I have checked that there is no difference between 5.5 and 5.5.2 (except in the value of LINUX_VERSION_CODE in linux/version.h), and in general as I understand there should be no difference across all 5.5.* series, therefore we should not update the headers from 5.5 to 5.5.x.

This seems to break the bootstrap for aarch64:

$ nix-build -A hello -j1
these derivations will be built:
 [...]
building '/nix/store/d5fhzi0dvfdqcjbawv21bhs8pnifxpfl-popt-1.16.drv'...
unpacking sources
unpacking source archive /nix/store/vd3nbw7w1d7yrb89c3vj5kwk3yb8sgf1-popt-1.16.tar.gz
source root is popt-1.16
setting SOURCE_DATE_EPOCH to timestamp 1273006578 of file popt-1.16/test-poptrc
patching sources
configuring
fixing libtool script ./ltmain.sh
configure flags: --disable-static --disable-dependency-tracking --prefix=/nix/store/88izzq2bqs47q7a9q9q7ahl9yi7l6p7c-popt-1.16
checking build system type... ./config.guess: unable to guess system type

This script, last modified 2009-11-20, has failed to recognize
the operating system you are using. It is advised that you
download the most up to date version of the config scripts from

  http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.guess;hb=HEAD
and
  http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.sub;hb=HEAD

If the version you run (./config.guess) is already up to date, please
send the following data and any information you think might be
pertinent to <config-patches@gnu.org> in order to provide the needed
information to handle your system.

config.guess timestamp = 2009-11-20

uname -m = aarch64
uname -r = 4.19.68
uname -s = Linux
uname -v = #1-NixOS SMP PREEMPT Thu Jan 1 00:00:01 UTC 1970

/usr/bin/uname -p =
/bin/uname -X     =

hostinfo               =
/bin/universe          =
/usr/bin/arch -k       =
/bin/arch              =
/usr/bin/oslevel       =
/usr/convex/getsysinfo =

UNAME_MACHINE = aarch64
UNAME_RELEASE = 4.19.68
UNAME_SYSTEM  = Linux
UNAME_VERSION = #1-NixOS SMP PREEMPT Thu Jan 1 00:00:01 UTC 1970
configure: error: cannot guess build type; you must specify one
builder for '/nix/store/d5fhzi0dvfdqcjbawv21bhs8pnifxpfl-popt-1.16.drv' failed with exit code 1
cannot build derivation '/nix/store/awz6s24220mxm03dqq3p61hn05x51mib-rsync-3.1.3.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/jjnrjx8qazviqvlg33i5b68jd0lagkdr-linux-headers-5.5.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/rlkn045ii7sxp91ly4whpbq3ks410kad-glibc-2.27.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/zrxffnf7iydi0ahqxl91jj653p9cc9x7-stdenv-linux.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/wi7kwn2hkghzf8p5ygcx008gz5h3bbj9-hello-2.10.drv': 1 dependencies couldn't be built
error: build of '/nix/store/wi7kwn2hkghzf8p5ygcx008gz5h3bbj9-hello-2.10.drv' failed

I can not reproduce this with nix-build . -A pkgsCross.aarch64-multiplatform.hello and I don't have access to an aarch64 machine to perform a native build. Could you save the build directory after nix-build -A hello --keep-failed and attach it as a tar.gz to this issue? It may also be worth to check if this update is the cause by testing it on top of master (git checkout master; git cherry-pick -n d48e99d9122119ce1741867b5a7ea682d134f713; nix-build . -A hello).

I think the problem is that rsync is now being required by an earlier stage during stdenv bootstrap. linuxHeaders is produced at stage2, so stage2 must be able to build rsync and its popt dependency.

Starting from stage3 onwards there's a hook that updates outdated config.guess files, which matches the error message from popt's configuration failure. Taking this hook and trivially applying it to stage2 is sufficient for me to nix-build -A hello on aarch64-linux. I also tested armv7l-linux, it was able to produce a linux headers 5.5, but glibc is currently broken so I couldn't build a full stdenv.

diff --git a/pkgs/stdenv/linux/default.nix b/pkgs/stdenv/linux/default.nix
index 60c0730dce1..cf8705bd891 100644
--- a/pkgs/stdenv/linux/default.nix
+++ b/pkgs/stdenv/linux/default.nix
@@ -255,6 +255,10 @@ in
         libc = getLibc self;
       };
     };
+    extraNativeBuildInputs = [ prevStage.patchelf ] ++
+      # Many tarballs come with obsolete config.sub/config.guess that don't recognize aarch64.
+      lib.optional (!localSystem.isx86 || localSystem.libc == "musl")
+                   prevStage.updateAutotoolsGnuConfigScriptsHook;
   })

(diff generated after cherry-picking onto master @ 6932a75)

That looks like a useful change, but since it is trivial to avoid rsync and this reduces the amount of packages needed to reach stage3, I have done just that. (I have checked that the result does not change.)

This update caused a few failures (not in master yet; based on this eval diff, checkboxes for fixes):

• lockdep — error: unknown type name '__aligned_u64' — maybe it defines __ASSEMBLY__?
• dnsmasq — error: 'SIOCGSTAMP' undeclared — probably does not include linux/sockios.h
• can-utils — same as dnsmasq
• dpdk — error: 'struct xsk_umem_config' has no member named 'flags' — it assumes that if linux-headers are recent than libbpf is recent; we shall update libbpf
• iptraf — error: linux/isdn.h: No such file or directory — isdn.h is gone since Linux 5.3
• klibc — error: 'MAP_PRIVATE' undeclared — probably does not include linux/mman.h

(discovered later)

• lkl — missing defines again; this one is a channel-blocker

dnsmasq was broken on master and I fixed it in 482642a (and 944775e).

Ok, this has reached master yesterday (2020-02-19 08:18 UTC).

I thought that this is a good candidate for 20.03 since it defaults to Linux 5.4 and the update will allow applications to use newer kernel features; but its backport depends on backporting fixes for the packages that are broken by the update.

Thanks @orivej for checking all the thorny cross cases!

Why 5.5, and not 5.4 (LTS)?

See https://sourceware.org/glibc/wiki/FAQ#What_version_of_the_Linux_kernel_headers_should_be_used.3F