projects / platform / kernel / linux-rpi.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 0cac21b) | patch
RISC-V: Upgrade smp_mb__after_spinlock() to iorw,iorw
authorPalmer Dabbelt <palmerdabbelt@google.com>
Thu, 16 Jul 2020 18:57:26 +0000 (11:57 -0700)
committerPalmer Dabbelt <palmerdabbelt@google.com>
Fri, 17 Jul 2020 16:28:35 +0000 (09:28 -0700)
commit38b7c2a3ffb1fce8358ddc6006cfe5c038ff9963
tree95be1d51d69478dbf93f5bb8c2a4e6c7729d21f9tree | snapshot
parent0cac21b02ba5f3095fd2dcc77c26a25a0b2432edcommit | diff
RISC-V: Upgrade smp_mb__after_spinlock() to iorw,iorw

While digging through the recent mmiowb preemption issue it came up that
we aren't actually preventing IO from crossing a scheduling boundary.
While it's a bit ugly to overload smp_mb__after_spinlock() with this
behavior, it's what PowerPC is doing so there's some precedent.

Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
arch/riscv/include/asm/barrier.h diff | blob | history
Domain: System / Kernel;