[PATCH v2 0/2] sparc64: D-cache alias flushing fixes

[Imre Kaloz]

Two fixes to D-cache alias flushing on sparc64 CPUs with a virtually
indexed L1 D-cache.

Patch 1 makes move_pte() and tlb_batch_add() flush only the page a PTE
maps, instead of every page of its folio once per PTE, so the number
of cross-calls for an mremap() of a large folio no longer grows with
the square of its size.

Patch 2 implements flush_cache_vmap() and flush_cache_vunmap(), which
have been no-ops on sparc64.  Stale lines left by a vmalloc mapping
corrupt BPF programs and module data on UltraSPARC III.  It uses
flush_dcache_page_all() as patch 1 restores it.

Both patches are tagged for stable.  Tested on a Sun Ultra 45.

Changes in v2:
- Patch 2: flush_cache_vmap() also runs on the vmemmap, from
  sparse_init() before init_IRQ() sets up the cross-call mondo
  blocks; skip addresses outside vmalloc and module space, since the
  vmemmap has no aliases to flush.  Picked up Stian Halseth's
  Tested-by.

Imre Kaloz (2):
  sparc64: flush only the aliased page, not its whole folio
  sparc64: flush vmalloc ranges from the D-cache on map and unmap

 arch/sparc/include/asm/cacheflush_64.h |  9 ++--
 arch/sparc/include/asm/pgtable_64.h    |  7 ++-
 arch/sparc/kernel/smp_64.c             | 42 ++++++++++++-----
 arch/sparc/mm/init_64.c                | 65 ++++++++++++++++++++++++++
 arch/sparc/mm/tlb.c                    |  2 +-
 arch/sparc/mm/ultra.S                  | 25 ++++++++++
 6 files changed, 132 insertions(+), 18 deletions(-)


base-commit: fe2ec83746e501645709761605c2464a44fd2929
--
2.47.3