diff mbox

[RESEND,v15,07/10] KVM: arm: page logging 2nd stage fault handling

Message ID 20150114103233.GI26222@cbox
State New
Headers show

Commit Message

Christoffer Dall Jan. 14, 2015, 10:32 a.m. UTC
On Tue, Jan 13, 2015 at 03:14:47PM -0800, Mario Smarduch wrote:

[...]

> >>>
> >>>
> >>> What I meant last time around concerning user_mem_abort was more
> >>> something like this:
> >>>
> >>> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> >>> index 1dc9778..38ea58e 100644
> >>> --- a/arch/arm/kvm/mmu.c
> >>> +++ b/arch/arm/kvm/mmu.c
> >>> @@ -935,7 +935,14 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
> >>>  		return -EFAULT;
> >>>  	}
> >>>  
> >>> -	if (is_vm_hugetlb_page(vma)) {
> >>> +	/*
> >>> +	 * Writes to pages in a memslot with logging enabled are always logged
> >>> +	 * on a singe page-by-page basis.
> >>> +	 */
> >>> +	if (memslot_is_logging(memslot) && write_fault)
> >>> +		force_pte = true;
> >>
> >> If it's a write you take the pte route and
> >> dissolves huge page, if it's a read you reconstruct the
> >> THP that seems to yield pretty bad results.
> > 
> > ok, then remove the ' && write_fault' part of the clause.
> Hi Christoffer,
>  couple comments/questions.
> 
>  setting force_pte here, disables huge pages for
> non-writable regions.
> 

hmmm, by a non-writable region you mean a read-only memslot? Can you set
dirty page logging for such one?  That doesn't make much sense to me.

Note, that if you receive writable == false from gfn_to_pfn_prot() that
doesn't mean that the page can never be written to, it just means that
the current mapping of the page is not a writable one, you can call that
same function again later with write_fault=true, and you either get a
writable page back or you simply get an error.

[...]

> >>>  	if (kvm_is_device_pfn(pfn))
> >>>  		mem_type = PAGE_S2_DEVICE;
> 
> If we're not setting the IOMAP flag do we have need
> this, since we're forfeiting error checking later
> in stage2_set_pte()?
> 

we still need this, remember the error checking is about
cache == NULL, not about the IOMAP flag.  I think I address this in the
new proposal below, but please check carefully.

Take a look at this one:


Thanks,
-Christoffer

Comments

Christoffer Dall Jan. 15, 2015, 10:20 a.m. UTC | #1
On Wed, Jan 14, 2015 at 03:10:11PM -0800, Mario Smarduch wrote:
> On 01/14/2015 02:32 AM, Christoffer Dall wrote:
> > On Tue, Jan 13, 2015 at 03:14:47PM -0800, Mario Smarduch wrote:
> > 
> > [...]
> > 
> >>>>>
> >>>>>
> >>>>> What I meant last time around concerning user_mem_abort was more
> >>>>> something like this:
> >>>>>
> >>>>> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> >>>>> index 1dc9778..38ea58e 100644
> >>>>> --- a/arch/arm/kvm/mmu.c
> >>>>> +++ b/arch/arm/kvm/mmu.c
> >>>>> @@ -935,7 +935,14 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
> >>>>>  		return -EFAULT;
> >>>>>  	}
> >>>>>  
> >>>>> -	if (is_vm_hugetlb_page(vma)) {
> >>>>> +	/*
> >>>>> +	 * Writes to pages in a memslot with logging enabled are always logged
> >>>>> +	 * on a singe page-by-page basis.
> >>>>> +	 */
> >>>>> +	if (memslot_is_logging(memslot) && write_fault)
> >>>>> +		force_pte = true;
> >>>>
> >>>> If it's a write you take the pte route and
> >>>> dissolves huge page, if it's a read you reconstruct the
> >>>> THP that seems to yield pretty bad results.
> >>>
> >>> ok, then remove the ' && write_fault' part of the clause.
> >> Hi Christoffer,
> >>  couple comments/questions.
> >>
> >>  setting force_pte here, disables huge pages for
> >> non-writable regions.
> >>
> > 
> 
> Hi Christoffer,
>  another round, although I'll go ahead and post another
> iteration, sorry but as you mentioned this code is
> important.
> 
> > hmmm, by a non-writable region you mean a read-only memslot? Can you set
> > dirty page logging for such one?  That doesn't make much sense to me.
> 
> Come to think of it that's  true.
> 
> It's bit fuzzyy when I was looking at the API for KVM_MEM_LOG_DIRTY_PAGES,
> it appears user space needs to check if region is read-only and set region
> size to 0(qemu). I don't see any checks in kernel to disable logging if
> region is read only and we're enabling dirty page logging. API doesn't say
> anything else. You may be able to enable logging
> for read-only region if you leave region size as is.
> 
> I guess this has been around for quite a while so we
> can just assume read-only slots will have logging disabled.
> 

I'm a bit confused, IIUC we don't make any explicit checks in the code
right now, so either there is some generic code that never sets the
logging flag on a read-only memregion or you can change the
implementation of memslot_is_logging() to return false if the memslot is
read-only.

> > 
> > Note, that if you receive writable == false from gfn_to_pfn_prot() that
> > doesn't mean that the page can never be written to, it just means that
> > the current mapping of the page is not a writable one, you can call that
> > same function again later with write_fault=true, and you either get a
> > writable page back or you simply get an error.
> 
> Yes that's true after studying hva_to_pfn_slow(),
> and __get_user_pages_fast(), a lot of conditions
> handled there.
> 
> > 
> > [...]
> > 
> >>>>>  	if (kvm_is_device_pfn(pfn))
> >>>>>  		mem_type = PAGE_S2_DEVICE;
> >>
> >> If we're not setting the IOMAP flag do we have need
> >> this, since we're forfeiting error checking later
> >> in stage2_set_pte()?
> >>
> > 
> > we still need this, remember the error checking is about
> > cache == NULL, not about the IOMAP flag.  I think I address this in the
> > new proposal below, but please check carefully.
> 
> Ok so mmu notifier may call stage2_set_pte() with
> null cache poiner and intermediate table entries may
> not be there so stage2_get_pud() may return NULL.
> With logging on it won't happen, but just in case
> we check.

yes, to easily catch programming errors in the future, that's why if you
make it a VM_BUG_ON the check won't be compiled unless you have kernel
memory debugging enabled.

> 
> And we'll continue to handle Device faults until
> further notice.
> 

yes.

> > 
> > Take a look at this one:
> 
> Looks good to me, concise.
> 

Thanks,
-Christoffer
diff mbox

Patch

diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
index 1dc9778..841e053 100644
--- a/arch/arm/kvm/mmu.c
+++ b/arch/arm/kvm/mmu.c
@@ -919,6 +919,7 @@  static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 	pfn_t pfn;
 	pgprot_t mem_type = PAGE_S2;
 	bool fault_ipa_uncached;
+	unsigned long flags = 0;
 
 	write_fault = kvm_is_write_fault(vcpu);
 	if (fault_status == FSC_PERM && !write_fault) {
@@ -976,8 +977,26 @@  static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 	if (is_error_pfn(pfn))
 		return -EFAULT;
 
-	if (kvm_is_device_pfn(pfn))
+	if (kvm_is_device_pfn(pfn)) {
 		mem_type = PAGE_S2_DEVICE;
+		flags |= KVM_S2PTE_FLAG_IS_IOMAP;
+	} else if (memslot_is_logging(memslot)) {
+		/*
+		 * Faults on pages in a memslot with logging enabled that can
+		 * should not be mapped with huge pages (it introduces churn
+		 * and performance degradation), so force a pte mapping.
+		 */
+		force_pte = true;
+		flags |= KVM_S2_FLAG_LOGGING_ACTIVE;
+
+		/*
+		 * Only actually map the page as writable if this was a write
+		 * fault.
+		 */
+		if (!write_fault)
+			writable = false;
+
+	}
 
 	spin_lock(&kvm->mmu_lock);
 	if (mmu_notifier_retry(kvm, mmu_seq))
@@ -1002,13 +1021,13 @@  static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
 		if (writable) {
 			kvm_set_s2pte_writable(&new_pte);
 			kvm_set_pfn_dirty(pfn);
+			mark_page_dirty(kvm, gfn);
 		}
 		coherent_cache_guest_page(vcpu, hva, PAGE_SIZE,
 					  fault_ipa_uncached);
-		ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte,
-			pgprot_val(mem_type) == pgprot_val(PAGE_S2_DEVICE));
-	}
 
+		ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags);
+	}
 
 out_unlock:
 	spin_unlock(&kvm->mmu_lock);