generate-config.pl's auto modules feature answers "m" to any Kconfig
question it thinks supports being a module. It detected this by
seeing if the help shown by make config (called "alts" by the script)
contained the string "/m", which it would in the case of e.g a
tristate option, where alts would be "N/m/y/?".
But then along came CONFIG_MODPROBE_PATH in Linux 5.13, with a default
value, shown in the make config help, of "/sbin/modprobe".
generate-config.pl would see the "/m" substring, and answer "m" to the
question, meaning (I think) that the built kernel would expect the
modprobe binary to be at /m. This broke the (non-NixOS) VM images I
build with Nix. NixOS was unaffected because it uses a different
mechanism to set the modprobe path.
With the current architecture, we can't 100% determine whether a
Kconfig option is a string or a tristate, but we can get a lot closer
by using a better regex. My new regex only accepts single word
characters, separated by slashes, with a "/?" at the end. This is
much less likely to ever end up as the default value of a string
option.
Tested by building linux_latest.configfile before and after my
changes, and checking the only difference is the correct default for
CONFIG_MODPROBE_PATH.
This enforces that the configuration generated will obey any/all flags
set in the platform/stdenv configuration. This is crucial, for example,
if you'd like to build a kernel using clang.
Without this patch, anything you set in
`stdenv.hostPlatform.linux-kernel.makeFlags` is wholly ignored during
config generation, causing (for example) any changes in the desired
toolchain (e.g. `LLVM`, `LLVM_IAS`) to not be reflected in the generated
config, and for the subsequent build to fail.
Prior to this change, the script could potentially get confused by
"menus" containing only one valid option. Thus, with
CPU_BIG_ENDIAN = no;
CPU_LITTLE_ENDIAN = yes;
ARM64_PA_BITS_48 = yes;
the endianness would be set incorrectly:
GOT: Physical address space size
GOT: > 1. 48-bit (ARM64_PA_BITS_48)
GOT: choice[1]: 1
GOT: Endianness
GOT: 1. Build big-endian kernel (CPU_BIG_ENDIAN)
GOT: > 2. Build little-endian kernel (CPU_LITTLE_ENDIAN)
CHOICE: 1-2?, ANSWER: 1
This commit fixes this error by forgetting previous menu choices if a
line that follows a menu option is neither another menu option nor the
prompt for choosing one of the options.
This makes it less frustrating to add a large group of kernel options
via boot.kernelPatches.*.extraConfig.
Signed-off-by: Anders Kaseorg <andersk@mit.edu>
In particular, this allows setting CONFIG_CMDLINE to a value with
multiple space-seperated parameters. Other variables also take free-form
strings and might similarly benefit.
- defined buildLinux as generic.nix instead of manual-config.nix. This
makes kernel derivations a tad more similar to your typical derivations.
- moved $buildRoot to within the source folder, this way it doesn't have to be created before the unpackPhase
and make it easier to work on kernel source without running the unpackPhase
This allows to use kernelAutoModules but still compile in any options that are set so in template config.
It's helpful for ARM and maybe other platforms where defaul configurations are useful because they compile in
modules that we and udev cannot autodetect now.
For instance, the current 3.10 kernel build fails at the end with:
unused option: BRCMFMAC_PCIE
unused option: FW_LOADER_USER_HELPER_FALLBACK
unused option: KEXEC_FILE
unused option: RANDOMIZE_BASE
However, it's not obvious that only the _last_ one is actually fatal to
the build. After this change it's at least somewhat better:
warning: unused option: BRCMFMAC_PCIE
warning: unused option: FW_LOADER_USER_HELPER_FALLBACK
warning: unused option: KEXEC_FILE
error: unused option: RANDOMIZE_BASE
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>
"make config" to print an additional question that the script
doesn't know how to answer ("Audio Excel DSP 16 [N/y] (NEW) ###").
Just ignore it.
svn path=/nixpkgs/branches/stdenv-updates/; revision=19713
has to be "/bin/sh" - other values such as "sh" or "bash" or "$(type
-tP sh)" cause a build error:
LD .tmp_vmlinux1
/nix/store/1yv8i1m76cvwk5w5i5wrk4gj5zyfj6vh-binutils-2.19.1/bin/ld:arch/um/kernel/vmlinux.lds:1: ignoring invalid character `#' in expression
/nix/store/1yv8i1m76cvwk5w5i5wrk4gj5zyfj6vh-binutils-2.19.1/bin/ld:arch/um/kernel/vmlinux.lds:1: syntax error
collect2: ld returned 1 exit status
This is caused by Bash 4.0
(http://bugzilla.kernel.org/show_bug.cgi?id=13343).
svn path=/nixpkgs/branches/kernel-config/; revision=18941
`generate-config.pl' runs `make config' to generate a Linux kernel
configuration file. For each question (i.e. kernel configuration
option), unless an override is provided, it answers "m" if possible,
and otherwise uses the default answer (as determined by the default
config for the architecture). This is safer than allmodconfig,
which answers "y" everywhere it can't answer "m" and thus ends up
enabling a lot of experimental or debug options. (For this reason,
a configuration generated by allmodconfig must be carefully checked
with every new release to ensure that nothing dangerous is enabled.
The default config should be safer wrt new kernel releases.)
Overrides are specified in the `config' argument to generic.nix,
which is a string that contains lines such as `EXT2_FS_POSIX_ACL y'.
The script warns about ignored options, and aborts if `make config'
selects an answer inconsistent with the one in `config'. This
allows us to be sure that `make config' doesn't silently override
our configuration values (e.g., depending on other options, it will
set FB_TILEBLITTING to "y" even if we want it to be "n").
svn path=/nixpkgs/branches/kernel-config/; revision=18910
