I've recently started getting an internal compiler error when building
an aarch64 kernel. It only happens once in a while, and re-running the
task usually just succeeds, so I don't know how to reproduce or trigger
this at will.

Two examples:

===
CC [M] drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm20b.o
CC [M] drivers/net/ethernet/mellanox/mlx5/core/rl.o
CC [M] drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp100.o
*** stack smashing detected ***: <unknown> terminated
In file included from .../kernel-source/arch/arm64/include/asm/atomic.h:15,
from .../kernel-source/include/linux/atomic.h:7,
from
.../kernel-source/include/asm-generic/bitops/atomic.h:5,
from .../kernel-source/arch/arm64/include/asm/bitops.h:26,
from .../kernel-source/include/linux/bitops.h:29,
from .../kernel-source/include/linux/kernel.h:12,
from .../kernel-source/include/linux/uio.h:8,
from .../kernel-source/include/linux/socket.h:8,
from .../kernel-source/include/uapi/linux/if.h:25,
from .../kernel-source/net/mac80211/led.c:7:
.../kernel-source/include/net/inet_sock.h: In function 'inet_sk_state_load':
.../kernel-source/arch/arm64/include/asm/barrier.h:114:8: internal
compiler error: Aborted
114 | union { __unqual_scalar_typeof(*p) __val; char __c[1]; } __u; \
| ^
.../kernel-source/include/asm-generic/barrier.h:142:29: note: in
expansion of macro '__smp_load_acquire'
142 | #define smp_load_acquire(p) __smp_load_acquire(p)
| ^~~~~~~~~~~~~~~~~~
.../kernel-source/include/net/inet_sock.h:312:9: note: in expansion of
macro 'smp_load_acquire'
312 | return smp_load_acquire(&sk->sk_state);
| ^~~~~~~~~~~~~~~~
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://gcc.gnu.org/bugs/> for instructions.
.../kernel-source/scripts/Makefile.build:279: recipe for target
'net/mac80211/led.o' failed
make[3]: *** [net/mac80211/led.o] Error 1
make[3]: *** Waiting for unfinished jobs....
CC [M] drivers/net/ethernet/mellanox/mlx5/core/lag.o
===
CC [M] drivers/gpu/drm/nouveau/nvkm/nvfw/ls.o
CC [M] drivers/gpu/drm/drm_modeset_helper.o
CC [M] drivers/gpu/drm/drm_scdc_helper.o
*** stack smashing detected ***: <unknown> terminated
In file included from
.../kernel-source/include/linux/regulator/consumer.h:35,
from
.../kernel-source/drivers/gpu/drm/nouveau/include/nvif/os.h:27,
from
.../kernel-source/drivers/gpu/drm/nouveau/include/nvkm/core/os.h:4,
from
.../kernel-source/drivers/gpu/drm/nouveau/include/nvkm/core/oclass.h:3,
from
.../kernel-source/drivers/gpu/drm/nouveau/include/nvkm/core/device.h:4,
from
.../kernel-source/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h:4,
from
.../kernel-source/drivers/gpu/drm/nouveau/nvkm/nvfw/ls.c:22:
.../kernel-source/include/linux/suspend.h:364:36: internal compiler
error: Aborted
364 | extern void mark_free_pages(struct zone *zone);
| ^~~~
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://gcc.gnu.org/bugs/> for instructions.
.../kernel-source/scripts/Makefile.build:280: recipe for target
'drivers/gpu/drm/nouveau/nvkm/nvfw/ls.o' failed
make[5]: *** [drivers/gpu/drm/nouveau/nvkm/nvfw/ls.o] Error 1
make[5]: *** Waiting for unfinished jobs....
CC [M] drivers/gpu/drm/drm_gem_framebuffer_helper.o
===

This is with hardknott, aarch64-oe-linux-gcc (GCC) 10.2.0, building
5.10.* kernels (5.10.45 and 5.10.65 in the cases above IIRC). The build
is visiting drivers/gpu/drm/ in both cases, but in the former case it's
not actually a TU in there that fails, but one in net/, so I'm not even
sure it it has to do with something peculiar to the drivers/gpu/drm/
modules.

Has anyone seem something like this, or any ideas for figuring out
what's going on?

Rasmus