
zswapd_vmstat_show Records events during the zswapd running. NA

icon-caution.gif CAUTION
Only avail_buffers proactively wakes up zswapd because the buffer waterline is adjusted. Other control APIs do not proactively wake up zswapd, but their configuration takes effect only after zswapd is woken up.

avail_buffers

The avail_buffers API sets the buffer range [min_avail_buffers, high_avail_buffers]. When the current buffer is less than the value of min_avail_buffers, zswapd will be woken up to reclaim anonymous pages. The expected amount of memory to reclaim is the difference between the value of high_avail_buffers and the current system buffer value. In fact, less memory is reclaimed due to reasons such as reclamation failure.

The parameters include the following:

  • avail_buffers indicates the expected buffer value.
  • free_swap_threshold indicates the threshold of the free capacity of the swap partition. After zswapd is woken up to reclaim memory, press events, such as medium press and critical press, will be recorded based on the current system status and the settings of these two parameters.

You can proactively adjust the values to trigger zswapd reclamation.

Example:

echo 1000 950 1050 0 > /dev/memcg/memory.avail_buffers

Default value:

avail_buffers: 0
min_avail_buffers: 0
high_avail_buffers: 0
free_swap_threshold: 0

Limit:

0<=min_avail_buffers<=avail_buffers<=high_avail_buffers

0<=free_swap_threashold

The values are all integers.

zswapd_single_memcg_param

zswapd_single_memcg_param sets the memcg configuration. The parameters include the following:

  • score indicates the current memcg reclamation priority.
  • ub_mem2zram_ratio indicates the memory compression ratio to zram.
  • ub_zram2ufs_ratio indicates the ratio of zram to ESwap.
  • refault_threshold indicates the refault threshold.

You can modify the parameters to control zram compression and ESwap.

Example:

echo 60 10 50 > memory.zswapd_single_memcg_param

Default value:

memcg score: 300
memcg ub_mem2zram_ratio: 60
memcg ub_zram2ufs_ratio: 10
memcg refault_threshold: 50

Limit:

0<=ub_mem2zram_ratio<=100

0<=ub_zram2ufs_ratio<=100

0<=refault_threshold<=100

The values are all integers.

zram_wm_ratio

zram_wm_ratio sets the zram swap-out waterline. When the size of the compressed anonymous page in the zram partition is greater than the total size of zram multiplied by zram_wm_ratio, the page is swapped out to the ESwap partition. The swap is performed after zswapd is woken up by the buffer waterline. The system defaults the value 0 as 37.

You can change the value as required.

Example:

echo 30 > /dev/memcg/memory.zram_wm_ratio

Default value:

zram_wm_ratio: 0

Limit:

0<=zram_wm_ratio<=100

The value is an integer.

zswapd_pressure_show

zswapd_pressure_show records the zswapd status. buffer_size indicates the current buffer size of the system, and recent_refault indicates the number of refaults occurred.

stat

In addition to memcg.stat, the stat API is added with Anon, File, zram, and Eswap to monitor ESwap in real time.

zswapd_vmstat_show

zswapd_vmstat_show records events occurred during the zswapd running.

Triggering zswapd

You can run cat /dev/memcg/memory.zswapd_pressure_show to check the current buffer value. For example, if the current buffer value is 1200, you can adjust the buffer range to a value greater than 1200 to wake up zswapd.

echo 1300 1250 1350 0 > /dev/memcg/memory.avail_buffers

ESwap Size

Set zram and ESwap partition sizes based on service requirements and hardware features. For the rk3568 board whose RAM is 2 GB, the zram and ESwap partitions are set to 512 MB.