| Message ID | 50056E59.4090003@linux.vnet.ibm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Jul 17, 2012 at 09:53:29PM +0800, Xiao Guangrong wrote: > If it have no indirect shadow pages we need not protect any gfn, > this is always true for direct mmu without nested > > Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Xiao, What is the motivation? Numbers please. In fact, what case was the original indirect_shadow_pages conditional in kvm_mmu_pte_write optimizing again? -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 07/20/2012 08:39 AM, Marcelo Tosatti wrote: > On Tue, Jul 17, 2012 at 09:53:29PM +0800, Xiao Guangrong wrote: >> If it have no indirect shadow pages we need not protect any gfn, >> this is always true for direct mmu without nested >> >> Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> > > Xiao, > > What is the motivation? Numbers please. > mmu_need_write_protect is the common path for both soft-mmu and hard-mmu, checking indirect_shadow_pages can skip hash-table walking for the case which is tdp is enabled without nested guest. I will post the Number after I do the performance test. > In fact, what case was the original indirect_shadow_pages conditional in > kvm_mmu_pte_write optimizing again? > They are the different paths, mmu_need_write_protect is the real page fault path, and kvm_mmu_pte_write is caused by mmio emulation. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
BTW, they are some bug fix patches on -master branch, but it is not existed on -next branch: commit: f411930442e01f9cf1bf4df41ff7e89476575c4d commit: 85b7059169e128c57a3a8a3e588fb89cb2031da1 It causes code conflict if we do the development on -next. On 07/20/2012 08:39 AM, Marcelo Tosatti wrote:can > On Tue, Jul 17, 2012 at 09:53:29PM +0800, Xiao Guangrong wrote: >> If it have no indirect shadow pages we need not protect any gfn, >> this is always true for direct mmu without nested >> >> Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> > > Xiao, > > What is the motivation? Numbers please. > > In fact, what case was the original indirect_shadow_pages conditional in > kvm_mmu_pte_write optimizing again? > > > -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Jul 20, 2012 at 10:34:28AM +0800, Xiao Guangrong wrote: > On 07/20/2012 08:39 AM, Marcelo Tosatti wrote: > > On Tue, Jul 17, 2012 at 09:53:29PM +0800, Xiao Guangrong wrote: > >> If it have no indirect shadow pages we need not protect any gfn, > >> this is always true for direct mmu without nested > >> > >> Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> > > > > Xiao, > > > > What is the motivation? Numbers please. > > > > mmu_need_write_protect is the common path for both soft-mmu and > hard-mmu, checking indirect_shadow_pages can skip hash-table walking > for the case which is tdp is enabled without nested guest. I mean motivation as observation that it is a bottleneck. > I will post the Number after I do the performance test. > > > In fact, what case was the original indirect_shadow_pages conditional in > > kvm_mmu_pte_write optimizing again? > > > > They are the different paths, mmu_need_write_protect is the real > page fault path, and kvm_mmu_pte_write is caused by mmio emulation. Sure. What i am asking is, what use case is the indirect_shadow_pages optimizing? What scenario, what workload? See the "When to optimize" section of http://en.wikipedia.org/wiki/Program_optimization. Can't remember why indirect_shadow_pages was introduced in kvm_mmu_pte_write. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Jul 20, 2012 at 11:45:59AM +0800, Xiao Guangrong wrote: > BTW, they are some bug fix patches on -master branch, but > it is not existed on -next branch: > commit: f411930442e01f9cf1bf4df41ff7e89476575c4d > commit: 85b7059169e128c57a3a8a3e588fb89cb2031da1 > > It causes code conflict if we do the development on -next. See auto-next branch. http://www.linux-kvm.org/page/Kvm-Git-Workflow -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 07/20/2012 07:09 PM, Marcelo Tosatti wrote: > On Fri, Jul 20, 2012 at 10:34:28AM +0800, Xiao Guangrong wrote: >> On 07/20/2012 08:39 AM, Marcelo Tosatti wrote: >>> On Tue, Jul 17, 2012 at 09:53:29PM +0800, Xiao Guangrong wrote: >>>> If it have no indirect shadow pages we need not protect any gfn, >>>> this is always true for direct mmu without nested >>>> >>>> Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> >>> >>> Xiao, >>> >>> What is the motivation? Numbers please. >>> >> >> mmu_need_write_protect is the common path for both soft-mmu and >> hard-mmu, checking indirect_shadow_pages can skip hash-table walking >> for the case which is tdp is enabled without nested guest. > > I mean motivation as observation that it is a bottleneck. > >> I will post the Number after I do the performance test. >> >>> In fact, what case was the original indirect_shadow_pages conditional in >>> kvm_mmu_pte_write optimizing again? >>> >> >> They are the different paths, mmu_need_write_protect is the real >> page fault path, and kvm_mmu_pte_write is caused by mmio emulation. > > Sure. What i am asking is, what use case is the indirect_shadow_pages > optimizing? What scenario, what workload? > Sorry, Marcelo, i do know why i completely misunderstood your mail. :( I am not sure whether this is a bottleneck, i just got it from code review, i will measure it to see if we can get benefit from it. :p > See the "When to optimize" section of > http://en.wikipedia.org/wiki/Program_optimization. > > Can't remember why indirect_shadow_pages was introduced in > kvm_mmu_pte_write. > Please refer to: https://lkml.org/lkml/2011/5/18/174 -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c index 28b12e2..a846a9c 100644 --- a/arch/x86/kvm/mmu.c +++ b/arch/x86/kvm/mmu.c @@ -2294,6 +2294,9 @@ static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn, struct hlist_node *node; bool need_unsync = false; + if (!vcpu->kvm->arch.indirect_shadow_pages) + return 0; + for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn, node) { if (!can_unsync) return 1;
If it have no indirect shadow pages we need not protect any gfn, this is always true for direct mmu without nested Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> --- arch/x86/kvm/mmu.c | 3 +++ 1 files changed, 3 insertions(+), 0 deletions(-)