We rely on the linker to create arrays for a number of things including kernel parameters and device-tree-match entries. The stride of these linker-section arrays obviously needs to match the expectations of the code accessing them or bad things will happen. One thing to watch out for is that gcc is known to increase the alignment of larger objects with static extent as an optimisation (on x86), but this can be suppressed by using the aligned attribute when declaring entries. We've been relying on this behaviour for 16 years for kernel parameters (and other structures) and it indeed hasn't changed since the introduction of the aligned attribute in gcc 3.1 (see align_variable() in [1]). Occasionally this gcc optimisation do cause problems which have instead been worked around in various creative ways including using indirection through an array of pointers. This was originally done for tracepoints [2] after a number of failed attempts to create properly aligned arrays, and the approach was later reused for module-version attributes [3] and earlycon entries. This series reverts the latter two workarounds in favour of the one-line fix of aligning the entries according to the requirement of the type. In principle, there shouldn't be anything preventing us from doing the same for tracepoints. The key observation here is that the arrays should be constructed using the alignment of the type in question (as given by __alignof__()) rather than some specific alignment such as sizeof(void *). This allows the structures to be stored efficiently, but more importantly prevents breakage on architectures like m68k where pointers are 2-byte aligned should the size or alignment of the type change (e.g. so that the size is no longer divisible by four). As a preventive measure in case the kernel-parameter structures are ever amended (or the code pattern is reused elsewhere), the final patches switches the parameter declarations to also use type alignment. The series has been tested using gcc 4.9 and 9.3 on x86_32 and x86_64 and using gcc 7.2 on arm; and has been compile-tested and verified using gcc 4.9 and 10.1 on aarch64, sparc and m68k. Note that the patches are mostly independent and can be merged through different subsystem trees. I decided to post them as a series to provide a common background and have a single thread for any general discussion. Rob and Greg, can you take patches 1 and 2 through your trees, respectively? Jessica, you said you could take the module and params patches through your tree? Who picks up the init.h one? Linus? Johan [1] https://github.com/gcc-mirror/gcc/blob/master/gcc/varasm.c [2] https://lore.kernel.org/lkml/20110126222622.GA10794@Krystal/ [3] https://lore.kernel.org/lkml/1297123347-2170-1-git-send-email-dtor@xxxxxxxxxx/ Changes in v2 - amend commit messages of patches 1, 2, 4, 5 and 7 slightly in order to make it more clear that the gcc optimisation is suppressed by specifying alignment when declaring variables v1 - https://lore.kernel.org/r/20201103175711.10731-1-johan@xxxxxxxxxx Johan Hovold (8): of: fix linker-section match-table corruption earlycon: simplify earlycon-table implementation module: drop version-attribute alignment module: simplify version-attribute handling init: use type alignment for kernel parameters params: drop redundant "unused" attributes params: use type alignment for kernel parameters params: clean up module-param macros drivers/of/fdt.c | 7 ++----- drivers/tty/serial/earlycon.c | 6 ++---- include/linux/init.h | 2 +- include/linux/module.h | 28 ++++++++++++++-------------- include/linux/moduleparam.h | 12 ++++++------ include/linux/of.h | 1 + include/linux/serial_core.h | 20 +++++++------------- kernel/params.c | 10 ++++------ 8 files changed, 37 insertions(+), 49 deletions(-) -- 2.26.2
