Skip to contentSkip to conversation

How do I remove a patch buried in your *-mm series?

DISCUSSION Paul Jackson 2 participants 5 messages Dec 4, 2005

lore
Share & tools
lore
#0 lore raw

I want to remove the patch in *-mm:

1) cpuset-change-marker-for-relative-numbering.patch

Unfortunately, it collides with another couple cpuset patches later in your stack:

2) cpuset-memory-pressure-meter.patch, cpuset-memory-pressure-meter-gcc-295-fix.patch

How should I do this so I minimize the amount of cussing you do in my general direction:

A. Just ask you to nuke patch (1) above; let you edit the mess. B. Ask you to nuke both (1) and (2); leave me to resend a (2) that applies. C. Send a reversing patch that applies on top of your current *-mm stack. D. Some other plan you would prefer.

I have verified that removing all the patches above applies cleanly and builds, with just a harmless -74 lines offset on one of the remaining cpuset patches.

So I recommend B.

Separately I will send a patch to remove the bit of Documentation/cpusets.txt that described this feature.

Details for the historical record:
I either need to go one step forward with it (fix a bug so that it zeros the marker_pid of the left behind cpuset when attach_task moves a task), or I need to go five steps backward, with a different approach. But I have other stuff to do first, so should avoid digging this "change-marker-for-relative-numbering" any deeper than it is for now.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#1 lore raw
Quoted 10 lines · Paul Jackson

I want to remove the patch in *-mm:

1) cpuset-change-marker-for-relative-numbering.patch

Unfortunately, it collides with another couple cpuset patches later in your stack:

2) cpuset-memory-pressure-meter.patch, cpuset-memory-pressure-meter-gcc-295-fix.patch

Quoted 7 lines

How should I do this so I minimize the amount of cussing you do in my general direction:

A. Just ask you to nuke patch (1) above; let you edit the mess. B. Ask you to nuke both (1) and (2); leave me to resend a (2) that applies. C. Send a reversing patch that applies on top of your current *-mm stack. D. Some other plan you would prefer.

E. You send a _minimal_ patch against lastest mm, telling me
"this fixes a bug in cpuset-change-marker-for-relative-numbering.patch". Then I insert it into the series with name cpuset-change-marker-for-relative-numbering-fix.patch and it gets folded into cpuset-change-marker-for-relative-numbering.patch prior to going to Linus.

Usually people forget to tell me which patch it fixes, but I work it out.

In this case, dropping cpuset-change-marker-for-relative-numbering.patch works fine too - it took 20 seconds to fix up the rejects. Mainly by simply omitting them, because all this stuff:

***************
*** 191,199 ****
.cpus_allowed = CPU_MASK_ALL, .mems_allowed = NODE_MASK_ALL, .marker_pid = 0,
- .fmeter.cnt = 0,
- .fmeter.val = 0,
- .fmeter.time = 0,
.count = ATOMIC_INIT(0),
.sibling = LIST_HEAD_INIT(top_cpuset.sibling), .children = LIST_HEAD_INIT(top_cpuset.children),
--- 191,201 ----
.cpus_allowed = CPU_MASK_ALL, .mems_allowed = NODE_MASK_ALL, .marker_pid = 0,
+ .fmeter = {
+ .cnt = 0,
+ .val = 0,
+ .time = 0,
+ },
.count = ATOMIC_INIT(0),
.sibling = LIST_HEAD_INIT(top_cpuset.sibling), .children = LIST_HEAD_INIT(top_cpuset.children),

It just redundant - the compiler does that.

Quoted 5 lines

I have verified that removing all the patches above applies cleanly and builds, with just a harmless -74 lines offset on one of the remaining cpuset patches.

So I recommend B.

Your call. E is preferred though.

#2 lore raw

Andrew wrote:
E. You send a _minimal_ patch ... prior
to going to Linus.

No! This isn't (for now at least) a fix. It's a nuke.

I don't want that "cpuset-change-marker-for-relative-numbering.patch" going to Linus, because I am hesitating whether I even want that "feature".

I want to nuke it now. You might see it again, fixed up, or you might never see it again (I doubt you'll miss it ;). I don't know yet.

Yes - the collisions resulting from removing this patch are easy to edit, if that fits your style. If you hadn't removed the useless silly zero initializers, I would have in my next set patches. It was on my todo list, thanks to an earlier comment of yours.

So ... not E ;). What's your preference now?

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#3 lore raw
Quoted 2 lines · Paul Jackson

So ... not E ;). What's your preference now?

http://www.zip.com.au/~akpm/linux/patches/stuff/x.bz2 is a patch against -rc4 containing everything in -mm up to and including

...
radix-tree-code-consolidation radix_tree-early-termination-of-tag-clearing radix-tree-reduce-tree-height-upon-partial-truncation slob-introduce-mm-utilc-for-shared-functions slob-introduce-the-slob-allocator slob-introduce-the-slob-allocator-fixes cpuset-better-bitmap-remap-defaults cpuset-mempolicy-one-more-nodemask-conversion cpuset-memory-pressure-meter cpuset-memory-pressure-meter-gcc-295-fix cpuset-document-additional-features

So a patch against that will be ideal.

(It's very easy for me to create such a rollup - one has but to ask...)

#4 lore raw
Quoted 1 line · Andrew

So a patch against that will be ideal.

Ok - the "change-marker-for-relative-numbering" patch I wanted gone is gone. Good.

I don't know what "patch against that" you have in mind. But I guess that doesn't matter.

I will make sure I am in sync with that rc4 + that mm rollup patch + whatever else follows it in mm, for whatever other work I do.

Thanks.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401