Gentoo Websites Logo
Go to: Gentoo Home Documentation Forums Lists Bugs Planet Store Wiki Get Gentoo!
Bug 479382 - sys-kernel/ck-sources-{3.9.11-r1;3.10.17} version bump
Summary: sys-kernel/ck-sources-{3.9.11-r1;3.10.17} version bump
Status: RESOLVED FIXED
Alias: None
Product: Gentoo Linux
Classification: Unclassified
Component: [OLD] Core system (show other bugs)
Hardware: All Linux
: Normal enhancement (vote)
Assignee: Eric F. GARIOUD
URL:
Whiteboard:
Keywords: EBUILD
Depends on:
Blocks:
 
Reported: 2013-08-01 11:02 UTC by Eric F. GARIOUD
Modified: 2013-11-18 19:49 UTC (History)
4 users (show)

See Also:
Package list:
Runtime testing required: ---


Attachments
ck-sources-3.9.11-r1.ebuild (ck-sources-3.9.11-r1.ebuild,2.52 KB, text/plain)
2013-08-01 11:05 UTC, Eric F. GARIOUD
Details
ck-sources-3.10.4.ebuild (ck-sources-3.10.4.ebuild,2.43 KB, text/plain)
2013-08-01 11:07 UTC, Eric F. GARIOUD
Details
ck-sources-3.9-3.10-config_rcu-aCOSwt_P8.patch (ck-sources-3.9-3.10-config_rcu-aCOSwt_P8.patch,643 bytes, patch)
2013-08-01 11:10 UTC, Eric F. GARIOUD
Details | Diff
ck-sources-3.9.11-r1.ebuild (ck-sources-3.9.11-r1.ebuild,2.44 KB, text/plain)
2013-11-18 17:20 UTC, Eric F. GARIOUD
Details
ck-sources-3.10.17.ebuild (ck-sources-3.10.17.ebuild,2.55 KB, text/plain)
2013-11-18 17:22 UTC, Eric F. GARIOUD
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Eric F. GARIOUD 2013-08-01 11:02:45 UTC
@proxy-maintainers : Please update the tree with :
- the ebuild for the 3.9.11-r1 release
- the ebuild for the 3.10.4 release
- the ck-sources-3.9-3.10-config_rcu-aCOSwt_P8.patch file.

Note : As written in Bug 472842, the BFS does not support the Full Dynticks kernel option.
CONFIG_NOHZ_FULL has been made depending on !SCHED_BFS as part of the 3.10-sched-bfs-440.patch, this directly disabling all options depending on CONFIG_NOHZ_FULL.

However :

There remain a couple of options not depending on CONFIG_NOHZ_FULL but related to the Full Dynticks development, options which actually break the build of the kernel if selected : CONFIG_RCU_USER_QS and CONFIG_RCU_NOCB_CPU.

The purpose of the attached patch is to make these two options !SCHED_BFS dependent.
This new patch also applies to 3.9.11-r1

BTW : As in Bug 472842, I acknowledge that this local patch should be part of the ck patchset. Be aware that CK acknowledged this too... a posteriori. 

Reproducible: Always
Comment 1 Eric F. GARIOUD 2013-08-01 11:05:07 UTC
Created attachment 354798 [details]
ck-sources-3.9.11-r1.ebuild

Ebuild for the 3.9.11 release.

Based on :

- Sources for the linux 3.9 branch
- genpatches rev 17
- 3.9-ck1 patchset including bfs-430
Comment 2 Eric F. GARIOUD 2013-08-01 11:07:46 UTC
Created attachment 354800 [details]
ck-sources-3.10.4.ebuild

Ebuild for the 3.10.4 release.

Based on :

- Sources for the linux 3.10 branch
- genpatches rev 8
- 3.10-ck1 patchset including bfs-440
Comment 3 Eric F. GARIOUD 2013-08-01 11:10:46 UTC
Created attachment 354802 [details, diff]
ck-sources-3.9-3.10-config_rcu-aCOSwt_P8.patch

Patch modifying init/Kconfig in order to make CONFIG_RCU_USER_QS and CONFIG_RCU_NOCB_CPU depend on !SCHED_BFS
Comment 4 Sergey Popov gentoo-dev 2013-08-04 13:18:26 UTC
phantom linux-3.9.11-ck # make

[snip]

Setup is 16028 bytes (padded to 16384 bytes).
System is 4324 kB
CRC 725f242b
Kernel: arch/x86/boot/bzImage is ready  (#1)

phantom linux-3.9.11-ck # cat .config | grep -E _BFS=\|RCU_USER_QS=\|RCU_NOCB_CPU=
CONFIG_SCHED_BFS=y
CONFIG_RCU_USER_QS=y
CONFIG_RCU_NOCB_CPU=y

So it builds successfully with this combination of options. Do not know about run-time problems, though...
Comment 5 Eric F. GARIOUD 2013-09-07 13:27:28 UTC
(In reply to Sergey Popov from comment #4)
Should I understand that you did not and will not commit the above ebuilds & patches because you find some code can be built under your "phantom" without the patch ?
Comment 6 Markos Chandras (RETIRED) gentoo-dev 2013-09-07 13:42:02 UTC
(In reply to Eric F. GARIOUD from comment #5)
> (In reply to Sergey Popov from comment #4)
> Should I understand that you did not and will not commit the above ebuilds &
> patches because you find some code can be built under your "phantom" without
> the patch ?

I've told you in the past that we should not carry Kconfig patches that are not upstream (ck-* upstream). If the ck-* upstream thinks that these symbols should depend on !*BFS then they should add it to their patches.

Having you patching the Kconfig this way, only delays the submission of the patchset since we need to verify that your custom patches actually make sense.
Comment 7 Sergey Popov gentoo-dev 2013-09-09 08:12:46 UTC
(In reply to Eric F. GARIOUD from comment #5)
> (In reply to Sergey Popov from comment #4)
> Should I understand that you did not and will not commit the above ebuilds &
> patches because you find some code can be built under your "phantom" without
> the patch ?

If it can be built(and run of course) without these patches, then, two options:

1) patches are completely wrong;
2) patches covers some failures by completely cut off some features(e.g. they should be more detailed, not just !BFS, some more options).

You choose. We do not want to commit patches in tree which is wrong by their nature. Could you post a link to discussion with upstream on some forum or maillist about this issue?
Comment 8 Eric F. GARIOUD 2013-09-09 08:26:47 UTC
(In reply to Sergey Popov from comment #7)

> Could you post a link to discussion with upstream on some forum or
> maillist about this issue?

http://ck-hack.blogspot.fr/2013/07/bfs-0440-ck-for-linux-310.html?showComment=1373570298381#c1615993220632377640
Comment 9 Markos Chandras (RETIRED) gentoo-dev 2013-09-09 08:38:05 UTC
There are hundreds of Kconfig symbol combinations that break the linux kernel build. It's not possible to restrict them all and it actually makes no sense to do it. I don't see the benefit of selectively disable certain symbols.
Comment 10 Tom Wijsman (TomWij) (RETIRED) gentoo-dev 2013-09-09 10:02:10 UTC
(In reply to Eric F. GARIOUD from comment #3)
> Created attachment 354802 [details, diff] [details, diff]
> ck-sources-3.9-3.10-config_rcu-aCOSwt_P8.patch
> 
> Patch modifying init/Kconfig in order to make CONFIG_RCU_USER_QS and
> CONFIG_RCU_NOCB_CPU depend on !SCHED_BFS

Actually, that is the other way around; SCHED_BFS should depend on !RCU_USER_QS and !RCU_NOCB_CPU, because that is the feature being introduced here and therefore that should be the feature checking if everything is alright for that feature to be introduced. Also make sure that BFS upstream is aware of this.
Comment 11 Sergey Popov gentoo-dev 2013-09-09 12:56:45 UTC
(In reply to Tom Wijsman (TomWij) from comment #10)
> (In reply to Eric F. GARIOUD from comment #3)
> > Created attachment 354802 [details, diff] [details, diff] [details, diff]
> > ck-sources-3.9-3.10-config_rcu-aCOSwt_P8.patch
> > 
> > Patch modifying init/Kconfig in order to make CONFIG_RCU_USER_QS and
> > CONFIG_RCU_NOCB_CPU depend on !SCHED_BFS
> 
> Actually, that is the other way around; SCHED_BFS should depend on
> !RCU_USER_QS and !RCU_NOCB_CPU, because that is the feature being introduced
> here and therefore that should be the feature checking if everything is
> alright for that feature to be introduced. Also make sure that BFS upstream
> is aware of this.

According to link above, upstream is aware and possibly planning to do it in next releases, but i suggest to commit current versions as-is, without the patch.
Comment 12 Markos Chandras (RETIRED) gentoo-dev 2013-10-19 13:12:00 UTC
3.10.4 is no longer applicable
Comment 13 Eric F. GARIOUD 2013-11-18 17:20:43 UTC
Created attachment 363500 [details]
ck-sources-3.9.11-r1.ebuild

Custom-Patch-Less-Ebuild for the 3.9.11-r1 release.

Based on :

- Sources for the linux 3.9 branch
- genpatches rev 17
- 3.9-ck1 patchset including bfs-430
Comment 14 Eric F. GARIOUD 2013-11-18 17:22:47 UTC
Created attachment 363502 [details]
ck-sources-3.10.17.ebuild

Custom-Patch-Less-Ebuild for the 3.10.17 release.

Based on :

- Sources for the linux 3.10 branch
- genpatches rev 24
- 3.10-ck1 patchset including bfs-440
Comment 15 Tom Wijsman (TomWij) (RETIRED) gentoo-dev 2013-11-18 18:13:24 UTC
Removed experimental from IUSE and set K_EXP_GENPATCHES_NOUSE, as changing the USE flag results did not result in any difference; because:

 - the first "use experimental" occurrence from the eclass is not evaluated due to K_EXP_GENPATCHES_PULL being set;

 - the second occurrence from the eclass results in the same behavior as the K_EXP_GENPATCHES_LIST definition is the same as the surrounding loop, and thus evaluating the loop or skipping the loop gives you the exact same result as nothing gets added to UNIPATCH_DROP in either case;

 - there are no USE flag conditional constructions in the ebuild itself.

If you have intended something different to happen than was defined: Can you clarify what you want to see happen? The re-addition of the USE flag will cause a rebuild so a revision bump isn't necessary to fix this up for different behavior.

+  18 Nov 2013; Tom Wijsman <TomWij@gentoo.org> +ck-sources-3.10.17.ebuild,
+  +ck-sources-3.9.11-r1.ebuild:
+  Version bumps 3.9.11-r1 (3.9 branch, genpatches rev 17, 3.9-ck1 patchset
+  including bfs-430) and 3.10.17 (3.10 branch, genpatches rev 24, 3.10-ck1
+  patchset including bfs-440), proxied commit for Eric F. GARIOUD.

Thank you for your contribution.
Comment 16 Eric F. GARIOUD 2013-11-18 19:49:01 UTC
(In reply to Tom Wijsman (TomWij) from comment #15)
Everything is absolutely fine as far as this bug is now closed.
Thank *you* Tom for *your* contribution.