diff --git a/Documentation/index.rst b/Documentation/index.rst
index d58f488d90693bc09c145bbb76b0c9b612055056..1f1d464c397c82abc6c3bab8b309e39a08d3d0a8 100644
--- a/Documentation/index.rst
+++ b/Documentation/index.rst
@@ -90,6 +90,7 @@ TKernel documentation
    tkernel/tkernel
    tkernel/irq-latency
    tkernel/cgroup-v2-mbuf
+   tkernel/emm
 
 Firmware-related documentation
 ==============================
diff --git a/Documentation/tkernel/emm.rst b/Documentation/tkernel/emm.rst
new file mode 100644
index 0000000000000000000000000000000000000000..867b65d63b9bd94499b5efb0c61b0e16510d49f9
--- /dev/null
+++ b/Documentation/tkernel/emm.rst
@@ -0,0 +1,87 @@
+.. SPDX-License-Identifier: GPL-2.0-only
+
+==========================================
+Enhanced Memory Management (EMM)
+==========================================
+
+Overview
+========
+
+EMM (``CONFIG_ENHANCED_MM``) is a collection of memory-management
+enhancements for memory cgroups maintained in the TKernel source tree.
+It builds on ``CONFIG_MEMCG`` and provides in-kernel APIs, extended
+cgroup interfaces, and additional per-cgroup data structures used by
+the RUE resource-isolation stack.  The individual features are enabled
+through their own Kconfig options and are described below.
+
+The EMM kernel-side code lives in ``kernel/tkernel/emm/`` (built when
+``CONFIG_EMM_RECLAIM`` is set) and in various ``mm/`` files guarded by
+the ``CONFIG_EMM_*`` options.
+
+Features
+========
+
+Force swappiness (``CONFIG_EMM_FORCE_SWAPPINESS``)
+---------------------------------------------------
+
+By default kswapd reclaims anonymous pages for the root cgroup even
+when ``vm.swappiness`` is 0, in order to avoid a system-wide OOM.
+This option disables that fallback so anon pages of the root cgroup
+are never reclaimed when swappiness is 0.
+
+Ramdisk swap tuning (``CONFIG_EMM_RAMDISK_SWAP``)
+-------------------------------------------------
+
+Enables a set of tunings for ramdisk-backed swap devices so that swap
+works better with a memcg-bounded ramdisk.
+
+Workingset tracking (``CONFIG_EMM_WORKINGSET_TRACKING``)
+--------------------------------------------------------
+
+Evaluates per-cgroup memory eviction status.  Because active
+shrinkers reduce the in-memory LRU size and change the workingset
+size, estimating a "raw workingset" helps analyse and improve memory
+usage.
+
+Enhanced memcg support (``CONFIG_EMM_MEMCG``)
+---------------------------------------------
+
+Adds extended memory-management interfaces for memory cgroups, such as
+the per-cgroup asynchronous reclaim knobs ``memory.async_ratio``,
+``memory.async_high``, ``memory.async_low`` and
+``memory.async_distance_factor``.  Asynchronous reclaim is gated by
+``vm.memcg_async`` (see the TKernel feature guide) and requires the
+RUE module to be installed.
+
+Enhanced reclaim (``CONFIG_EMM_RECLAIM``)
+-----------------------------------------
+
+Adds the enhanced reclaim path used by the above interfaces and pulls
+in the EMM module tree under ``kernel/tkernel/emm/``.
+
+Enhanced ZRAM (``CONFIG_EMM_ZRAM_CONF``)
+----------------------------------------
+
+A placeholder option that selects the compression backends
+(CRYPTO_LZO, CRYPTO_ZSTD, CRYPTO_LZ4, CRYPTO_LZ4HC), ``ZSMALLOC`` and
+``MEMCG_ZRAM`` required by the enhanced ZRAM build maintained under
+``kernel/tkernel/emm/driver/block/zram/``.
+
+Batched TLB flush for dirty pages (``CONFIG_EMM_BATCH_DIRTY_TLB_FLUSH``)
+-------------------------------------------------------------------------
+
+The kernel flushes dirty pages one by one on pageout, which can cause
+heavy IPI cost on large machines.  This option batches the TLB
+flushes; it is controlled at runtime through
+``vm.batch_dirty_tlb_flush`` and is experimental.
+
+Module build
+============
+
+The ``kernel/tkernel/emm/`` tree builds three loadable modules
+(``emm_extentions``, ``emm_coreutils`` and ``emm_zram``) with::
+
+    make M=kernel/tkernel/emm modules
+
+The same tree is built into the kernel build when
+``CONFIG_EMM_RECLAIM`` is enabled.