cache-v7: Disable preemption when reading CCSIDR
authorStephen Boyd <(address hidden)>
Fri, 10 Feb 2012 11:19:11 +0000 (16:19 +0530)
committerSimone Willett <swillett@nvidia.com>
Wed, 15 Feb 2012 20:44:49 +0000 (12:44 -0800)
commit6dfe0d880acf300419da399d4973fe38cdd42318
tree2a27b7e199700c64566fbac507eb0edbf2df8bc7
parenta138e0615700e7d31a6595aa01f5b3fe1a620d7c
cache-v7: Disable preemption when reading CCSIDR

armv7's flush_cache_all() flushes caches via set/way. To
determine the cache attributes (line size, number of sets,
etc.) the assembly first writes the CSSELR register to select a
cache level and then reads the CCSIDR register. The CSSELR register
is banked per-cpu and is used to determine which cache level CCSIDR
reads. If the task is migrated between when the CSSELR is written and
the CCSIDR is read the CCSIDR value may be for an unexpected cache
level (for example L1 instead of L2) and incorrect cache flushing
could occur.

Disable interrupts across the write and read so that the correct
cache attributes are read and used for the cache flushing
routine. We disable interrupts instead of disabling preemption
because the critical section is only 3 instructions and we want
to call v7_dcache_flush_all from __v7_setup which doesn't have a
full kernel stack with a struct thread_info.

This fixes a problem we see in scm_call() when flush_cache_all()
is called from preemptible context and sometimes the L2 cache is
not properly flushed out.

Signed-off-by: Stephen Boyd <(address hidden)>
Acked-by: Catalin Marinas <(address hidden)>
Reviewed-by: Nicolas Pitre <(address hidden)>
Cc: stable@vger.kernel.org
Change-Id: I34a54ac396929d9e4f9abb43fbeaeb71d5514b63
Reviewed-on: http://git-master/r/83094
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Scott Williams <scwilliams@nvidia.com>
Reviewed-by: Aleksandr Frid <afrid@nvidia.com>
Reviewed-by: Prashant Gaikwad <pgaikwad@nvidia.com>
Tested-by: Prashant Gaikwad <pgaikwad@nvidia.com>
arch/arm/mm/cache-v7.S