Software Suspend - Older News

Yet another new release, incorporating more bug fixes and cleanups. One important change here is work towards moving away from using the PageNosave flag. This will reduce further the number of changes needed outside kernel/power. One thing I haven't been able to test is that pages which are HighMem and Reserved get properly handled by the new code. If you see a message about this, please check your logs and email me so I can know that it works ok or fix it.

I'm mildly hopeful that this might be the last rc before 2.2, but as always, I'll wait and see how it goes on machines my fingers don't touch :)

A release for 2.6.15-rc5 is now available. This includes patches posted to the devel list since the 2.6.14 release was prepared.

rc14 is available. I really shouldn't have started calling them rcs as soon as I decided it was time to work towards 2.2, but there you go. Cleanups dominate the changelog in this release. There has been a huge refactoring of the writers and io code, resulting in the removal of about 400 lines in this release. Fewer places for things to go wrong is always good.

Remounting wasn't reliable on all filesystems, so we've got a new candidate, which looks good so far: freezing filesystems. In addition, there are various other bug fixes and cleanups.

Those using the new fglrx driver have been finding they need to increase Suspend's allowance for extra memory allocated by suspending drivers. This is now a /proc setting. You can set it to zero to make Suspend do a test suspend/resume of drivers at the start, thereby automatically setting the value.

rc12 continues the current bug fix only tradition. It includes fixes for the filewriter and for the remounting code introduced in rc11.

Support for highmem was broken in rc10, and I also forgot about addressing the issues with XFS not syncing data when told to, so rc11 is out quite quickly. This version has working highmem support, and remounts filesystems read only during a cycle. Thanks to Geoffrey Levand at Sony, who gave me the idea at OLS earlier in the year.

I have received a patch implementing support for ARM, but haven't had the time to incorporate it yet.

2.2-rc10 has been released. This version includes a preliminary version of the last new feature that I want to add before merging: discontiguous memory support. It also includes a number of other bug fixes and cleanups. See the changelog for full details.

2.2-rc9 is available. The changelog is primarily bug fixing, but support has been added for a 'userspace storage manager' (usm). USMs will be used to set up access to networked storage. No USMs are currently available. Nigel did hope to include discontiguous memory support in this release, but it's not ready yet.

Those using the filewriter should be aware of the following important warning:

WARNING FOR FILEWRITER + NON EXT2/3 USERS: The filewriter has been designed and tested on ext2/3. XFS (and maybe other filesystems) appears to factor in block sizes in a different way. If you use the filewriter with XFS, you should expect corruption. I'll work to find a way to fix this before rc10.

This revision contains a couple of bug fixes for rc7, plus corrections to the driver model calls and a large number of cleanups. In addition, it changes the name of the /proc directory from software_suspend to suspend2, and removes the /proc/suspend2/interface_version entry. Both of these changes mean that you will need to upgrade to hibernate script 1.12 or newer.

Some more bug fixing and one new little feature. Our sanity check now also compares the root filesystem device prior to resuming, comparing it to that used when suspending began. If they're the same, it will instead invalidate the image. There is one case where the would be the wrong thing to do - for embedded systems with a ram based filesystem, there is a new proc entry to disable this check.

An important bug fix in this release: With rc5, if powering down fails and the powerdown method is not suspend to ram, we fail to reload overwritten pages. This causes memory corruption and on-disk corruption too. Humble apologies to all affected. The release is called Hamilton for later reference - I'm away from Brisbane at the moment and only have intermittent internet access.

A few more bug fixes in this release. The main ones are: We remove a work around in the powering down code that's no longer needed, fix a bug in the swap allocation code, handle drivers reenabling interrupts during resume more nicely (print a message and don't panic.

One important bug found in rc3, which caused an oops in udev has been fixed in this release, along with a few other smaller issues.

2.6.13 is out and so is release candidate 3 of Suspend2 2.2. Lots of bug blatting has been going on. All going well, we're running out of rcs.

Since 2.6.13-rc7 has come out and I already have a few little fixes from 2.2-rc1, I thought I'd do a quick update. Don't feel that you need to race to get this if you've only just installed rc1.

Having finally found the bug that appears to be at the source of some recent problems, I'm happy to announce 2.2-rc1. Bug fixes only in the tree for a while as I work towards what will hopefully be a merge into mm and then mainline.

A few bug fixes against the 2.1.9.13 release here. Hopefully we're not too far from a stable release.

2.1.9.13, prepared for the 2.6.13-rc6 kernel, is now available. This release includes a number of bug fixes and the switchover of the refrigerator to Christoph Lameter's todo list algorithm. Since I'm in the process of splitting suspend up to loooooots of patches, which will then be put into a git tree, this version has only one monolithic patch.

This release is against 2.6.13-rc5. It adds fixes for some refrigerator issues (hopefully the corruption problems), switches I/O waiting code to a new method (addressing another ocassional oops) and a few other less significant bugs.

Users who have been experiencing corruption problems should definitely give this release a try.

This release is against 2.6.13-rc4. Hotplug CPU support has now been tested and gotten working. In addition, the code which tests whether it is safe to resume by looking at filesystem types, has been removed. There are a number of other bug fixes and improvements.

This development release is for 2.6.13-rc3. It includes new support for hotplug cpu. Since Nigel is in Canada at the moment, he has been unable to test this on a true SMP/HT machine. If you get problems, please delete/reverse the hotplug cpu patch.

Other changes in this edition include a fix for some stack corruption introduced earlier in this series, and the reversal of the 'check it is safe to resume' patch. The method by which workqueues are made freezable has also changed, but this shouldn't affect anyone adversely.

The netlink message id changes in this version, so you'll need an updated userui package.

Lots of cleanups from the merge attempt here. You'll only really want to upgrade if you've been experiencing oopses when resuming. This version adds some printks to help with diagnosing that issue.

Lots of small changes in this release, and a few big ones. Kernel user interface components are gone. This means that, to get a 'nice' interface, you now need the userspace user interface programs (also available via the download page).

Test releases have been coming thick and fast! I promise to slow down now. .14 adds a compilation fix for highmem, and leverages the stack moving changes applied earlier in this devel series. With this release, you should see a small increase in the speed of the atomic copies, because we now use memcpy instead of a loop copying 1024 unsigned longs per page.

2.1.8.13 adds bug fixes and further improvements over 2.1.8.10.

2.1.8.10 adds bug fixes against 2.1.8.9. Nothing exciting. I'm aiming for 2.1.9.

This release adds support for replacing the stack when copying the original kernel data back at resume time. All going well, you shouldn't see any difference.

More fixes after the big changes in 2.1.8.5. The swapwriter is no longer deprecated.

A couple of compilation errors have been found in 2.1.8.6. This release addresses those.

A new release is ready for you to test. There are a large number of cleanups in this version, and a few bug fixes. Nigel had hoped to include new CPU hotplugging and SMP S3 support patches from -mm in this release, as well as a rework of the powerdown code. Unfortunately this wasn't yet possible.

Are available here. If you missed out on linux.conf.au 2005, don't worry - it'll be in Dunedin, New Zealand in 2006!

2.1.8.5 is out, containing a few bug fixes and an enhancement over last night's release.

After a loooong time of anticipation, I'm delighted to announce the first public release of Suspend2 that includes the new filewriter. The filewriter allows you to write your image to normal files such as /dev/hda1, /mnt/test/myswapfile or /mnt/nfs/my_network_image... Well, I overstate slightly - the network support is not yet tested except on NBD, where it wasn't able to read the last portion of the image because of issues in the protocol. Nevertheless... this is a good step forward, and one closer to dealing with the objections against getting Suspend2 merged!

Please try out the new userspace user interface components too! You'll need the userspace package, also found on the downloads page.

If you want to feel safe and secure, the old standards are still there, they are just marked as deprecated.

Well, 2.1.8.1 wasn't adequately tested. At least we get to slip a few more fixes in. Humble apologies from Nigel.

Scary! Nigel is updating web pages!

A 2.1.8.1 release is out. This release addresses a few small but significant issues encountered in 2.1.8.

It also incorporates updates to the Suspend userspace interface module. Bernard and I are working toward the userspace user interface module becoming the main user interface module. This means that in 2.1.9, Lord willing, the existing text mode and bootsplash modules will be marked deprecated (as you'll see in menuconfig now), and removed in 2.1.10.

The file writer is also slowly coming along. In similar fashion to the userspace interface changes, the swapwriter will be deprecated in 2.1.9, and go out-of-tree in 2.1.10. It will continue to be available for those who require it, as a separate patch on the web site.

All of this is in preparation for another attempt at merging into mainline!

Bitkeeper repositories for 2.4 and 2.6 kernels are now available. You should be able to pull with bk pull bk://bk.tuxonice.net:40026 (or 40024 for the 2.4 tree). Please note that bk.tuxonice.net is Nigel's machine, so he'd really like it if you pull updates from Linus' and Marcelo's trees first :).

Suspend2 has made the move to become self-hosting at suspend2.net (since renamed tuxonice.net). The new website is at www.tuxonice.net. The mailing lists have also moved, though existing subscribers need not worry about resubscribing, only changing your mail filters. The wiki is also now at wiki.tuxonice.net, along with a new bugzilla for bug-tracking at bugzilla.tuxonice.net.

The main change in this release is new code to improve the speed of freeing memory.

[Download for 2.6.11]

Changes since 2.1.6 are minimal, but includes a fix for a potential source of crashes upon resume, and limiting concurrent I/O transacations which may solve some hangs whilst resuming. It also includes an alpha version of a userspace user interface which will ultimately supercede the bootsplash module - details on setting this up can be found on the wiki in UserUIHowto.

[Download for 2.6.11-rc4] [Download for 2.6.10]

Much work has gone into this release since 2.1.5. Amongst many internal changes, some visible changes include:

• Fewer hang-causing bugs.
• Checking for mounted filesystems before resuming (prevents faulty initrds from corrupting your data).
• Added timeouts to early-boot prompts (BIG FAT WARNINGs).
• GZIP support has been deprecated.

[Download for 2.6.11-rc3]

Cyclades Corporation has agreed to allow Nigel to spend a significant part of his paid hours each week working on issues related to the Suspend2 patches and Power Management in general (in addition to my own time spent on these projects).
Full announcement here.

Suspend2 2.1.5.7B for kernel 2.4.28 is now available. It is intended to be the last release of Suspend2 for the 2.4 kernel series (except for forward porting and bugfixes).

Thanks to Michel Mengis and the The Swiss Federal Institute of Technology, we have a mirror of both the Suspend2 website, and patches! If you are in Europe, you may want to bookmark suspend2.epfl.ch instead! Thanks Michel.

This version adds better implementation for device mapper support, protection against inserting modules when functionality is compiled in, and some more work on fixing the hanging I/O issues.
Announcement     Download (for 2.6.9)

This patch may fix the filesystem corruption observed by some people, amongst other things.
Announcement     Download (for 2.6.9)

This release adds support for the device-mapper, enabling you to use encrypted swap and other media provided by the device-mapper. (Using this requires a statically compiled dmsetup in your initrd).
Announcement     Download (for 2.6.9)

Important: In order to facilitate initramfs and early-userspace support, this release (and onwards) uses do_suspend and do_resume in place of the original activate entry in /proc/software_suspend/. This means you will have to modify your hibernate script (or grab the latest hibernate script v1.00-rc2), and your initrd script also if you are using modular Suspend2.

This release includes a fix for an Oops when aborting whilst writing kernel and process data, extra improvements on SMP, I/O routines, and more.
Announcement     Download (for 2.6.9)

Nigel has put out Suspend2 2.1.1 to resolve a couple of issues people were having, along with a bundle of other changes.
Announcement     Download (for 2.6.9)

It's been far from quiet in the last few months as Nigel has been working tirelessly on Suspend2, and towards getting it merged into the mainstream kernel. Suspend2 2.1 is the latest release to be officially marked as stable (though people have been running development patches for some time quite happily!).

This version includes tons of changes since 2.0, possibly most importantly, the ability to build Suspend2 as modules. This allows it to be loaded from an initrd, opening up all sorts of possibilities.

Nigel's announcement     Download

We're making the switch to http://softwaresuspend.berlios.de/. The mailing lists have also changed; please see the link for new details. The SourceForge lists won't be used for much longer.

A wiki has been created for sharing information relevant to Suspend2. Such items include hardware compatibility, known issues with compilers, known issues with certain patch versions, etc. It is not a place to ask questions - they belong on the mailing list. Please keep it tidy :)

While the development of Suspend2 has been very active, until now the only patches available for 2.4.26 were test patches. A patch for the 2.0 release has been created for 2.4.26, and is available here.

For those not-quite-bleeding-edge junkies that want to hear about the latest, but not be flooded by a mass of bug reports and discussion, a new mailing list has been created. The swsusp-announce list will be very low volume, with only announcements of new versions of Suspend2, patches, and suspend scripts. Click here to subscribe.

Software Suspend 2.0 for Linux 2.4 and 2.6 kernels is now available. The 2.0 release is a major advance over previous versions and includes the following features:

• Support for Highmem (up to 4GB), SMP (on 2.4, 2.6 soon) and preemptive kernels.
• Support for any number of swap partitions and/or files.
• Full asynchronous I/O and readahead for synchronous I/O for maximum throughput.
• Image compression (LZF and GZIP - the former is very fast and highly recommended).
• Support for saving a full image of your memory, resulting in a fast, responsive system after resuming.
• Support for plugins: data transformers (compression, encryption) and new storage backends (NFS support is planned).
• Nice user interface (Bootsplash compatible).
• The ability to specify a maximum image size.
• The ability to cancel a suspend by pressing Escape (for security, this can be disabled).
• Speed and reliability - Software Suspend 2.0 has been extensively tested in a variety of configurations over many months. It is not guaranteed to be perfect, but bugs found will be hunted and fixed quickly.

This website includes extensive documentation, including known issues (primarily DRI and USB support) and FAQs. A well-used mailing list is also available.

Note that two patches are required to use suspend: one for the particular kernel version you are using and a second (applied afterwards) contains the core files.

Special thanks go to Michael Frank for many months of extensive testing of the code, to Marc Lehmann for the work on the LZF compressor, and to Bernard Blackham for maintaining the website and especially to LinuxFund.org for their sponsorship of the project, which has allowed me to work full time on Software Suspend over the last four months.
Announcement. Download (for 2.6.9).

Nigel writes: Core patch 2.0rc5 is now available on Sourceforge. I'm really truly expecting it to be the last rc. This is not to say that I think we've got it completely bug free, but that it's close enough for me to feel happy about calling it 2.0. The main changes since rc4:

• Console printing calls reworked in response to discussions on LKML.
• Initial SCSI support. Your mileage will vary, depending upon how/whether your SCSI driver implemented power management support.
• Confirmed success running on an 8 processor box (except that SCSI power management was broken on the box, so I couldn't get past starting to try to re-read the caches).
• Fixed an issue with reserved Highmem pages causing an oops in copy_pageset1.
• Fixed preemption issues with 2.6.
• Keys for debugging reworked so serial console and normal keyboard use the same keys: P for pausing, S for single step, Space to continue when paused.

Nigel writes: Release candidate 4 is ready for you to test. It should be used with the following updated version specific patches, which are also now available:

• 2.4.21 revision 1
• 2.4.22 revision 1
• 2.4.23 revision 5
• 2.4.24

A comparison table has been compiled by Harry Rockefeller and Nigel comparing the 3 different implementations available in the 2.6 kernel and may clear up some confusion.

No, it hasn't been quiet at all! Thanks to many test patches and our bleeding edge testers, Nigel has announced swsusp 2.0-rc3. There are now two patches one needs to apply to obtain Suspend2. The first is a version-specific patch for your kernel, whether it be 2.4.23 2.4.22 2.4.21 or 2.6.0-test11. The second is the core patch to be applied regardless of your kernel version.

NOTE: The kernel command line parameters have changed. resume= and noresume have been deprecated, in favour of resume2= and noresume2 to avoid clashing with other 2.6 parameters, and which will allow for alternate methods of storing the suspend image (eg, on an NFS mount) in the future. If your resume line currently reads resume=/dev/hda1 then it should be changed to resume2=swap:/dev/hda1. For swap files, similar changes have occured - read Nigel's announcement for the details.

Furthermore, /proc/sys/kernel/swsusp has finally been moved into /proc/swsusp/all_settings. This means you will need to update your suspend scripts accordingly too.

For more gory details, read Nigel's announcement.

This version removes lots of unneeded code leftover from 2.4, fixes some compilation errors, documentation updates and more. For more details, see Nigel's announcement, including release notes.
Download it here (incremental patch against 2.0-2.6-alpha1).

Nigel writes: "I'm pleased to be able to announce the first test release of a port of the current 2.0 pre-release Software Suspend code to 2.6."
For more details, see Nigel's announcement, including release notes.
Download it here (patch against 2.6.0-test8).

2.0-rc2 fixes and squashes more bugs, including:

• Various fixes/improvements to the freezer
• Fix for ACPI continually generating lid events post-resume (subject to further investigation, but this works and has no negative side effects)
• Fix for no progress bar during reading kernel image back
• Fix for bug in switching from bootsplash to text mode progress bar
• Fix/rework for allocating zlib memory when compression support enabled
• Fixing support for swapfiles on file systems with blocksize != 1024 and not aligned on 4096 byte boundaries
• Fix for memory corruption when number of range pages == 53

Read Nigel's announcement
Download incremental patch from 2.0-rc1
Download whole patch against 2.4.22

Due to the many substantial changes in swsusp since 1.0, it was decided that the next release be called 2.0, not 1.1 as originally planned. There will be no 1.1 release.

This patch adds:

• support for suspending lots of kernel daemons (so that modules do not need to be unloaded)
• the Control key now acts as if the noresume option had been passed on the kernel command line
• a "softer" power down (switching off no longer ends in a "clunk" for some users)
• a slight speed increase
• more bugfixes & code cleanups

Read Nigel's announcement
Download incremental patch from 1.1-rc10
Download whole patch against 2.4.22

Improvements on the freezer, extra documentation, shuffling of options patches, and more. Lots of feedback is more than welcome!

Download package of incremental, whole & options patches
Fixes a reject from the above patch (apply after)

This patch fixes two bugs to vastly improve reliability, along with a reworking of the freezer. It also introduces some extra options - one to allow you to specify the expected compression of your swap image (so you no longer require as much swap as the full uncompressed image size to be available).

The second new option allows you to resume multiple times from the same suspend image - this requires you have a read-only root partition that is identically distributed across multiple machines on a network.

For the full story, read Nigel's announcement.
Download incremental patch from 1.1-rc8
Download whole patch against 2.4.22

Another set of bugfixes. Read Nigel's announcement. Apologies for the delayed update of this page!
Download incremental patch from 1.1-rc7
Download whole patch against 2.4.22

"This is a serious contender for 1.1 in my eyes!" -- Nigel.
Download incremental patch from 1.1-rc6
Download whole patch against 2.4.22

Due to increasing spamming on the swsusp-devel list and the fact that instances of SOBIG worm have been found in some of the posts, I had to enable the limitation of posting to subscribed members only. Please, subscribe to the list in order to be able to post.
Sorry for the inconvenience.

rc6 contains "tons of small bug fixes and improvements", writes Nigel. Download the incremental patch from rc5 or a complete patch against 2.4.22. To apply against 2.4.21, you'll need to apply the fixes patch after.

This patch fixes the potential sync corruption (below), completes support for 8139too suspension, and moves the code from kernel/suspend to kernel/power to be inline with the 2.6 kernel (hence it's large size).
Announcement
Download incremental patch from 1.1-rc4
Download whole patch from 2.4.21

A possible source of corruption has been discovered in all versions of swsusp between beta18, through to 1.1-rc4. It is recommended that you read and apply this patch to your current swsusp source to avoid corruption. This is also fixed in swsusp 1.1-rc5.

Fixes compression (broke in rc3), improves speed, and contains an untested patch for suspending the 8139too driver. Announcement. Download.

Small speed increases and bugfixes. You can read the full announcement here. This patch has been distributed as a tarball containing both a patch against a vanilla 2.4.21 kernel, and an incremental patch from 1.1-rc2. Download.

Small bugfixes. Read the announcement and download it here.

Entitled the "Look Mum, no swap partition!" release. Notable changes are:

• support for swapfiles without a swap partition (see note below)
• handle the suspend/resume screen on wide consoles correctly
• support for booting using initrd's (needs testing)

Please take note that this is a development release, and may contain bugs. If you are using swapfile support you should ensure that your data is backed up!

For the weary, read the announcement from Nigel, then follow this link to download it.

A couple of cleanups, bug fixes, and getting ready for swapfile support. Click here for the announcement, and here for the incremental patch against 1.0.2.

Patch against 1.0.1 - "End result should be that you can suspend with compression enabled and shouldn't get errors triggered by fork that were before." Announced by Nigel.

Some minor bugfixes, as announced by Nigel. Changes include preemption fixes, and extra sanity checks. Incremental patch from 1.0 is available here.

Download it here! If you're new to Software Suspend, it's recommended that you peruse the Options, HOWTO, and FAQ sections of this website.