Re: [PATCH v2 2/3] KVM: x86: introduce KVM_MEM_PCI_HOLE memory

[Sean Christopherson]

On Fri, Aug 07, 2020 at 04:12:31PM +0200, Vitaly Kuznetsov wrote:
> PCIe config space can (depending on the configuration) be quite big but
> usually is sparsely populated. Guest may scan it by accessing individual
> device's page which, when device is missing, is supposed to have 'pci
> hole' semantics: reads return '0xff' and writes get discarded. Compared
> to the already existing KVM_MEM_READONLY, VMM doesn't need to allocate
> real memory and stuff it with '0xff'.
> 
> Suggested-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
> ---
>  Documentation/virt/kvm/api.rst  | 18 ++++++++++-----
>  arch/x86/include/uapi/asm/kvm.h |  1 +
>  arch/x86/kvm/mmu/mmu.c          |  5 ++++-
>  arch/x86/kvm/mmu/paging_tmpl.h  |  3 +++
>  arch/x86/kvm/x86.c              | 10 ++++++---
>  include/linux/kvm_host.h        |  3 +++
>  include/uapi/linux/kvm.h        |  2 ++
>  virt/kvm/kvm_main.c             | 39 +++++++++++++++++++++++++++------
>  8 files changed, 64 insertions(+), 17 deletions(-)
> 
> diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
> index 644e5326aa50..dc4172352635 100644
> --- a/Documentation/virt/kvm/api.rst
> +++ b/Documentation/virt/kvm/api.rst
> @@ -1241,6 +1241,7 @@ yet and must be cleared on entry.
>    /* for kvm_memory_region::flags */
>    #define KVM_MEM_LOG_DIRTY_PAGES	(1UL << 0)
>    #define KVM_MEM_READONLY	(1UL << 1)
> +  #define KVM_MEM_PCI_HOLE		(1UL << 2)
>  
>  This ioctl allows the user to create, modify or delete a guest physical
>  memory slot.  Bits 0-15 of "slot" specify the slot id and this value
> @@ -1268,12 +1269,17 @@ It is recommended that the lower 21 bits of guest_phys_addr and userspace_addr
>  be identical.  This allows large pages in the guest to be backed by large
>  pages in the host.
>  
> -The flags field supports two flags: KVM_MEM_LOG_DIRTY_PAGES and
> -KVM_MEM_READONLY.  The former can be set to instruct KVM to keep track of
> -writes to memory within the slot.  See KVM_GET_DIRTY_LOG ioctl to know how to
> -use it.  The latter can be set, if KVM_CAP_READONLY_MEM capability allows it,
> -to make a new slot read-only.  In this case, writes to this memory will be
> -posted to userspace as KVM_EXIT_MMIO exits.
> +The flags field supports the following flags: KVM_MEM_LOG_DIRTY_PAGES,
> +KVM_MEM_READONLY, KVM_MEM_PCI_HOLE:
> +- KVM_MEM_LOG_DIRTY_PAGES: log writes.  Use KVM_GET_DIRTY_LOG to retreive
> +  the log.
> +- KVM_MEM_READONLY: exit to userspace with KVM_EXIT_MMIO on writes.  Only
> +  available when KVM_CAP_READONLY_MEM is present.
> +- KVM_MEM_PCI_HOLE: always return 0xff on reads, exit to userspace with
> +  KVM_EXIT_MMIO on writes.  Only available when KVM_CAP_PCI_HOLE_MEM is
> +  present.  When setting the memory region 'userspace_addr' must be NULL.
> +  This flag is mutually exclusive with KVM_MEM_LOG_DIRTY_PAGES and with
> +  KVM_MEM_READONLY.
>  
>  When the KVM_CAP_SYNC_MMU capability is available, changes in the backing of
>  the memory region are automatically reflected into the guest.  For example, an
> diff --git a/arch/x86/include/uapi/asm/kvm.h b/arch/x86/include/uapi/asm/kvm.h
> index 17c5a038f42d..cf80a26d74f5 100644
> --- a/arch/x86/include/uapi/asm/kvm.h
> +++ b/arch/x86/include/uapi/asm/kvm.h
> @@ -48,6 +48,7 @@
>  #define __KVM_HAVE_XSAVE
>  #define __KVM_HAVE_XCRS
>  #define __KVM_HAVE_READONLY_MEM
> +#define __KVM_HAVE_PCI_HOLE_MEM
>  
>  /* Architectural interrupt line count. */
>  #define KVM_NR_INTERRUPTS 256
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index fef6956393f7..4a2a7fface1e 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -3254,7 +3254,7 @@ static int kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, gfn_t gfn,
>  		return PG_LEVEL_4K;
>  
>  	slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, true);
> -	if (!slot)
> +	if (!slot || (slot->flags & KVM_MEM_PCI_HOLE))

This is unnecessary since you're setting disallow_lpage in
kvm_alloc_memslot_metadata().

>  		return PG_LEVEL_4K;
>  
>  	max_level = min(max_level, max_huge_page_level);
> @@ -4105,6 +4105,9 @@ static int direct_page_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code,
>  
>  	slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
>  
> +	if (!write && slot && (slot->flags & KVM_MEM_PCI_HOLE))

I'm confused.  Why does this short circuit reads but not writes?

> +		return RET_PF_EMULATE;
> +
>  	if (try_async_pf(vcpu, slot, prefault, gfn, gpa, &pfn, write,
>  			 &map_writable))
>  		return RET_PF_RETRY;
> diff --git a/arch/x86/kvm/mmu/paging_tmpl.h b/arch/x86/kvm/mmu/paging_tmpl.h
> index 5c6a895f67c3..27abd69e69f6 100644
> --- a/arch/x86/kvm/mmu/paging_tmpl.h
> +++ b/arch/x86/kvm/mmu/paging_tmpl.h
> @@ -836,6 +836,9 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gpa_t addr, u32 error_code,
>  
>  	slot = kvm_vcpu_gfn_to_memslot(vcpu, walker.gfn);
>  
> +	if (!write_fault && slot && (slot->flags & KVM_MEM_PCI_HOLE))
> +		return RET_PF_EMULATE;
> +
>  	if (try_async_pf(vcpu, slot, prefault, walker.gfn, addr, &pfn,
>  			 write_fault, &map_writable))
>  		return RET_PF_RETRY;
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index dc4370394ab8..538bc58a22db 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -3515,6 +3515,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>  	case KVM_CAP_EXCEPTION_PAYLOAD:
>  	case KVM_CAP_SET_GUEST_DEBUG:
>  	case KVM_CAP_LAST_CPU:
> +	case KVM_CAP_PCI_HOLE_MEM:
>  		r = 1;
>  		break;
>  	case KVM_CAP_SYNC_REGS:
> @@ -10114,9 +10115,11 @@ static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
>  		ugfn = slot->userspace_addr >> PAGE_SHIFT;
>  		/*
>  		 * If the gfn and userspace address are not aligned wrt each
> -		 * other, disable large page support for this slot.
> +		 * other, disable large page support for this slot. Also,
> +		 * disable large page support for KVM_MEM_PCI_HOLE slots.
>  		 */
> -		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
> +		if ((slot->flags & KVM_MEM_PCI_HOLE) || ((slot->base_gfn ^ ugfn) &
> +				      (KVM_PAGES_PER_HPAGE(level) - 1))) {
>  			unsigned long j;
>  
>  			for (j = 0; j < lpages; ++j)
> @@ -10178,7 +10181,8 @@ static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
>  	 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
>  	 * See comments below.
>  	 */
> -	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
> +	if ((change != KVM_MR_FLAGS_ONLY) ||
> +	    (new->flags & (KVM_MEM_READONLY | KVM_MEM_PCI_HOLE)))
>  		return;
>  
>  	/*
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index 989afcbe642f..de1faa64a8ef 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -1081,6 +1081,9 @@ __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
>  static inline unsigned long
>  __gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
>  {
> +	/* Should never be called with a KVM_MEM_PCI_HOLE slot */
> +	BUG_ON(!slot->userspace_addr);

So _technically_, userspace can hit this by allowing virtual address 0,
which is very much non-standard, but theoretically legal.  It'd probably be
better to use a value that can't possibly be a valid userspace_addr, e.g. a
non-canonical value.

> +
>  	return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
>  }
>  

...

> @@ -2318,6 +2338,11 @@ static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
>  	int r;
>  	unsigned long addr;
>  
> +	if (unlikely(slot && (slot->flags & KVM_MEM_PCI_HOLE))) {
> +		memset(data, 0xff, len);
> +		return 0;
> +	}

This feels wrong, shouldn't we be treating PCI_HOLE as MMIO?  Given that
this is performance oriented, I would think we'd want to leverage the
GPA from the VMCS instead of doing a full translation.

That brings up a potential alternative to adding a memslot flag.  What if
we instead add a KVM_MMIO_BUS device similar to coalesced MMIO?  I think
it'd be about the same amount of KVM code, and it would provide userspace
with more flexibility, e.g. I assume it would allow handling even writes
wholly within the kernel for certain ranges and/or use cases, and it'd
allow stuffing a value other than 0xff (though I have no idea if there is
a use case for this).

Speaking of which, why do writes go to userspace in this series?

> +
>  	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
>  	if (kvm_is_error_hva(addr))
>  		return -EFAULT;
> -- 
> 2.25.4
>