Ying Huang continues to work on his kexec-based hibernation patches. Currently only supporting the i386 architecture, Ying notes, "the setup of hibernation/restore is fairly complex. I will continue working on simplifying." Following up to the latest round of kexec-based hibernation patches posted to the Linux Kernel mailing list it was asked how performance would compare to other hibernation solutions. Ying suggested that with not-yet implemented optimizations it should offer comparable performance:

"In general, for kexec based hibernation, what increases hibernation/wakeup time: One extra Linux boot is needed to hibernate and wakeup. What decreases hibernation/wakeup time: Most hibernation/wakeup work is done in full functional user space program, so it is possible to do some optimization, such as parallel compression.

"So, I think the kexec based hibernation may be slower than original implementation in general. In this prototype implementation, the hibernation/wakeup time is much longer than original hibernation/wakeup implementation. But it has much to be optimized and I think it can approach the speed of the original implementation after optimization."

From:	Huang, Ying
Subject: [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 09:14:05 +0800

Kexec base hibernation has some potential advantages over uswsusp and
TuxOnIce (suspend2). Some most obvious advantages are:

1. The hibernation image size can exceed half of memory size easily.

2. The hibernation image can be written to and read from almost
   anywhere, such as USB disk, NFS.

3. It is possible to eliminate freezer from kexec based hibernation
   implementation.

4. Based on kexec/kdump implementation, the kernel code needed is
   less.


This patch set implements a prototype of kexec based hibernation. The
kernel functionalities added are as follow:

1. Jumping from kexeced kernel back to original kernel. This is used
   by hibernation to save/load necessary state in original kernel and
   jumping back to original kernel after restore the memory of
   original kernel.

2. Add writing support to /dev/oldmem. This is used by hibernation to
   restore the memory of original kernel.


The hibernation process with the patch set is as follow:

1. Boot a kernel A

2. Work under kernel A

3. Kexec another kernel B (crash dump enabled) in kernel A.

4. Save the memory image of kernel A through crash dump (such as "cp
   /proc/vmcore ~"). Save the "jump buffer pfn".

5. Shutdown or reboot


The restore process with the patch set is as follow:

1. Boot a kernel C

2. Kexec another kernel D (crash dump enabled) in kernel C. The memory
   area used by kernel D must be a subset of memory area used by
   kernel B.

3. Restore the memory image of kernel A through /dev/oldmem. Restore
   the "jump buffer pfn".

4. Jump from kernel D back to kernel A

5. Continue work under kernel A


The following user-space tools are needed to implement hibernation and
restore.

1. kexec-tools needs to be patched to support kexec jump. The patch is
   attached in this mail. The precompiled kexec can be download from:
       http://linux-mcr700.sourceforge.net/kjump/kexec

2. Memory image saving tool. Currently, the memory image saving is
   done through: "cp /proc/vmcore <image file>". This will save all
   memory pages of original kernel including the free pages. Maybe the
   crash dump tool "makedumpfile" can be used for this, but it has not
   been tested.

3. Memory image restore tool. A simplest memory image restoring tool
   named "krestore" is implemented. It can be downloaded from
   following URL:
       source: http://linux-mcr700.sourceforge.net/kjump/krestore.tar.gz
       binary: http://linux-mcr700.sourceforge.net/kjump/krestore


Known issues:

1. The kernel B run as a crashdump kernel in reserved memory
   region. This is the biggest constrains of the patch set and planed
   to be eliminated in the future version. That is, instead of
   reserving memory region previously, the needed memory region is
   backupped before kexec and restored after jumping back.

2. Another constrains of the patch set is that the CONFIG_ACPI must be
   turned off to make kexec jump work. Because ACPI will put devices
   into low power state, the kexeced kernel can not be booted properly
   under it. This constrains can be eliminated by separating the
   suspend method and hibernate method of the devices as proposed
   earlier in the LKML.

3. The setup of hibernation/restore is fairly complex. I will continue
   working on simplifying.

4. Memory pages including free pages of kernel A are saved. I think
   the "makedumpfile" tool can be used to exclude "free pages", but I
   have not tested it.

Now, only the i386 architecture is supported. The patch is based on
Linux kernel 2.6.23-rc3-mm1, and has been tested on my IBM T42.


Usage:

1. Compile kernel with following options selected:

CONFIG_X86_32=y
CONFIG_RELOCATABLE=y # not needed strictly, but it is more convenient with it
CONFIG_KEXEC=y
CONFIG_CRASH_DUMP=y # only needed by kexeced kernel to save/restore memory image
CONFIG_PM=y
CONFIG_KEXEC_JUMP=y

2. Download the kexec-tools-testing git tree, apply the kexec-tools
   kjump patch and compile. The kexec-tools kjump patch is appended
   with the mail. The precompiled kexec can be download from:
       http://linux-mcr700.sourceforge.net/kjump/kexec

3. Download and compile the krestore tool. The source code and
   precompiled binary can be downloaded from the following URL:
       source: http://linux-mcr700.sourceforge.net/kjump/krestore.tar.gz
       binary: http://linux-mcr700.sourceforge.net/kjump/krestore

4. Prepare 3 root partition used by kernel A, kernel B, kernel C,
   refered as /dev/hda, /dev/hdb, /dev/hdc in following text. This is
   not strictly necessary, I use this scheme for testing during
   development.

5. Boot kernel compiled for normal usage (kernel A), the reserved
   crash kernel memory region must be added to kernel command line as
   follow:

   crashkernel=<XX>M@<XX>M

   Where, <XX> should be replaced by the real memory size and
   position. I use 16M@32M in my testing.

6. Load kernel compiled for hibernating usage (kernel B) as a
   crashdump kernel with kexec, the same kernel as that of 5 can be
   used if CONFIG_RELOCATABLE=y and CONFIG_CRASH_DUMP=y are
   selected. The kernel command line option as follow should be
   appended to kernel command line.

   kexec_jump_buf_pfn=`cat /sys/kernel/kexec_jump_buf_pfn`

   The --elf64-core-headers should be specified in command line of
   kexec, because only the 64bit ELF is supported by krestore tool.

   For example, the shell command line can be as follow:

   kexec -p /boot/bzImage --elf64-core-headers --append="root=/dev/hdb
       single kexec_jump_buf_pfn=`cat /sys/kernel/kexec_jump_buf_pfn`"

7. Jump to the hibernating kernel (kernel B) with following shell
   command line:

   kexec -j

8. In the hibernating kernel (kernel B), the memory image of
   hibernated kernel (kernel A) can be saved as follow:

   cp /proc/vmcore .
   cp /sys/kernel/kexec_jump_buf_pfn .

9. Shutdown or reboot in hibernating kernel (kernel B).

10. Boot kernel (kernel C) compiled for normal usage in another root
    file system (/dev/hdc).

11. Load restore kernel (kernel D) in kernel C through kexec. This is
    almost same as step 6. The root file system of kernel D should be
    /dev/hdb.

12. In restore kernel (kernel D), the memory image of kernel A can be
    restored as follow:

    # get backup address through kernel command line
    backup_addr=`cat /proc/cmdline | tr ' ' '\n' | grep kexec_backup | cut -d '=' -f 2`
    cp kexec_jump_buf_pfn /sys/kernel/kexec_jump_buf_pfn
    krestore -b $backup_addr vmcore

13. Jump back to hibernated kernel (kernel A)

    kexec -b


---

Changelog of kexec-tools

A command line option -j/--jump-to is added to support jumping
to/executing the currently loaded kernel. A command line option
-b/--jump-back is added to support jumping back to the original
kernel. The implementation of options is to call corresponding
syscall.

For i386 architecture, the address of backup of 0~640k is passed to
kexeced kernel as a kernel command line option. This is used by
kexeced kernel to restore the backup.

This patch is against kexec-tools-testing git tree.



From:	Hu, Fenghua [email blocked]
Subject: RE: [linux-pm] [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 09:28:35 +0800

One quick question is, can it improve hiberation/wakeup time?

Best regards
Hu, Fenghua



From:	Huang, Ying [email blocked]
Subject: RE: [linux-pm] [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 10:16:37 +0800

On Mon, 2007-08-27 at 09:28 +0800, Hu, Fenghua wrote:
> One quick question is, can it improve hiberation/wakeup time?

In general, for kexec based hibernation, what increases
hibernation/wakeup time:

- One extra Linux boot is needed to hibernate and wakeup.


What decreases hibernation/wakeup time:

- Most hibernation/wakeup work is done in full functional user space
program, so it is possible to do some optimization, such as parallel
compression.


So, I think the kexec based hibernation may be slower than original
implementation in general. In this prototype implementation, the
hibernation/wakeup time is much longer than original hibernation/wakeup
implementation. But it has much to be optimized and I think it can
approach the speed of the original implementation after optimization.


For example, the hibernation image is an ordinary ELF file, so it can be
loaded by bootloader directly.

Best Regards,
Huang Ying



From: Nick Piggin [email blocked]
Subject: Re: [linux-pm] [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Wed, 29 Aug 2007 12:41:25 +1000

Huang, Ying wrote:
> On Mon, 2007-08-27 at 09:28 +0800, Hu, Fenghua wrote:
> 
>>One quick question is, can it improve hiberation/wakeup time?
> 
> 
> In general, for kexec based hibernation, what increases
> hibernation/wakeup time:
> 
> - One extra Linux boot is needed to hibernate and wakeup.
> 
> 
> What decreases hibernation/wakeup time:
> 
> - Most hibernation/wakeup work is done in full functional user space
> program, so it is possible to do some optimization, such as parallel
> compression.

- It does not have to reclaim pagecache before suspend?

- It does not have to restore working set afterwards?

(You could do this to reduce image size, of course, but it can
be optional which is nice).


> So, I think the kexec based hibernation may be slower than original
> implementation in general. In this prototype implementation, the
> hibernation/wakeup time is much longer than original hibernation/wakeup
> implementation. But it has much to be optimized and I think it can
> approach the speed of the original implementation after optimization.

Also, don't just look at the time to do a simple suspend/resume cycle,
but the full cost of going from working state to working state (eg.
grep a kernel tree or two!).

Although the kexec details are out of my league, I really like
everything about the concept :) Nice work.

-- 
SUSE Labs, Novell Inc.



From:	Vivek Goyal [email blocked]
Reply-To: [email blocked]
Subject: Re: [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 10:30:27 +0530

On Mon, Aug 27, 2007 at 09:14:05AM +0800, Huang, Ying wrote:
> Kexec base hibernation has some potential advantages over uswsusp and
> TuxOnIce (suspend2). Some most obvious advantages are:
> 
> 1. The hibernation image size can exceed half of memory size easily.
> 
> 2. The hibernation image can be written to and read from almost
>    anywhere, such as USB disk, NFS.
> 
> 3. It is possible to eliminate freezer from kexec based hibernation
>    implementation.
> 
> 4. Based on kexec/kdump implementation, the kernel code needed is
>    less.
> 
> 
> This patch set implements a prototype of kexec based hibernation. The
> kernel functionalities added are as follow:
> 
> 1. Jumping from kexeced kernel back to original kernel. This is used
>    by hibernation to save/load necessary state in original kernel and
>    jumping back to original kernel after restore the memory of
>    original kernel.
> 
> 2. Add writing support to /dev/oldmem. This is used by hibernation to
>    restore the memory of original kernel.
> 
> 
> The hibernation process with the patch set is as follow:
> 
> 1. Boot a kernel A
> 
> 2. Work under kernel A
> 
> 3. Kexec another kernel B (crash dump enabled) in kernel A.
> 
> 4. Save the memory image of kernel A through crash dump (such as "cp
>    /proc/vmcore ~"). Save the "jump buffer pfn".
> 
> 5. Shutdown or reboot
> 
> 
> The restore process with the patch set is as follow:
> 
> 1. Boot a kernel C
> 
> 2. Kexec another kernel D (crash dump enabled) in kernel C. The memory
>    area used by kernel D must be a subset of memory area used by
>    kernel B.
> 

If one compiles the kernel C to boot from reserved memory area (subset
of memory area used by kernel B), then I can skip the step of kexecing
from C to D? (COFIG_PHYSICAL_START)

Alternatively, can we give hint to kernel C to run from a specified address
at run time with the help of command line parameters. What I mean boot-loader
can load the kernel at any address, but kernel will move itself to run
from a different location depending on command line parameter.  For example,
let say kernel_run_addr = 0x1000000. This parameter will tell the kernel
to move itself to 16MB address and run from there. I think it can be made
to work with little work in existing setup of relocatable kernel.

Kernel run address can be put by some user space script which will save
the hibernated image of original kernel. So after saving the /proc/vmcore,
script can modify the boot loader config file to append the right
command line to the kernel (kernel_run_addr).

After hibernation, user will shutdown/reboot. Next time the kernel boots
it will load at 16MB addr (because of kernel_run_addr) and then it can
restore the previously saved image.

We shall have to get rid of (kernel_run_addr) parameter from command
line while resuming. One can restore the image (krestore) and then edit
the boot loader config file to get rid of command line param, kernel_run_addr.

In this scheme, with the help of relocatable kernel, we can use a single
kernel for everything. (A, B, C, D). We will also avoid additional kexec
from kernel C to kernel D.

I think in the long run we shall have to work out so that a user does
not have to maintain multiple kernels. 

Does this make sense?

Thanks
Vivek



From:	"Huang, Ying" <ying.huang@intel.com>
Subject: Re: [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 14:18:49 +0800

On Mon, 2007-08-27 at 10:30 +0530, Vivek Goyal wrote:
> On Mon, Aug 27, 2007 at 09:14:05AM +0800, Huang, Ying wrote:
> > Kexec base hibernation has some potential advantages over uswsusp and
> > TuxOnIce (suspend2). Some most obvious advantages are:
> > 
> > 1. The hibernation image size can exceed half of memory size easily.
> > 
> > 2. The hibernation image can be written to and read from almost
> >    anywhere, such as USB disk, NFS.
> > 
> > 3. It is possible to eliminate freezer from kexec based hibernation
> >    implementation.
> > 
> > 4. Based on kexec/kdump implementation, the kernel code needed is
> >    less.
> > 
> > 
> > This patch set implements a prototype of kexec based hibernation. The
> > kernel functionalities added are as follow:
> > 
> > 1. Jumping from kexeced kernel back to original kernel. This is used
> >    by hibernation to save/load necessary state in original kernel and
> >    jumping back to original kernel after restore the memory of
> >    original kernel.
> > 
> > 2. Add writing support to /dev/oldmem. This is used by hibernation to
> >    restore the memory of original kernel.
> > 
> > 
> > The hibernation process with the patch set is as follow:
> > 
> > 1. Boot a kernel A
> > 
> > 2. Work under kernel A
> > 
> > 3. Kexec another kernel B (crash dump enabled) in kernel A.
> > 
> > 4. Save the memory image of kernel A through crash dump (such as "cp
> >    /proc/vmcore ~"). Save the "jump buffer pfn".
> > 
> > 5. Shutdown or reboot
> > 
> > 
> > The restore process with the patch set is as follow:
> > 
> > 1. Boot a kernel C
> > 
> > 2. Kexec another kernel D (crash dump enabled) in kernel C. The memory
> >    area used by kernel D must be a subset of memory area used by
> >    kernel B.
> > 
> 
> If one compiles the kernel C to boot from reserved memory area (subset
> of memory area used by kernel B), then I can skip the step of kexecing
> from C to D? (COFIG_PHYSICAL_START)

Yes. I think so.

> Alternatively, can we give hint to kernel C to run from a specified address
> at run time with the help of command line parameters. What I mean boot-loader
> can load the kernel at any address, but kernel will move itself to run
> from a different location depending on command line parameter.  For example,
> let say kernel_run_addr = 0x1000000. This parameter will tell the kernel
> to move itself to 16MB address and run from there. I think it can be made
> to work with little work in existing setup of relocatable kernel.
> 
> Kernel run address can be put by some user space script which will save
> the hibernated image of original kernel. So after saving the /proc/vmcore,
> script can modify the boot loader config file to append the right
> command line to the kernel (kernel_run_addr).
> 
> After hibernation, user will shutdown/reboot. Next time the kernel boots
> it will load at 16MB addr (because of kernel_run_addr) and then it can
> restore the previously saved image.
> 
> We shall have to get rid of (kernel_run_addr) parameter from command
> line while resuming. One can restore the image (krestore) and then edit
> the boot loader config file to get rid of command line param, kernel_run_addr.
> 
> In this scheme, with the help of relocatable kernel, we can use a single
> kernel for everything. (A, B, C, D). We will also avoid additional kexec
> from kernel C to kernel D.
> 
> I think in the long run we shall have to work out so that a user does
> not have to maintain multiple kernels. 

Only one relocatable kernel image is needed. In fact, I use one
relocatable kernel image in testing during development.

> Does this make sense?

Yes, this is a sensible optimization. But I think it may be better to
make bootloader load kernel D directly into a specified memory location.
For example, we can add a option to "kernel" command of grub. 

And, I think we can do more in bootloader. Such as we can prepare two
kernel command line for one grub menu item, one for normal kernel, the
other for restore kernel. After hibernating successfully, the effective
kernel command line is switched.

Best Regards,
Huang Ying



From:	Vivek Goyal [email blocked]
Reply-To: [email blocked]
Subject: Re: [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 12:16:04 +0530

On Mon, Aug 27, 2007 at 02:18:49PM +0800, Huang, Ying wrote:

[..]
> > 
> > If one compiles the kernel C to boot from reserved memory area (subset
> > of memory area used by kernel B), then I can skip the step of kexecing
> > from C to D? (COFIG_PHYSICAL_START)
> 
> Yes. I think so.
> 
> > Alternatively, can we give hint to kernel C to run from a specified address
> > at run time with the help of command line parameters. What I mean boot-loader
> > can load the kernel at any address, but kernel will move itself to run
> > from a different location depending on command line parameter.  For example,
> > let say kernel_run_addr = 0x1000000. This parameter will tell the kernel
> > to move itself to 16MB address and run from there. I think it can be made
> > to work with little work in existing setup of relocatable kernel.
> > 
> > Kernel run address can be put by some user space script which will save
> > the hibernated image of original kernel. So after saving the /proc/vmcore,
> > script can modify the boot loader config file to append the right
> > command line to the kernel (kernel_run_addr).
> > 
> > After hibernation, user will shutdown/reboot. Next time the kernel boots
> > it will load at 16MB addr (because of kernel_run_addr) and then it can
> > restore the previously saved image.
> > 
> > We shall have to get rid of (kernel_run_addr) parameter from command
> > line while resuming. One can restore the image (krestore) and then edit
> > the boot loader config file to get rid of command line param, kernel_run_addr.
> > 
> > In this scheme, with the help of relocatable kernel, we can use a single
> > kernel for everything. (A, B, C, D). We will also avoid additional kexec
> > from kernel C to kernel D.
> > 
> > I think in the long run we shall have to work out so that a user does
> > not have to maintain multiple kernels. 
> 
> Only one relocatable kernel image is needed. In fact, I use one
> relocatable kernel image in testing during development.
> 
> > Does this make sense?
> 
> Yes, this is a sensible optimization. But I think it may be better to
> make bootloader load kernel D directly into a specified memory location.
> For example, we can add a option to "kernel" command of grub. 
> 

IIUC, you mean a command line option which is parsed by boot-loader and
then boot-loader loads the kernel at user specified address?

I think it might not be a very good idea as hibernation becomes boot-loader
dependent scheme. How many boot-loaders will one modify and hibernation
will not work with older versions of boot-loader. 

I think it is better to make kernel relocate to user specified address and
keep hibernation mechanism independent of specific boot-loader(grup, lilo,...)
and boot-loader version.

Thanks
Vivek



From:	Pavel Machek [email blocked]
Subject: Re: [RFC][PATCH 0/2 -mm] kexec based hibernation
Date:	Mon, 27 Aug 2007 09:53:17 +0200

Hi!

> > Does this make sense?
> 
> Yes, this is a sensible optimization. But I think it may be better to
> make bootloader load kernel D directly into a specified memory location.
> For example, we can add a option to "kernel" command of grub. 
> 
> And, I think we can do more in bootloader. Such as we can prepare
> two

Yes, that would be nice.

It will mean quite a bit of work, but I guess it should be the long
term goal. Loading restore kernel directly from bootloader means:

1) it is fast -- no need to boot another kernel

2) it is "classical" way of doing things

On the other hand, we loose flexibility that way:

1) it locks you onto one bootloader

2) you no longer have userland there to do uncompression, decryption,
etc..
									Pave
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

• Archive of above thread
• KernelTrap interview with Nick Piggin