Re: [PATCH RFC 0/4] KVM: MIPS: Provide arch-specific kvm_flush_remote_tlbs()

[Paolo Bonzini <pbonzini@redhat.com>]

On 07/02/20 23:35, Peter Xu wrote:
> [This series is RFC because I don't have MIPS to compile and test]
> 
> kvm_flush_remote_tlbs() can be arch-specific, by either:
> 
> - Completely replace kvm_flush_remote_tlbs(), like ARM, who is the
>   only user of CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL so far
> 
> - Doing something extra before kvm_flush_remote_tlbs(), like MIPS VZ
>   support, however still wants to have the common tlb flush to be part
>   of the process.  Could refer to kvm_vz_flush_shadow_all().  Then in
>   MIPS it's awkward to flush remote TLBs: we'll need to call the mips
>   hooks.
> 
> It's awkward to have different ways to specialize this procedure,
> especially MIPS cannot use the genenal interface which is quite a
> pity.  It's good to make it a common interface.
> 
> This patch series removes the 2nd MIPS usage above, and let it also
> use the common kvm_flush_remote_tlbs() interface.  It should be
> suggested that we always keep kvm_flush_remote_tlbs() be a common
> entrance for tlb flushing on all archs.
> 
> This idea comes from the reading of Sean's patchset on dynamic memslot
> allocation, where a new dirty log specific hook is added for flushing
> TLBs only for the MIPS code [1].  With this patchset, logically the
> new hook in that patch can be dropped so we can directly use
> kvm_flush_remote_tlbs().
> 
> TODO: We can even extend another common interface for ranged TLB, but
> let's see how we think about this series first.
> 
> Any comment is welcomed, thanks.
> 
> Peter Xu (4):
>   KVM: Provide kvm_flush_remote_tlbs_common()
>   KVM: MIPS: Drop flush_shadow_memslot() callback
>   KVM: MIPS: Replace all the kvm_flush_remote_tlbs() references
>   KVM: MIPS: Define arch-specific kvm_flush_remote_tlbs()
> 
>  arch/mips/include/asm/kvm_host.h |  7 -------
>  arch/mips/kvm/Kconfig            |  1 +
>  arch/mips/kvm/mips.c             | 22 ++++++++++------------
>  arch/mips/kvm/trap_emul.c        | 15 +--------------
>  arch/mips/kvm/vz.c               | 14 ++------------
>  include/linux/kvm_host.h         |  1 +
>  virt/kvm/kvm_main.c              | 10 ++++++++--
>  7 files changed, 23 insertions(+), 47 deletions(-)
> 

Compile-tested and queued.

MIPS folks, I see that arch/mips/kvm/mmu.c uses pud_index, so it's not
clear to me if it's meant to only work if CONFIG_PGTABLE_LEVELS=4 or
it's just bit rot.  Should I add a "depends on PGTABLE_LEVEL=4" to
arch/mips/Kconfig?

Paolo