lkml.org 
[lkml]   [2010]   [Apr]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 6/6] KVM MMU: optimize synchronization shadow pages
On 04/12/2010 11:06 AM, Xiao Guangrong wrote:
> - chain all unsync shadow pages then we can fetch them quickly
> - flush local/remote tlb after all shadow page synced
>
> Signed-off-by: Xiao Guangrong<xiaoguangrong@cn.fujitsu.com>
> ---
> arch/x86/include/asm/kvm_host.h | 1 +
> arch/x86/kvm/mmu.c | 82 ++++++++++++++++++---------------------
> 2 files changed, 39 insertions(+), 44 deletions(-)
>
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index d463bc6..ae543fb 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -207,6 +207,7 @@ struct kvm_mmu_page {
> #define MMU_PAGE_UNSYNC 0x2
> unsigned int flags;
> unsigned int unsync_children;
> + struct list_head unsync_link;
> union {
> u64 *parent_pte; /* !multimapped */
> struct hlist_head parent_ptes; /* multimapped, kvm_pte_chain */
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index 18eceb2..fcb6299 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -177,6 +177,8 @@ typedef int (*mmu_parent_walk_fn) (struct kvm_mmu_page *sp, u64 *spte);
> static struct kmem_cache *pte_chain_cache;
> static struct kmem_cache *rmap_desc_cache;
> static struct kmem_cache *mmu_page_header_cache;
> +static struct list_head unsync_mmu_page_list =
> + LIST_HEAD_INIT(unsync_mmu_page_list);
>
>

Does this really need to be global? Should be per guest.

> static u64 __read_mostly shadow_trap_nonpresent_pte;
> static u64 __read_mostly shadow_notrap_nonpresent_pte;
> @@ -950,6 +952,7 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
> bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
> sp->flags = 0;
> sp->parent_pte = parent_pte;
> + INIT_LIST_HEAD(&sp->unsync_link);
> --vcpu->kvm->arch.n_free_mmu_pages;
> return sp;
> }
> @@ -1200,12 +1203,14 @@ static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp)
> {
> WARN_ON(!mmu_page_is_unsync(sp));
> mmu_page_clear_unsync(sp);
> + list_del(&sp->unsync_link);
> --kvm->stat.mmu_unsync;
> }
>

Locking?

> - if (protected)
> - kvm_flush_remote_tlbs(vcpu->kvm);
> + if (list_empty(&unsync_mmu_page_list))
> + return;
>

Will this ever happen? In my experience we usually have a ton of
unsyncs lying around.

> + list_for_each_safe(p, next,&unsync_mmu_page_list) {
> + struct kvm_mmu_page *sp;
> + sp = list_entry(p, struct kvm_mmu_page, unsync_link);
> + __kvm_sync_page(vcpu, sp,&flush_local_tlb,&flush_remote_tlb);
> }
>

That's counter to the point of unsync. We only want to sync the minimum
number of pages - all pages reachable by the new root. The others we
want to keep writeable to reduce our fault count.

--
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.



\
 
 \ /
  Last update: 2010-04-12 10:45    [W:0.082 / U:0.512 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site