Based on access analysis with strace, I determined an essentially
minimal required set of files from the kernel source that was needed to
build all current kernel packages on 3.10, which ultimately resulted in
keeping 30M of source. Generalizing from that minimal set, which
required ad-hoc specifications of which headers outside of include/ and
arch/*/include and which files in the scripts/ directory should be kept,
to a policy of keeping all non-arch-specific headers that aren't part of
the drivers/ directory and the entire scripts/ directory added an
additional 17M, but there was nothing in the analysis that indicated
that that ad-hoc specification was at all complete so I think the extra
hit is worth the likely greater compatibility.
For reference, we now keep:
* All headers that are NOT in arch/${notTargetArch}/include or drivers/
* The scripts/ directory
* Makefile
* arch/${targetArch}/Makefile
IMO the most likely cause of future problems are the headers in
drivers/, but hopefully they won't actually be needed as they add 50M
Ideally kernel packages would only use include and
arch/${targetArch}/include, but alas this is observably not the case.
master:
* $out
* size: 234M
* references-closure: linux-headers, glibc, attr, acl, zlib, gcc,
coreutils, perl, bash
merge-kernel-builds:
* $out
* size: 152M
* references-closure: none
* $dev
* size: 57M
* references-closure: linux-headers, glibc, zlib, gcc
So even with the non-minimal set we still beat out master. Keeping the
drivers headers would make us only slightly bigger.
Signed-off-by: Shea Levy <shea@shealevy.com>
In most cases, this just meant changing kernelDev (now removed from
linuxPackagesFor) to kernel.dev. Some packages needed more work (though
whether that was because of my changes or because they were already
broken, I'm not sure). Specifics:
* psmouse-alps builds on 3.4 but not 3.10, as noted in the comments that
were already there
* blcr builds on 3.4 but not 3.10, as noted in comments that were
already there
* open-iscsi, ati-drivers, wis-go7007, and openafsClient don't build on
3.4 or 3.10 on this branch or on master, so they're marked broken
* A version-specific kernelHeaders package was added
The following packages were removed:
* atheros/madwifi is superceded by official ath*k modules
* aufs is no longer used by any of our kernels
* broadcom-sta v6 (which was already packaged) replaces broadcom-sta
* exmap has not been updated since 2011 and doesn't build
* iscis-target has not been updated since 2010 and doesn't build
* iwlwifi is part of mainline now and doesn't build
* nivida-x11-legacy-96 hasn't been updated since 2008 and doesn't build
Everything not specifically mentioned above builds successfully on 3.10.
I haven't yet tested on 3.4, but will before opening a pull request.
Signed-off-by: Shea Levy <shea@shealevy.com>
This makes the disk usage footprint of building the kernel smaller in 3
ways:
1) There is no separate kernel source derivation
2) Rather than using the entire build tree, only the output of make
modules_prepare is kept in the $dev output (plus the module symbol
versioning file generated during the build)
3) Only the subset of the source tree known to be needed for external
builds is kept in $dev
Note that while 2) is supported by official kernel documentation, I
couldn't find any source describing what we need to keep for 3). I've
started with the bare minimum (the main Makefile is called by the
Makefile generated by make modules_prepare) and we can/should add more
as needed for kernelPackages.
Signed-off-by: Shea Levy <shea@shealevy.com>
This has three major benefits:
1. We no longer have two kernel build processes to maintain
2. The build process is (IMO) cleaner and cleaves more closely to
upstream. In partuclar, we use make install to install the kernel and
development source/build trees, eliminating the guesswork about which
files to copy.
3. The derivation has multiple outputs: the kernel and modules are in
the default `out' output, while the build and source trees are in a
`dev' output. This makes it possible for the full source and build tree
to be kept (which is expected by out-of-tree modules) without bloating
the closure of the system derivation.
In addition, if a solution for how to handle queries in the presence of
imports from derivations ever makes it into nix, a framework for
querying the full configuration of the kernel in nix expressions is
already in place.
Signed-off-by: Shea Levy <shea@shealevy.com>
In the most general case, the cross and native kernel may differ in
patches and configuration file as well as architecture, kernel target,
etc. It's probably overkill to support that case, but since it was
doable without much duplication and it will make integrating with the
existing cross-compilation support in the generic kernel I decided to
implement it anyway.
Signed-off-by: Shea Levy <shea@shealevy.com>
nix's version parsing treats the previous name as a package named
`linux' with version `${version}-source', when we really want a package
named `linux-source' with version `${version}'
Signed-off-by: Shea Levy <shea@shealevy.com>
This only affects the `oldaskconfig' make target, so it shouldn't really
affect current manual-config users, but it does make it more
straightforward to implement the generic kernel build on top of
manual-config.
Signed-off-by: Shea Levy <shea@shealevy.com>
If the config attrset is manually specified, we still want isYes,
isModule, etc. to work. But we let the passed in config attrset take
precedence, if for some reason the caller wants to provide their own
implementation of one or more of these functions.
Signed-off-by: Shea Levy <shea@shealevy.com>
With this, I was able to successfully compile a defconfig kernel for the
sheevaplug, though I didn't actually try to run it (not having a
sheevaplug myself).
For native compiles, the most significant difference is that the
platform's kernel target is built directly rather than hoping the
default make target will pull it in.
Also some stylistic improvements along the way.
Signed-off-by: Shea Levy <shea@shealevy.com>
The function ‘mkDerivation’ now checks whether the current platform
type is included in a package's meta.platform field. If not, it
throws an exception:
$ nix-build -A linux --argstr system x86_64-darwin
error: user-thrown exception: the package ‘linux-3.10.15’ is not supported on ‘x86_64-darwin’
These packages also no longer show up in ‘nix-env -qa’ output. This
means, for instance, that the number of packages shown on
x86_64-freebsd has dropped from 9268 to 4764.
Since meta.platforms was also used to prevent Hydra from building some
packages, there now is a new attribute meta.hydraPlatforms listing the
platforms on which Hydra should build the package (which defaults to
meta.platforms).
Although this is a release canidate version of kernel 3.12, there are
reasons for merging this anyway, as discussed in #1010 and #1006.
Thanks to @offlinehacker for this and the initial pull request.
it helps, but is incomplete.
more fixes are coming, but including these would change too much
generic btrfs code, which might cause trouble for others.
so the best advice is not to use btrfs send yet and wait for 3.11 or 3.12
It has been submitted for inclusion in mainline, so it will probably
make it into 3.11 (or 3.12 as 3.11 is fairly close to release).
It is very local, only affecting people who use the "send" feature.
Without it, send is unstable/unsafe to use incrementally.
It can probably be applied to 3.9 and 3.8 as well, but as I only
tested it against 3.10, so I didn't bother.
It's bad to have the kernel config scattered across two places. (This
should also be done for the other architectures.)
Also, restore Xen and KVM guest support in Linux 3.10.
Having N different copies of the NixOS kernel configuration is bad
because these copies tend to diverge. For instance, our 3.10 config
lacked some modules that were enabled in older configs, probably
because the 3.10 config had been copied off an earlier version of some
older kernel config.
So now there is a single kernel config in common-config.nix. It has a
few conditionals to deal with new/removed kernel options, but
otherwise it's pretty straightforward.
Also, a lot of cut&paste boilerplate between the kernel Nix
expressions is gone (such as preConfigure).
9p (with caching enabled) is much faster than CIFS and doesn't require
Samba or virtual networking. For instance, building GNU Hello with
CIFS takes ~323s on my laptop, but with 9p it takes 54s.
More measurements will be needed to see if "cache=fscache" is really
faster than "cache=loose" (the former seems to be a little bit
faster).
Starting with 3.10, #! script handling can be built modularly (or not
at all). By default the nixpkgs builder sets everything modular, but
since our initird init is a #! script this creates a chicken-and-egg
problem on NixOS.
Signed-off-by: Shea Levy <shea@shealevy.com>
First, pass in `self' again so that overriding works properly (thanks
for pointing that out, @edolstra)
Second, instead of having linuxPackages*.kernel mean something different
inside the set and out, add a new attribute linuxPackages*.kernelDev,
which for the generic kernel is simply linuxPackages*.kernel but for the
manual-config kernel is the `dev' output (which has the build tree,
source tree, etc.)
The second change required trivial modifications in a bunch of
expressions, I verified that all of the linuxPackages* sets defined in
all-packages.nix have the same drv paths before and after the change.
Signed-off-by: Shea Levy <shea@shealevy.com>
This makes it possible to still build out-of-tree modules without making a system using this kernel depend on the full source and build tree at runtime.
Note that references to the source tree are removed from kernel modules after build.
Ideally, this would be accomplished by modifying the Makefile that puts the reference there in the first place, but I haven't tracked that down yet.
As timeconst.pl was replaced by a bc script in the merge window for kernel 3.9,
we also need bc during build time, for more information about the reasons,
please have a look at the following commit:
torvalds/linux@70730bca13
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Conflicts:
pkgs/applications/networking/browsers/chromium/default.nix
pkgs/top-level/all-packages.nix
Merge conflicts seemed trivial, but a look from viric and aszlig would be nice.
NFS_FSCACHE and CIFS_FSCACHE are required to enable the local cache support for
NFS and CIFS (the 'fsc' mount option). The reasoning for enabling it from 3.2
and forward is that at least Ubuntu 12.04 (linux 3.2) has it turned on. Could
probably be enabled for earlier kernels too.
Conflicts:
pkgs/development/compilers/gcc/4.6/default.nix
pkgs/development/compilers/gcc/4.7/default.nix
The 4.7 had some weird parameters added in crossAttrs; I've removed
them, but I don't understand where they come from.
This is for consistency with terminology in stdenv (and the terms
"hostDrv" and "buildDrv" are not very intuitive, even if they're
consistent with GNU terminology).
The previos version did a for loop over the output of set, which spits out _all_
defined variables and their contents. This not only is dangerous if there is a
variable starting with CONFIG_ but also can't handle whitespace, as the IFS is
set to any (horizontal _and_ vertical) whitespace by default.
So, imagine (actually don't imagine, something like this is the case in a lot of
kernel configuration files) you have the following variable:
CONFIG_ARCH_HWEIGHT_CFLAGS="-fcall-saved-rdi -fcall-saved-rsi ..."
A loop with for and the default IFS would result in the following variable
pieces:
0: CONFIG_ARCH_HWEIGHT_CFLAGS="-fcall-saved-rdi
1: -fcall-saved-rsi
2: ..."
This obviously leads to the problem that this config variable is being cut off
at the first whitespace.
Another downside of this approach is that set not only returns variables but
functions as well. This could lead to quite a lot of unexpected behaviour and
confusion.
So the new approach doesn't source the kernel configuration anymore but uses
`read` to parse the file line-by line, setting IFS to '=', thus splitting all
configuration lines into key/value pairs.
Using parameter expansion, we ensure that we only read lines starting with
"CONFIG_". This particularily has the advantage of not being bash-specific,
should we choose to change to a different default shell someday.
Now, after we got a correct "CONFIG_" line, we're using a temporary variable to
split off the first quote from the result. Particularily the reason behind this
is shell compatibility again, as ${${foo#"}%"} only works in Bash, Zsh and
whatnot but not in plain SH.
And within the next line we obviously insert the no_firstquote variable without
it's last quote removed.
But, what about escaping?
First of all, if we'd just eval the $val variable, we would correctly unescape
the value, but this has the downside that variables within the content would be
expanded, for example look at this:
CONFIG_DEFCONFIG_LIST="/lib/modules/$UNAME_RELEASE/.config"
Well, obviously this is a bad example at the Nix sense, but just to show that
variables within kernel configuration entries aren't impossible.
And second, which would have been a show stopper if \" would be within $val: It
simply would end up being an invalid Nix expression, because \" would end up as
a ploin " within a double quoted string.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
