| 1 |
===== |
| 2 |
zswap |
| 3 |
===== |
| 4 |
|
| 5 |
Overview |
| 6 |
======== |
| 7 |
|
| 8 |
Zswap is a lightweight compressed cache for swap pages. It takes pages that are |
| 9 |
in the process of being swapped out and attempts to compress them into a |
| 10 |
dynamically allocated RAM-based memory pool. zswap basically trades CPU cycles |
| 11 |
for potentially reduced swap I/O. This trade-off can also result in a |
| 12 |
significant performance improvement if reads from the compressed cache are |
| 13 |
faster than reads from a swap device. |
| 14 |
|
| 15 |
Some potential benefits: |
| 16 |
|
| 17 |
* Desktop/laptop users with limited RAM capacities can mitigate the |
| 18 |
performance impact of swapping. |
| 19 |
* Overcommitted guests that share a common I/O resource can |
| 20 |
dramatically reduce their swap I/O pressure, avoiding heavy handed I/O |
| 21 |
throttling by the hypervisor. This allows more work to get done with less |
| 22 |
impact to the guest workload and guests sharing the I/O subsystem |
| 23 |
* Users with SSDs as swap devices can extend the life of the device by |
| 24 |
drastically reducing life-shortening writes. |
| 25 |
|
| 26 |
Zswap evicts pages from compressed cache on an LRU basis to the backing swap |
| 27 |
device when the compressed pool reaches its size limit. This requirement had |
| 28 |
been identified in prior community discussions. |
| 29 |
|
| 30 |
Whether Zswap is enabled at the boot time depends on whether |
| 31 |
the ``CONFIG_ZSWAP_DEFAULT_ON`` Kconfig option is enabled or not. |
| 32 |
This setting can then be overridden by providing the kernel command line |
| 33 |
``zswap.enabled=`` option, for example ``zswap.enabled=0``. |
| 34 |
Zswap can also be enabled and disabled at runtime using the sysfs interface. |
| 35 |
An example command to enable zswap at runtime, assuming sysfs is mounted |
| 36 |
at ``/sys``, is:: |
| 37 |
|
| 38 |
echo 1 > /sys/module/zswap/parameters/enabled |
| 39 |
|
| 40 |
When zswap is disabled at runtime it will stop storing pages that are |
| 41 |
being swapped out. However, it will _not_ immediately write out or fault |
| 42 |
back into memory all of the pages stored in the compressed pool. The |
| 43 |
pages stored in zswap will remain in the compressed pool until they are |
| 44 |
either invalidated or faulted back into memory. In order to force all |
| 45 |
pages out of the compressed pool, a swapoff on the swap device(s) will |
| 46 |
fault back into memory all swapped out pages, including those in the |
| 47 |
compressed pool. |
| 48 |
|
| 49 |
Design |
| 50 |
====== |
| 51 |
|
| 52 |
Zswap receives pages for compression from the swap subsystem and is able to |
| 53 |
evict pages from its own compressed pool on an LRU basis and write them back to |
| 54 |
the backing swap device in the case that the compressed pool is full. |
| 55 |
|
| 56 |
Zswap makes use of zsmalloc for the managing the compressed memory pool. Each |
| 57 |
allocation in zsmalloc is not directly accessible by address. Rather, a handle is |
| 58 |
returned by the allocation routine and that handle must be mapped before being |
| 59 |
accessed. The compressed memory pool grows on demand and shrinks as compressed |
| 60 |
pages are freed. The pool is not preallocated. |
| 61 |
|
| 62 |
When a swap page is passed from swapout to zswap, zswap maintains a mapping of |
| 63 |
the swap entry, a combination of the swap type and swap offset, to the zsmalloc |
| 64 |
handle that references that compressed swap page. This mapping is achieved |
| 65 |
with an xarray per swap type. The swap offset is the search key for the xarray |
| 66 |
nodes. |
| 67 |
|
| 68 |
During a page fault on a PTE that is a swap entry, the swapin code calls the |
| 69 |
zswap load function to decompress the page into the page allocated by the page |
| 70 |
fault handler. |
| 71 |
|
| 72 |
Once there are no PTEs referencing a swap page stored in zswap (i.e. the count |
| 73 |
in the swap_map goes to 0) the swap code calls the zswap invalidate function |
| 74 |
to free the compressed entry. |
| 75 |
|
| 76 |
Zswap seeks to be simple in its policies. Sysfs attributes allow for one user |
| 77 |
controlled policy: |
| 78 |
|
| 79 |
* max_pool_percent - The maximum percentage of memory that the compressed |
| 80 |
pool can occupy. |
| 81 |
|
| 82 |
The default compressor is selected in ``CONFIG_ZSWAP_COMPRESSOR_DEFAULT`` |
| 83 |
Kconfig option, but it can be overridden at boot time by setting the |
| 84 |
``compressor`` attribute, e.g. ``zswap.compressor=lzo``. |
| 85 |
It can also be changed at runtime using the sysfs "compressor" |
| 86 |
attribute, e.g.:: |
| 87 |
|
| 88 |
echo lzo > /sys/module/zswap/parameters/compressor |
| 89 |
|
| 90 |
When the compressor parameter is changed at runtime, any existing compressed |
| 91 |
pages are not modified; they are left in their own pool. When a request is |
| 92 |
made for a page in an old pool, it is uncompressed using its original |
| 93 |
compressor. Once all pages are removed from an old pool, the pool and its |
| 94 |
compressor are freed. |
| 95 |
|
| 96 |
Some of the pages in zswap are same-value filled pages (i.e. contents of the |
| 97 |
page have same value or repetitive pattern). These pages include zero-filled |
| 98 |
pages and they are handled differently. During store operation, a page is |
| 99 |
checked if it is a same-value filled page before compressing it. If true, the |
| 100 |
compressed length of the page is set to zero and the pattern or same-filled |
| 101 |
value is stored. |
| 102 |
|
| 103 |
To prevent zswap from shrinking pool when zswap is full and there's a high |
| 104 |
pressure on swap (this will result in flipping pages in and out zswap pool |
| 105 |
without any real benefit but with a performance drop for the system), a |
| 106 |
special parameter has been introduced to implement a sort of hysteresis to |
| 107 |
refuse taking pages into zswap pool until it has sufficient space if the limit |
| 108 |
has been hit. To set the threshold at which zswap would start accepting pages |
| 109 |
again after it became full, use the sysfs ``accept_threshold_percent`` |
| 110 |
attribute, e. g.:: |
| 111 |
|
| 112 |
echo 80 > /sys/module/zswap/parameters/accept_threshold_percent |
| 113 |
|
| 114 |
Setting this parameter to 100 will disable the hysteresis. |
| 115 |
|
| 116 |
Some users cannot tolerate the swapping that comes with zswap store failures |
| 117 |
and zswap writebacks. Swapping can be disabled entirely (without disabling |
| 118 |
zswap itself) on a cgroup-basis as follows:: |
| 119 |
|
| 120 |
echo 0 > /sys/fs/cgroup/<cgroup-name>/memory.zswap.writeback |
| 121 |
|
| 122 |
Note that if the store failures are recurring (for e.g if the pages are |
| 123 |
incompressible), users can observe reclaim inefficiency after disabling |
| 124 |
writeback (because the same pages might be rejected again and again). |
| 125 |
|
| 126 |
When there is a sizable amount of cold memory residing in the zswap pool, it |
| 127 |
can be advantageous to proactively write these cold pages to swap and reclaim |
| 128 |
the memory for other use cases. By default, the zswap shrinker is disabled. |
| 129 |
User can enable it as follows:: |
| 130 |
|
| 131 |
echo Y > /sys/module/zswap/parameters/shrinker_enabled |
| 132 |
|
| 133 |
This can be enabled at the boot time if ``CONFIG_ZSWAP_SHRINKER_DEFAULT_ON`` is |
| 134 |
selected. |
| 135 |
|
| 136 |
A debugfs interface is provided for various statistic about pool size, number |
| 137 |
of pages stored, same-value filled pages and various counters for the reasons |
| 138 |
pages are rejected. |