Skip to contentSkip to conversation

ckpt-v20-dev, ckpt-v21-rc1

DISCUSSION Oren Laadan 4 participants 10 messages Mar 30, 2010 – Apr 1, 2010 5 messages not in the archive

lore
Share & tools
lore
#0 lore raw

I pulled all the recent patches in linux-cr (except for ipv6 fixup set), and created the following two branches:

ckpt-v20-dev - patches applied onto pof v20 ckpt-v21-rc1 - patches folded into a clean patchset

Likewise with user-cr (but with more exceptions - working now on pulling more patches in).

This is totally untested except for successful compilation...

Oren.

#2 lore raw
In reply to a message not in the archive
Quoted 7 lines · Oren Laadan

I pulled all the recent patches in linux-cr (except for ipv6 fixup set), and created the following two branches:

ckpt-v20-dev - patches applied onto pof v20 ckpt-v21-rc1 - patches folded into a clean patchset

Quoted 6 lines

Likewise with user-cr (but with more exceptions - working now on pulling more patches in).

This is totally untested except for successful compilation...

Oren.

ckpt-v21-rc1 (aside from the export_symbol patch) passes all tests for me on powerpc, x86-64, and s390, and all manner of config combos with lsms, namespaces, checkpoint=y/n and debug=y/n compile with no errors

-serge

#3 lore raw
In reply to a message not in the archive
Quoted 7 lines · Oren Laadan

I pulled all the recent patches in linux-cr (except for ipv6 fixup set), and created the following two branches:

ckpt-v20-dev - patches applied onto pof v20 ckpt-v21-rc1 - patches folded into a clean patchset

Quoted 6 lines

Likewise with user-cr (but with more exceptions - working now on pulling more patches in).

This is totally untested except for successful compilation...

Oren.

v21 with the recent patches applied seems rock-solid to me, on x86-64, s390x, and powerpc.

recent patches means:
skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

for kernel and

	Add --nonetns switch to user-cr checkpoint
	fix vpids
	user-c/r: get rid of ckpt_hdr_vpids

to user-cr#ckpt-v20-dev

Do we want to try and incorporate Matt's patchset to clear out linux-2.6/checkpoint/ next, or ship what we have?

-serge

#5 lore raw
In reply to a message not in the archive
Quoted 23 lines · Serge E. Hallyn
Quoted 6 lines · Oren Laadan

I pulled all the recent patches in linux-cr (except for ipv6 fixup set), and created the following two branches:

ckpt-v20-dev - patches applied onto pof v20 ckpt-v21-rc1 - patches folded into a clean patchset

Quoted 6 lines

Likewise with user-cr (but with more exceptions - working now on pulling more patches in).

This is totally untested except for successful compilation...

Oren.

v21 with the recent patches applied seems rock-solid to me, on x86-64, s390x, and powerpc.

recent patches means:
skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

Quoted 7 lines

for kernel and

	Add --nonetns switch to user-cr checkpoint
	fix vpids
	user-c/r: get rid of ckpt_hdr_vpids

to user-cr#ckpt-v20-dev

I pushed linux-cr:ckpt-v21-rc2 with:

Quoted 4 lines

skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

and user-cr:ckpt-v20-dev with:

Quoted 2 lines

fix vpids
user-c/r: get rid of ckpt_hdr_vpids

plus Suka's two recent patches.

Dan's --nonetns is pending - see my reply.

Quoted 3 lines

Do we want to try and incorporate Matt's patchset to clear out linux-2.6/checkpoint/ next, or ship what we have?

IIRC there is at least one more comment from fsdevel that I need to address, need to dig into emails again (file_pos_ ?).

I hope to start tackling the transformation to matt's-format by tomorrow, and will see how fast it goes.

Oren.

#7 lore raw
In reply to a message not in the archive
Quoted 50 lines · Oren Laadan
Quoted 23 lines · Serge E. Hallyn
Quoted 6 lines · Oren Laadan

I pulled all the recent patches in linux-cr (except for ipv6 fixup set), and created the following two branches:

ckpt-v20-dev - patches applied onto pof v20 ckpt-v21-rc1 - patches folded into a clean patchset

Quoted 6 lines

Likewise with user-cr (but with more exceptions - working now on pulling more patches in).

This is totally untested except for successful compilation...

Oren.

v21 with the recent patches applied seems rock-solid to me, on x86-64, s390x, and powerpc.

recent patches means:
skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

Quoted 7 lines

for kernel and

	Add --nonetns switch to user-cr checkpoint
	fix vpids
	user-c/r: get rid of ckpt_hdr_vpids

to user-cr#ckpt-v20-dev

I pushed linux-cr:ckpt-v21-rc2 with:

Quoted 4 lines

skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

and user-cr:ckpt-v20-dev with:

Quoted 2 lines

fix vpids
user-c/r: get rid of ckpt_hdr_vpids

plus Suka's two recent patches.

Dan's --nonetns is pending - see my reply.

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Do we want to try and incorporate Matt's patchset to clear out linux-2.6/checkpoint/ next, or ship what we have?

IIRC there is at least one more comment from fsdevel that I need to address, need to dig into emails again (file_pos_ ?).

I hope to start tackling the transformation to matt's-format by tomorrow, and will see how fast it goes.

Alas after I sent this Nathan reported trouble on x86. I haven't gotten a x86-32 partition running yet so can't reproduce. Does anyone else have x86-32 with f12 they can test on?

-serge

#9 lore raw
In reply to a message not in the archive
Quoted 3 lines · Serge E. Hallyn

Alas after I sent this Nathan reported trouble on x86. I haven't gotten a x86-32 partition running yet so can't reproduce. Does anyone else have x86-32 with f12 they can test on?

More detail on this. Seeing programs crash after restart on an i386 KVM guest with Fedora 12 userspace, e.g.

bash-simple.sh[3627] general protection ip:b76d8197 sp:bfe73904 error:0 in libc-2.11.1.so[b760e000+16f000]

kernel: ckpt-v21-rc1 (v2.6.33-108-g9f4401c)

user-cr: ckpt-v20 ( 70a5c7630ce8dd933e60174aba6dc5cb08ea4b41) with CHECKPOINT_NONETNS flag added to checkpoint calls

cr_tests: master (7bd883e17f25d36d66f9b550194c445df2f63e7e) with CHECKPOINT_NONETNS flag added to simple/ckpt.c.

The "simple" (self-checkpoint) testcase restarts successfully without crashing. Everything else seems to get crashes as above. As far as I can tell the memory map is restored correctly and from the kernel's POV the restart is successful.

FWIW, I don't see these crashes on another KVM guest where the only difference is that it is x86_64.

#10 lore raw
Quoted 8 lines · Nathan Lynch
Quoted 3 lines · Serge E. Hallyn

Alas after I sent this Nathan reported trouble on x86. I haven't gotten a x86-32 partition running yet so can't reproduce. Does anyone else have x86-32 with f12 they can test on?

More detail on this. Seeing programs crash after restart on an i386 KVM guest with Fedora 12 userspace, e.g.

Quoted 7 lines

bash-simple.sh[3627] general protection ip:b76d8197 sp:bfe73904 error:0 in libc-2.11.1.so[b760e000+16f000]

kernel: ckpt-v21-rc1 (v2.6.33-108-g9f4401c)

user-cr: ckpt-v20 ( 70a5c7630ce8dd933e60174aba6dc5cb08ea4b41) with CHECKPOINT_NONETNS flag added to checkpoint calls

Confirmed that behavior is the same with user-cr ckpt-v20-dev (7bd883e17f25d36d66f9b550194c445df2f63e7e).

Quoted 11 lines

cr_tests: master (7bd883e17f25d36d66f9b550194c445df2f63e7e) with CHECKPOINT_NONETNS flag added to simple/ckpt.c.

The "simple" (self-checkpoint) testcase restarts successfully without crashing. Everything else seems to get crashes as above. As far as I can tell the memory map is restored correctly and from the kernel's POV the restart is successful.

FWIW, I don't see these crashes on another KVM guest where the only difference is that it is x86_64.

#11 lore raw
Quoted 12 lines · Nathan Lynch
Quoted 11 lines · Nathan Lynch
Quoted 3 lines · Serge E. Hallyn

Alas after I sent this Nathan reported trouble on x86. I haven't gotten a x86-32 partition running yet so can't reproduce. Does anyone else have x86-32 with f12 they can test on?

More detail on this. Seeing programs crash after restart on an i386 KVM guest with Fedora 12 userspace, e.g.

bash-simple.sh[3627] general protection ip:b76d8197 sp:bfe73904 error:0 in libc-2.11.1.so[b760e000+16f000]

Quoted 20 lines
Quoted 4 lines

kernel: ckpt-v21-rc1 (v2.6.33-108-g9f4401c)

user-cr: ckpt-v20 ( 70a5c7630ce8dd933e60174aba6dc5cb08ea4b41) with CHECKPOINT_NONETNS flag added to checkpoint calls

Confirmed that behavior is the same with user-cr ckpt-v20-dev (7bd883e17f25d36d66f9b550194c445df2f63e7e).

Quoted 11 lines

cr_tests: master (7bd883e17f25d36d66f9b550194c445df2f63e7e) with CHECKPOINT_NONETNS flag added to simple/ckpt.c.

The "simple" (self-checkpoint) testcase restarts successfully without crashing. Everything else seems to get crashes as above. As far as I can tell the memory map is restored correctly and from the kernel's POV the restart is successful.

FWIW, I don't see these crashes on another KVM guest where the only difference is that it is x86_64.

I just (finally) finished install and setup of f12 x86 kvm container, and it seems to be passing. (well cloop_parallel seems to need more than the 10 second timeout to check all tasks are restarted, but they do restart - I do have kvm constrained with memory and cpu cgroups :).

-serge

#13 lore raw
In reply to a message not in the archive
Quoted 30 lines · Serge E. Hallyn
Quoted 29 lines · Nathan Lynch
Quoted 8 lines · Nathan Lynch
Quoted 3 lines · Serge E. Hallyn

Alas after I sent this Nathan reported trouble on x86. I haven't gotten a x86-32 partition running yet so can't reproduce. Does anyone else have x86-32 with f12 they can test on?

More detail on this. Seeing programs crash after restart on an i386 KVM guest with Fedora 12 userspace, e.g.

Quoted 7 lines

bash-simple.sh[3627] general protection ip:b76d8197 sp:bfe73904 error:0 in libc-2.11.1.so[b760e000+16f000]

kernel: ckpt-v21-rc1 (v2.6.33-108-g9f4401c)

user-cr: ckpt-v20 ( 70a5c7630ce8dd933e60174aba6dc5cb08ea4b41) with CHECKPOINT_NONETNS flag added to checkpoint calls

Confirmed that behavior is the same with user-cr ckpt-v20-dev (7bd883e17f25d36d66f9b550194c445df2f63e7e).

Quoted 9 lines

cr_tests: master (7bd883e17f25d36d66f9b550194c445df2f63e7e) with CHECKPOINT_NONETNS flag added to simple/ckpt.c.

The "simple" (self-checkpoint) testcase restarts successfully without crashing. Everything else seems to get crashes as above. As far as I can tell the memory map is restored correctly and from the kernel's POV the restart is successful.

Quoted 8 lines
Quoted 3 lines
Quoted 2 lines

FWIW, I don't see these crashes on another KVM guest where the only difference is that it is x86_64.

I just (finally) finished install and setup of f12 x86 kvm container, and it seems to be passing. (well cloop_parallel seems to need more than the 10 second timeout to check all tasks are restarted, but they do restart - I do have kvm constrained with memory and cpu cgroups :).

On the problem guest I booted a 64-bit kernel configured similarly to the i386 one, and the crashes do not occur. That is, I'm running 32-bit userspace on a 64-bit kernel. So the issue is likely specific to the i386 c/r implementation and my particular version of kvm or host kernel (which is also Fedora 12).

#14 lore raw
In reply to a message not in the archive
Quoted 50 lines · Oren Laadan
Quoted 23 lines · Serge E. Hallyn
Quoted 6 lines · Oren Laadan

I pulled all the recent patches in linux-cr (except for ipv6 fixup set), and created the following two branches:

ckpt-v20-dev - patches applied onto pof v20 ckpt-v21-rc1 - patches folded into a clean patchset

Quoted 6 lines

Likewise with user-cr (but with more exceptions - working now on pulling more patches in).

This is totally untested except for successful compilation...

Oren.

v21 with the recent patches applied seems rock-solid to me, on x86-64, s390x, and powerpc.

recent patches means:
skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

Quoted 7 lines

for kernel and

	Add --nonetns switch to user-cr checkpoint
	fix vpids
	user-c/r: get rid of ckpt_hdr_vpids

to user-cr#ckpt-v20-dev

I pushed linux-cr:ckpt-v21-rc2 with:

Quoted 4 lines

skip down interfaces v2,
put fops->checkpoint under ifdef
get rid of ckpt_hdr_vpids,
export net checkpoint fns

and user-cr:ckpt-v20-dev with:

Quoted 2 lines

fix vpids
user-c/r: get rid of ckpt_hdr_vpids

plus Suka's two recent patches.

Dan's --nonetns is pending - see my reply.

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Do we want to try and incorporate Matt's patchset to clear out linux-2.6/checkpoint/ next, or ship what we have?

IIRC there is at least one more comment from fsdevel that I need to address, need to dig into emails again (file_pos_ ?).

I hope to start tackling the transformation to matt's-format by tomorrow, and will see how fast it goes.

Well, my transformation doesn't touch eclone or the cgroup-freezer-related patches so the initial 20 patches won't need to change.

It would be great if we could reduce the number of patches. Some ideas:

Can we push Dave's Namespace menu patch independently? I suspect it's good enough even without c/r at this point.

The cgroup freezer fix patch has gone to Rafael. It only affects frozen cgroups if we do:
echo FROZEN > /cgroup/foo/freezer.state
<suspend system>
<resume system> <-- unfreezes cgroup too!

I've only submitted patches via one tree and then "waited" patiently -- can/should we drop this patch? My instinct is to trust the maintainers to merge things properly and keep it so we don't get a cgroup freezer bug report. But I would like to shorten the number of non-c/r patches in this series...

Could Serge's "c/r: split core function out of some set*{u,g}id functions" or "c/r: break out new_user_ns()" be considered code cleanups or at all useful without c/r? If so, perhaps we can push those separately and with c/r.

I think Nick's comments and the LWN article clearly suggest that the current c/r code organization is a barrier to review. I didn't get the impression that any of the patches added for v21 were complicated enough to warrant another round of limited review amongst ourselves. The longer we wait the harder it will be to do. So I think reorganizing the code now, before v21 is the way to go.

Cheers,
-Matt Helsley