Since the RFC I accounted for feedback, and also made the patch more
complete so we can actually boot and run a 64s/radix kernel without
using tlbie at all.
KVM and bare metal hash are harder to support. Bare metal hash because
it does some TLB invation with interrupts disabled (can't use IPIs).
It might be possible those invalidates could be avoided or the paths
changed to enable interrupts, but it would be a much bigger change.
KVM with radix guests might be acutally quite simple to support, so I
can look at that next if we get this merged.
KVM with hash guests might not be feasible to do in a performant way,
because of the design of virtualised hash architecture (host flushes
guest effective addresses), at least it would need special real mode
IPI handlers for TLB flushes.
Thanks,
Nick
Nicholas Piggin (3):
powerpc/64s: Remove mmu_partition_table_set
powerpc/64s/radix: all CPUs should flush local translation structure
before turning MMU on
powerpc/64s: introduce options to disable use of the tlbie instruction
.../admin-guide/kernel-parameters.txt | 4 +
arch/powerpc/include/asm/book3s/64/tlbflush.h | 9 +
arch/powerpc/include/asm/mmu.h | 2 -
arch/powerpc/kvm/book3s_hv.c | 10 +-
arch/powerpc/kvm/book3s_hv_nested.c | 35 +++-
arch/powerpc/mm/book3s64/hash_utils.c | 4 +-
arch/powerpc/mm/book3s64/mmu_context.c | 4 +-
arch/powerpc/mm/book3s64/pgtable.c | 77 +++++---
arch/powerpc/mm/book3s64/radix_pgtable.c | 22 +--
arch/powerpc/mm/book3s64/radix_tlb.c | 182 ++++++++++++++++--
arch/powerpc/mm/pgtable_64.c | 5 +-
drivers/misc/cxl/main.c | 4 +
drivers/misc/ocxl/main.c | 4 +
13 files changed, 282 insertions(+), 80 deletions(-)
--
2.22.0
