From: Paul Durrant pdurrant@amazon.com

There is no need for the existing kvm_gpc_XXX() functions to be exported. Clean up now before additional functions are added in subsequent patches.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: Sean Christopherson seanjc@google.com Cc: David Woodhouse dwmw2@infradead.org Cc: Paolo Bonzini pbonzini@redhat.com

v8: - New in this version. --- virt/kvm/pfncache.c | 5 ----- 1 file changed, 5 deletions(-)

diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c index 10842f1eeeae..f3571f44d9af 100644 --- a/virt/kvm/pfncache.c +++ b/virt/kvm/pfncache.c @@ -94,7 +94,6 @@ bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len)

return true; } -EXPORT_SYMBOL_GPL(kvm_gpc_check);

static void *gpc_map(kvm_pfn_t pfn) { @@ -346,7 +345,6 @@ int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len) { return __kvm_gpc_refresh(gpc, gpc->gpa, len); } -EXPORT_SYMBOL_GPL(kvm_gpc_refresh);

void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm, struct kvm_vcpu *vcpu, enum pfn_cache_usage usage) @@ -363,7 +361,6 @@ void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm, gpc->pfn = KVM_PFN_ERR_FAULT; gpc->uhva = KVM_HVA_ERR_BAD; } -EXPORT_SYMBOL_GPL(kvm_gpc_init);

int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len) { @@ -388,7 +385,6 @@ int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len) } return __kvm_gpc_refresh(gpc, gpa, len); } -EXPORT_SYMBOL_GPL(kvm_gpc_activate);

void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc) { @@ -426,4 +422,3 @@ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc) gpc_unmap(old_pfn, old_khva); } } -EXPORT_SYMBOL_GPL(kvm_gpc_deactivate);

From: Paul Durrant pdurrant@amazon.com

At the moment pages are marked dirty by open-coded calls to mark_page_dirty_in_slot(), directly deferefencing the gpa and memslot from the cache. After a subsequent patch these may not always be set so add a helper now so that caller will protected from the need to know about this detail.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: David Woodhouse dwmw2@infradead.org Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Ingo Molnar mingo@redhat.com Cc: Borislav Petkov bp@alien8.de Cc: Dave Hansen dave.hansen@linux.intel.com Cc: "H. Peter Anvin" hpa@zytor.com Cc: x86@kernel.org

v8: - Make the helper a static inline. --- arch/x86/kvm/x86.c | 2 +- arch/x86/kvm/xen.c | 6 +++--- include/linux/kvm_host.h | 11 +++++++++++ 3 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 27e23714e960..0a0ac91a494f 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -3156,7 +3156,7 @@ static void kvm_setup_guest_pvclock(struct kvm_vcpu *v,

guest_hv_clock->version = ++vcpu->hv_clock.version;

- mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT); + kvm_gpc_mark_dirty(gpc); read_unlock_irqrestore(&gpc->lock, flags);

trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c index b63bf54bb376..34c48d0029c1 100644 --- a/arch/x86/kvm/xen.c +++ b/arch/x86/kvm/xen.c @@ -453,11 +453,11 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic) }

if (user_len2) { - mark_page_dirty_in_slot(v->kvm, gpc2->memslot, gpc2->gpa >> PAGE_SHIFT); + kvm_gpc_mark_dirty(gpc2); read_unlock(&gpc2->lock); }

- mark_page_dirty_in_slot(v->kvm, gpc1->memslot, gpc1->gpa >> PAGE_SHIFT); + kvm_gpc_mark_dirty(gpc1); read_unlock_irqrestore(&gpc1->lock, flags); }

@@ -565,7 +565,7 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v) WRITE_ONCE(vi->evtchn_upcall_pending, 1); }

- mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT); + kvm_gpc_mark_dirty(gpc); read_unlock_irqrestore(&gpc->lock, flags);

/* For the per-vCPU lapic vector, deliver it as MSI. */ diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index 7e7fd25b09b3..f3bb9e0a81fe 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -1399,6 +1399,17 @@ int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len); */ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc);

+/** + * kvm_gpc_mark_dirty - mark a cached page as dirty. + * + * @gpc: struct gfn_to_pfn_cache object. + */ +static inline void kvm_gpc_mark_dirty(struct gfn_to_pfn_cache *gpc) +{ + lockdep_assert_held(&gpc->lock); + mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT); +} + void kvm_sigset_activate(struct kvm_vcpu *vcpu); void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);

From: Paul Durrant pdurrant@amazon.com

Some code in pfncache uses offset_in_page() but in other places it is open- coded. Use offset_in_page() consistently everywhere.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: David Woodhouse dwmw2@infradead.org

v8: - New in this version. --- virt/kvm/pfncache.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c index 6f4b537eb25b..0eeb034d0674 100644 --- a/virt/kvm/pfncache.c +++ b/virt/kvm/pfncache.c @@ -48,7 +48,7 @@ bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len) if (!gpc->active) return false;

- if ((gpc->gpa & ~PAGE_MASK) + len > PAGE_SIZE) + if (offset_in_page(gpc->gpa) + len > PAGE_SIZE) return false;

if (gpc->generation != slots->generation || kvm_is_error_hva(gpc->uhva)) @@ -192,7 +192,7 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)

gpc->valid = true; gpc->pfn = new_pfn; - gpc->khva = new_khva + (gpc->gpa & ~PAGE_MASK); + gpc->khva = new_khva + offset_in_page(gpc->gpa);

/* * Put the reference to the _new_ pfn. The pfn is now tracked by the @@ -213,7 +213,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len) { struct kvm_memslots *slots = kvm_memslots(gpc->kvm); - unsigned long page_offset = gpa & ~PAGE_MASK; + unsigned long page_offset = offset_in_page(gpa); bool unmap_old = false; unsigned long old_uhva; kvm_pfn_t old_pfn;

From: Paul Durrant pdurrant@amazon.com

Some pfncache pages may actually be overlays on guest memory that have a fixed HVA within the VMM. It's pointless to invalidate such cached mappings if the overlay is moved so allow a cache to be activated directly with the HVA to cater for such cases. A subsequent patch will make use of this facility.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: David Woodhouse dwmw2@infradead.org

v11: - Fixed kvm_gpc_check() to ignore memslot generation if the cache is not activated with a GPA. (This breakage occured during the re-work for v8).

v9: - Pass both GPA and HVA into __kvm_gpc_refresh() rather than overloading the address paraneter and using a bool flag to indicated what it is.

v8: - Re-worked to avoid messing with struct gfn_to_pfn_cache. --- include/linux/kvm_host.h | 20 +++++++++++++++++++- virt/kvm/pfncache.c | 40 +++++++++++++++++++++++++++++++--------- 2 files changed, 50 insertions(+), 10 deletions(-)

diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index f2354f808d04..7994c4d16783 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -1344,6 +1344,22 @@ void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm); */ int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len);

+/** + * kvm_gpc_activate_hva - prepare a cached kernel mapping and HPA for a given HVA. + * + * @gpc: struct gfn_to_pfn_cache object. + * @hva: userspace virtual address to map. + * @len: sanity check; the range being access must fit a single page. + * + * @return: 0 for success. + * -EINVAL for a mapping which would cross a page boundary. + * -EFAULT for an untranslatable guest physical address. + * + * The semantics of this function are the same as those of kvm_gpc_activate(). It + * merely bypasses a layer of address translation. + */ +int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long hva, unsigned long len); + /** * kvm_gpc_check - check validity of a gfn_to_pfn_cache. * @@ -1398,7 +1414,9 @@ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc); static inline void kvm_gpc_mark_dirty(struct gfn_to_pfn_cache *gpc) { lockdep_assert_held(&gpc->lock); - mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT); + + if (gpc->gpa != KVM_XEN_INVALID_GPA) + mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT); }

void kvm_sigset_activate(struct kvm_vcpu *vcpu); diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c index 97eec8ee3449..ae822bff812f 100644 --- a/virt/kvm/pfncache.c +++ b/virt/kvm/pfncache.c @@ -48,7 +48,10 @@ bool kvm_gpc_check(struct gfn_to_pfn_cache *gpc, unsigned long len) if (!gpc->active) return false;

- if (gpc->generation != slots->generation || kvm_is_error_hva(gpc->uhva)) + if (gpc->gpa != KVM_XEN_INVALID_GPA && gpc->generation != slots->generation) + return false; + + if (kvm_is_error_hva(gpc->uhva)) return false;

if (offset_in_page(gpc->uhva) + len > PAGE_SIZE) @@ -209,11 +212,13 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc) return -EFAULT; }

-static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, +static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long uhva, unsigned long len) { struct kvm_memslots *slots = kvm_memslots(gpc->kvm); - unsigned long page_offset = offset_in_page(gpa); + unsigned long page_offset = (gpa != KVM_XEN_INVALID_GPA) ? + offset_in_page(gpa) : + offset_in_page(uhva); bool unmap_old = false; unsigned long old_uhva; kvm_pfn_t old_pfn; @@ -246,9 +251,15 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, old_khva = (void *)PAGE_ALIGN_DOWN((uintptr_t)gpc->khva); old_uhva = PAGE_ALIGN_DOWN(gpc->uhva);

- /* Refresh the userspace HVA if necessary */ - if (gpc->gpa != gpa || gpc->generation != slots->generation || - kvm_is_error_hva(gpc->uhva)) { + if (gpa == KVM_XEN_INVALID_GPA) { + gpc->gpa = KVM_XEN_INVALID_GPA; + gpc->uhva = PAGE_ALIGN_DOWN(uhva); + + if (gpc->uhva != old_uhva) + hva_change = true; + } else if (gpc->gpa != gpa || + gpc->generation != slots->generation || + kvm_is_error_hva(gpc->uhva)) { gfn_t gfn = gpa_to_gfn(gpa);

gpc->gpa = gpa; @@ -319,7 +330,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa,

int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len) { - return __kvm_gpc_refresh(gpc, gpc->gpa, len); + return __kvm_gpc_refresh(gpc, gpc->gpa, gpc->uhva, len); }

void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm) @@ -332,7 +343,8 @@ void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm) gpc->uhva = KVM_HVA_ERR_BAD; }

-int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len) +static int __kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long uhva, + unsigned long len) { struct kvm *kvm = gpc->kvm;

@@ -353,7 +365,17 @@ int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len) gpc->active = true; write_unlock_irq(&gpc->lock); } - return __kvm_gpc_refresh(gpc, gpa, len); + return __kvm_gpc_refresh(gpc, gpa, uhva, len); +} + +int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len) +{ + return __kvm_gpc_activate(gpc, gpa, KVM_HVA_ERR_BAD, len); +} + +int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long uhva, unsigned long len) +{ + return __kvm_gpc_activate(gpc, KVM_XEN_INVALID_GPA, uhva, len); }

void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc)

From: Paul Durrant pdurrant@amazon.com

A subsequent patch will allow shared_info to be initialized using either a GPA or a user-space (i.e. VMM) HVA. To make that patch cleaner, separate the initialization of the shared_info content from the activation of the pfncache.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Ingo Molnar mingo@redhat.com Cc: Borislav Petkov bp@alien8.de Cc: Dave Hansen dave.hansen@linux.intel.com Cc: "H. Peter Anvin" hpa@zytor.com Cc: David Woodhouse dwmw2@infradead.org Cc: x86@kernel.org

v11: - Fix accidental regression from commit 5d6d6a7d7e66a ("KVM: x86: Refine calculation of guest wall clock to use a single TSC read").

v10: - New in this version. --- arch/x86/kvm/xen.c | 55 +++++++++++++++++++++++++++------------------- 1 file changed, 32 insertions(+), 23 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c index f9b1e494c430..df53fea73747 100644 --- a/arch/x86/kvm/xen.c +++ b/arch/x86/kvm/xen.c @@ -34,41 +34,32 @@ static bool kvm_xen_hcall_evtchn_send(struct kvm_vcpu *vcpu, u64 param, u64 *r);

DEFINE_STATIC_KEY_DEFERRED_FALSE(kvm_xen_enabled, HZ);

-static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn) +static int kvm_xen_shared_info_init(struct kvm *kvm) { struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache; struct pvclock_wall_clock *wc; - gpa_t gpa = gfn_to_gpa(gfn); u32 *wc_sec_hi; u32 wc_version; u64 wall_nsec; int ret = 0; int idx = srcu_read_lock(&kvm->srcu);

- if (gfn == KVM_XEN_INVALID_GFN) { - kvm_gpc_deactivate(gpc); - goto out; - } + read_lock_irq(&gpc->lock); + while (!kvm_gpc_check(gpc, PAGE_SIZE)) { + read_unlock_irq(&gpc->lock);

- do { - ret = kvm_gpc_activate(gpc, gpa, PAGE_SIZE); + ret = kvm_gpc_refresh(gpc, PAGE_SIZE); if (ret) goto out;

- /* - * This code mirrors kvm_write_wall_clock() except that it writes - * directly through the pfn cache and doesn't mark the page dirty. - */ - wall_nsec = kvm_get_wall_clock_epoch(kvm); - - /* It could be invalid again already, so we need to check */ read_lock_irq(&gpc->lock); + }

- if (gpc->valid) - break; - - read_unlock_irq(&gpc->lock); - } while (1); + /* + * This code mirrors kvm_write_wall_clock() except that it writes + * directly through the pfn cache and doesn't mark the page dirty. + */ + wall_nsec = kvm_get_wall_clock_epoch(kvm);

/* Paranoia checks on the 32-bit struct layout */ BUILD_BUG_ON(offsetof(struct compat_shared_info, wc) != 0x900); @@ -639,12 +630,30 @@ int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data) } break;

- case KVM_XEN_ATTR_TYPE_SHARED_INFO: + case KVM_XEN_ATTR_TYPE_SHARED_INFO: { + int idx; + mutex_lock(&kvm->arch.xen.xen_lock); - r = kvm_xen_shared_info_init(kvm, data->u.shared_info.gfn); + + idx = srcu_read_lock(&kvm->srcu); + + if (data->u.shared_info.gfn == KVM_XEN_INVALID_GFN) { + kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache); + r = 0; + } else { + r = kvm_gpc_activate(&kvm->arch.xen.shinfo_cache, + gfn_to_gpa(data->u.shared_info.gfn), + PAGE_SIZE); + } + + srcu_read_unlock(&kvm->srcu, idx); + + if (!r && kvm->arch.xen.shinfo_cache.active) + r = kvm_xen_shared_info_init(kvm); + mutex_unlock(&kvm->arch.xen.xen_lock); break; - + } case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR: if (data->u.vector && data->u.vector < 0x10) r = -EINVAL;

From: Paul Durrant pdurrant@amazon.com

The shared_info page is not guest memory as such. It is a dedicated page allocated by the VMM and overlaid onto guest memory in a GFN chosen by the guest and specified in the XENMEM_add_to_physmap hypercall. The guest may even request that shared_info be moved from one GFN to another by re-issuing that hypercall, but the HVA is never going to change.

Because the shared_info page is an overlay the memory slots need to be updated in response to the hypercall. However, memory slot adjustment is not atomic and, whilst all vCPUs are paused, there is still the possibility that events may be delivered (which requires the shared_info page to be updated) whilst the shared_info GPA is absent. The HVA is never absent though, so it makes much more sense to use that as the basis for the kernel's mapping.

Hence add a new KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA attribute type for this purpose and a KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA flag to advertize its availability. Don't actually advertize it yet though. That will be done in a subsequent patch, which will also add tests for the new attribute type.

Also update the KVM API documentation with the new attribute and also fix it up to consistently refer to 'shared_info' (with the underscore).

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: David Woodhouse dwmw2@infradead.org Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Ingo Molnar mingo@redhat.com Cc: Borislav Petkov bp@alien8.de Cc: Dave Hansen dave.hansen@linux.intel.com Cc: "H. Peter Anvin" hpa@zytor.com Cc: x86@kernel.org

v8: - Re-base.

v2: - Define the new attribute and capability but don't advertize the capability yet. - Add API documentation. --- Documentation/virt/kvm/api.rst | 25 ++++++++++++++++------ arch/x86/kvm/xen.c | 39 ++++++++++++++++++++++++++-------- include/uapi/linux/kvm.h | 6 +++++- 3 files changed, 54 insertions(+), 16 deletions(-)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst index 3ec0b7a455a0..3372be85b335 100644 --- a/Documentation/virt/kvm/api.rst +++ b/Documentation/virt/kvm/api.rst @@ -372,7 +372,7 @@ The bits in the dirty bitmap are cleared before the ioctl returns, unless KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 is enabled. For more information, see the description of the capability.

-Note that the Xen shared info page, if configured, shall always be assumed +Note that the Xen shared_info page, if configured, shall always be assumed to be dirty. KVM will not explicitly mark it such.

@@ -5487,8 +5487,9 @@ KVM_PV_ASYNC_CLEANUP_PERFORM __u8 long_mode; __u8 vector; __u8 runstate_update_flag; - struct { + union { __u64 gfn; + __u64 hva; } shared_info; struct { __u32 send_port; @@ -5516,10 +5517,10 @@ type values:

KVM_XEN_ATTR_TYPE_LONG_MODE Sets the ABI mode of the VM to 32-bit or 64-bit (long mode). This - determines the layout of the shared info pages exposed to the VM. + determines the layout of the shared_info page exposed to the VM.

KVM_XEN_ATTR_TYPE_SHARED_INFO - Sets the guest physical frame number at which the Xen "shared info" + Sets the guest physical frame number at which the Xen shared_info page resides. Note that although Xen places vcpu_info for the first 32 vCPUs in the shared_info page, KVM does not automatically do so and instead requires that KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO be used @@ -5528,7 +5529,7 @@ KVM_XEN_ATTR_TYPE_SHARED_INFO not be aware of the Xen CPU id which is used as the index into the vcpu_info[] array, so may know the correct default location.

- Note that the shared info page may be constantly written to by KVM; + Note that the shared_info page may be constantly written to by KVM; it contains the event channel bitmap used to deliver interrupts to a Xen guest, amongst other things. It is exempt from dirty tracking mechanisms — KVM will not explicitly mark the page as dirty each @@ -5537,9 +5538,21 @@ KVM_XEN_ATTR_TYPE_SHARED_INFO any vCPU has been running or any event channel interrupts can be routed to the guest.

- Setting the gfn to KVM_XEN_INVALID_GFN will disable the shared info + Setting the gfn to KVM_XEN_INVALID_GFN will disable the shared_info page.

+KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA + If the KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA flag is also set in the + Xen capabilities, then this attribute may be used to set the + userspace address at which the shared_info page resides, which + will always be fixed in the VMM regardless of where it is mapped + in guest physical address space. This attribute should be used in + preference to KVM_XEN_ATTR_TYPE_SHARED_INFO as it avoids + unnecessary invalidation of an internal cache when the page is + re-mapped in guest physcial address space. + + Setting the hva to zero will disable the shared_info page. + KVM_XEN_ATTR_TYPE_UPCALL_VECTOR Sets the exception vector used to deliver Xen event channel upcalls. This is the HVM-wide vector injected directly by the hypervisor diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c index d595d476a5b3..813d63a8703a 100644 --- a/arch/x86/kvm/xen.c +++ b/arch/x86/kvm/xen.c @@ -617,7 +617,6 @@ int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data) { int r = -ENOENT;

- switch (data->type) { case KVM_XEN_ATTR_TYPE_LONG_MODE: if (!IS_ENABLED(CONFIG_64BIT) && data->u.long_mode) { @@ -638,20 +637,32 @@ int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data) } break;

- case KVM_XEN_ATTR_TYPE_SHARED_INFO: { + case KVM_XEN_ATTR_TYPE_SHARED_INFO: + case KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA: { int idx;

mutex_lock(&kvm->arch.xen.xen_lock);

idx = srcu_read_lock(&kvm->srcu);

- if (data->u.shared_info.gfn == KVM_XEN_INVALID_GFN) { - kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache); - r = 0; + if (data->type == KVM_XEN_ATTR_TYPE_SHARED_INFO) { + if (data->u.shared_info.gfn == KVM_XEN_INVALID_GFN) { + kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache); + r = 0; + } else { + r = kvm_gpc_activate(&kvm->arch.xen.shinfo_cache, + gfn_to_gpa(data->u.shared_info.gfn), + PAGE_SIZE); + } } else { - r = kvm_gpc_activate(&kvm->arch.xen.shinfo_cache, - gfn_to_gpa(data->u.shared_info.gfn), - PAGE_SIZE); + if (data->u.shared_info.hva == 0) { + kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache); + r = 0; + } else { + r = kvm_gpc_activate_hva(&kvm->arch.xen.shinfo_cache, + data->u.shared_info.hva, + PAGE_SIZE); + } }

srcu_read_unlock(&kvm->srcu, idx); @@ -715,13 +726,23 @@ int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data) break;

case KVM_XEN_ATTR_TYPE_SHARED_INFO: - if (kvm->arch.xen.shinfo_cache.active) + if (kvm->arch.xen.shinfo_cache.active && + kvm->arch.xen.shinfo_cache.gpa != KVM_XEN_INVALID_GPA) data->u.shared_info.gfn = gpa_to_gfn(kvm->arch.xen.shinfo_cache.gpa); else data->u.shared_info.gfn = KVM_XEN_INVALID_GFN; r = 0; break;

+ case KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA: + if (kvm->arch.xen.shinfo_cache.active && + kvm->arch.xen.shinfo_cache.gpa == KVM_XEN_INVALID_GPA) + data->u.shared_info.hva = kvm->arch.xen.shinfo_cache.uhva; + else + data->u.shared_info.hva = 0; + r = 0; + break; + case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR: data->u.vector = kvm->arch.xen.upcall_vector; r = 0; diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h index c3308536482b..ac5caba313d1 100644 --- a/include/uapi/linux/kvm.h +++ b/include/uapi/linux/kvm.h @@ -1246,6 +1246,7 @@ struct kvm_x86_mce { #define KVM_XEN_HVM_CONFIG_EVTCHN_SEND (1 << 5) #define KVM_XEN_HVM_CONFIG_RUNSTATE_UPDATE_FLAG (1 << 6) #define KVM_XEN_HVM_CONFIG_PVCLOCK_TSC_UNSTABLE (1 << 7) +#define KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA (1 << 8)

struct kvm_xen_hvm_config { __u32 flags; @@ -1744,9 +1745,10 @@ struct kvm_xen_hvm_attr { __u8 long_mode; __u8 vector; __u8 runstate_update_flag; - struct { + union { __u64 gfn; #define KVM_XEN_INVALID_GFN ((__u64)-1) + __u64 hva; } shared_info; struct { __u32 send_port; @@ -1788,6 +1790,8 @@ struct kvm_xen_hvm_attr { #define KVM_XEN_ATTR_TYPE_XEN_VERSION 0x4 /* Available with KVM_CAP_XEN_HVM / KVM_XEN_HVM_CONFIG_RUNSTATE_UPDATE_FLAG */ #define KVM_XEN_ATTR_TYPE_RUNSTATE_UPDATE_FLAG 0x5 +/* Available with KVM_CAP_XEN_HVM / KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA */ +#define KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA 0x6

/* Per-vCPU Xen attributes */ #define KVM_XEN_VCPU_GET_ATTR _IOWR(KVMIO, 0xca, struct kvm_xen_vcpu_attr)

From: Paul Durrant pdurrant@amazon.com

If the guest does not explicitly set the GPA of vcpu_info structure in memory then, for guests with 32 vCPUs or fewer, the vcpu_info embedded in the shared_info page may be used. As described in a previous commit, the shared_info page is an overlay at a fixed HVA within the VMM, so in this case it also more optimal to activate the vcpu_info cache with a fixed HVA to avoid unnecessary invalidation if the guest memory layout is modified.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: David Woodhouse dwmw2@infradead.org Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Ingo Molnar mingo@redhat.com Cc: Borislav Petkov bp@alien8.de Cc: Dave Hansen dave.hansen@linux.intel.com Cc: "H. Peter Anvin" hpa@zytor.com Cc: x86@kernel.org

v8: - Re-base.

v5: - New in this version. --- Documentation/virt/kvm/api.rst | 26 +++++++++++++++++++++----- arch/x86/kvm/xen.c | 34 ++++++++++++++++++++++++++++------ include/uapi/linux/kvm.h | 3 +++ 3 files changed, 52 insertions(+), 11 deletions(-)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst index 3372be85b335..bd93cafd3e4e 100644 --- a/Documentation/virt/kvm/api.rst +++ b/Documentation/virt/kvm/api.rst @@ -5523,11 +5523,12 @@ KVM_XEN_ATTR_TYPE_SHARED_INFO Sets the guest physical frame number at which the Xen shared_info page resides. Note that although Xen places vcpu_info for the first 32 vCPUs in the shared_info page, KVM does not automatically do so - and instead requires that KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO be used - explicitly even when the vcpu_info for a given vCPU resides at the - "default" location in the shared_info page. This is because KVM may - not be aware of the Xen CPU id which is used as the index into the - vcpu_info[] array, so may know the correct default location. + and instead requires that KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO or + KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA be used explicitly even when + the vcpu_info for a given vCPU resides at the "default" location + in the shared_info page. This is because KVM may not be aware of + the Xen CPU id which is used as the index into the vcpu_info[] + array, so may know the correct default location.

Note that the shared_info page may be constantly written to by KVM; it contains the event channel bitmap used to deliver interrupts to @@ -5649,6 +5650,21 @@ KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO on dirty logging. Setting the gpa to KVM_XEN_INVALID_GPA will disable the vcpu_info.

+KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA + If the KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA flag is also set in the + Xen capabilities, then this attribute may be used to set the + userspace address of the vcpu_info for a given vCPU. It should + only be used when the vcpu_info resides at the "default" location + in the shared_info page. In this case it is safe to assume the + userspace address will not change, because the shared_info page is + an overlay on guest memory and remains at a fixed host address + regardless of where it is mapped in guest physical address space + and hence unnecessary invalidation of an internal cache may be + avoided if the guest memory layout is modified. + If the vcpu_info does not reside at the "default" location then + it is not guaranteed to remain at the same host address and + hence the aforementioned cache invalidation is required. + KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO Sets the guest physical address of an additional pvclock structure for a given vCPU. This is typically used for guest vsyscall support. diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c index 813d63a8703a..c99008d217f4 100644 --- a/arch/x86/kvm/xen.c +++ b/arch/x86/kvm/xen.c @@ -779,20 +779,33 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)

switch (data->type) { case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO: + case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA: /* No compat necessary here. */ BUILD_BUG_ON(sizeof(struct vcpu_info) != sizeof(struct compat_vcpu_info)); BUILD_BUG_ON(offsetof(struct vcpu_info, time) != offsetof(struct compat_vcpu_info, time));

- if (data->u.gpa == KVM_XEN_INVALID_GPA) { - kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache); - r = 0; - break; + if (data->type == KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO) { + if (data->u.gpa == KVM_XEN_INVALID_GPA) { + kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache); + r = 0; + break; + } + + r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_info_cache, + data->u.gpa, sizeof(struct vcpu_info)); + } else { + if (data->u.hva == 0) { + kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache); + r = 0; + break; + } + + r = kvm_gpc_activate_hva(&vcpu->arch.xen.vcpu_info_cache, + data->u.hva, sizeof(struct vcpu_info)); }

- r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_info_cache, - data->u.gpa, sizeof(struct vcpu_info)); if (!r) kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);

@@ -1021,6 +1034,15 @@ int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data) r = 0; break;

+ case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA: + if (vcpu->arch.xen.vcpu_info_cache.active && + vcpu->arch.xen.vcpu_info_cache.gpa == KVM_XEN_INVALID_GPA) + data->u.hva = vcpu->arch.xen.vcpu_info_cache.uhva; + else + data->u.hva = 0; + r = 0; + break; + case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO: if (vcpu->arch.xen.vcpu_time_info_cache.active) data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa; diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h index ac5caba313d1..d2665319db6e 100644 --- a/include/uapi/linux/kvm.h +++ b/include/uapi/linux/kvm.h @@ -1809,6 +1809,7 @@ struct kvm_xen_vcpu_attr { union { __u64 gpa; #define KVM_XEN_INVALID_GPA ((__u64)-1) + __u64 hva; __u64 pad[8]; struct { __u64 state; @@ -1839,6 +1840,8 @@ struct kvm_xen_vcpu_attr { #define KVM_XEN_VCPU_ATTR_TYPE_VCPU_ID 0x6 #define KVM_XEN_VCPU_ATTR_TYPE_TIMER 0x7 #define KVM_XEN_VCPU_ATTR_TYPE_UPCALL_VECTOR 0x8 +/* Available with KVM_CAP_XEN_HVM / KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA */ +#define KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA 0x9

/* Secure Encrypted Virtualization command */ enum sev_cmd_id {

From: Paul Durrant pdurrant@amazon.com

Now that all relevant kernel changes and selftests are in place, enable the new capability.

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Ingo Molnar mingo@redhat.com Cc: Borislav Petkov bp@alien8.de Cc: Dave Hansen dave.hansen@linux.intel.com Cc: "H. Peter Anvin" hpa@zytor.com Cc: David Woodhouse dwmw2@infradead.org Cc: x86@kernel.org

v2: - New in this version. --- arch/x86/kvm/x86.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 19e6cc1dadfe..4864251dca41 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -4678,7 +4678,8 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) KVM_XEN_HVM_CONFIG_SHARED_INFO | KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL | KVM_XEN_HVM_CONFIG_EVTCHN_SEND | - KVM_XEN_HVM_CONFIG_PVCLOCK_TSC_UNSTABLE; + KVM_XEN_HVM_CONFIG_PVCLOCK_TSC_UNSTABLE | + KVM_XEN_HVM_CONFIG_SHARED_INFO_HVA; if (sched_info_on()) r |= KVM_XEN_HVM_CONFIG_RUNSTATE | KVM_XEN_HVM_CONFIG_RUNSTATE_UPDATE_FLAG;

From: Paul Durrant pdurrant@amazon.com

As described in [1] compiling with CONFIG_PROVE_RAW_LOCK_NESTING shows that kvm_xen_set_evtchn_fast() is blocking on pfncache locks in IRQ context. There is only actually blocking with PREEMPT_RT because the locks will turned into mutexes. There is no 'raw' version of rwlock_t that can be used to avoid that, so use read_trylock() and treat failure to lock the same as an invalid cache.

[1] https://lore.kernel.org/lkml/99771ef3a4966a01fefd3adbb2ba9c3a75f97cf2.camel@...

Signed-off-by: Paul Durrant pdurrant@amazon.com Reviewed-by: David Woodhouse dwmw@amazon.co.uk --- Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Ingo Molnar mingo@redhat.com Cc: Borislav Petkov bp@alien8.de Cc: Dave Hansen dave.hansen@linux.intel.com Cc: "H. Peter Anvin" hpa@zytor.com Cc: David Woodhouse dwmw2@infradead.org Cc: x86@kernel.org

v11: - Amended the commit comment.

v10: - New in this version. --- arch/x86/kvm/xen.c | 30 ++++++++++++++++++++---------- 1 file changed, 20 insertions(+), 10 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c index 5ce02699f44c..9168a6ec88fd 100644 --- a/arch/x86/kvm/xen.c +++ b/arch/x86/kvm/xen.c @@ -1673,10 +1673,13 @@ static int set_shinfo_evtchn_pending(struct kvm_vcpu *vcpu, u32 port) unsigned long flags; int rc = -EWOULDBLOCK;

- read_lock_irqsave(&gpc->lock, flags); - if (!kvm_gpc_check(gpc, PAGE_SIZE)) + local_irq_save(flags); + if (!read_trylock(&gpc->lock)) goto out;

+ if (!kvm_gpc_check(gpc, PAGE_SIZE)) + goto out_unlock; + if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) { struct shared_info *shinfo = gpc->khva;

@@ -1698,8 +1701,10 @@ static int set_shinfo_evtchn_pending(struct kvm_vcpu *vcpu, u32 port) rc = 1; /* It is newly raised */ }

+ out_unlock: + read_unlock(&gpc->lock); out: - read_unlock_irqrestore(&gpc->lock, flags); + local_irq_restore(flags); return rc; }

@@ -1709,21 +1714,23 @@ static bool set_vcpu_info_evtchn_pending(struct kvm_vcpu *vcpu, u32 port) struct gfn_to_pfn_cache *gpc = &vcpu->arch.xen.vcpu_info_cache; unsigned long flags; bool kick_vcpu = false; + bool locked;

- read_lock_irqsave(&gpc->lock, flags); + local_irq_save(flags); + locked = read_trylock(&gpc->lock);

/* * Try to deliver the event directly to the vcpu_info. If successful and * the guest is using upcall_vector delivery, send the MSI. - * If the pfncache is invalid, set the shadow. In this case, or if the - * guest is using another form of event delivery, the vCPU must be - * kicked to complete the delivery. + * If the pfncache lock is contended or the cache is invalid, set the + * shadow. In this case, or if the guest is using another form of event + * delivery, the vCPU must be kicked to complete the delivery. */ if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) { struct vcpu_info *vcpu_info = gpc->khva; int port_word_bit = port / 64;

- if (!kvm_gpc_check(gpc, sizeof(*vcpu_info))) { + if ((!locked || !kvm_gpc_check(gpc, sizeof(*vcpu_info)))) { if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel)) kick_vcpu = true; goto out; @@ -1737,7 +1744,7 @@ static bool set_vcpu_info_evtchn_pending(struct kvm_vcpu *vcpu, u32 port) struct compat_vcpu_info *vcpu_info = gpc->khva; int port_word_bit = port / 32;

- if (!kvm_gpc_check(gpc, sizeof(*vcpu_info))) { + if ((!locked || !kvm_gpc_check(gpc, sizeof(*vcpu_info)))) { if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel)) kick_vcpu = true; goto out; @@ -1756,7 +1763,10 @@ static bool set_vcpu_info_evtchn_pending(struct kvm_vcpu *vcpu, u32 port) }

out: - read_unlock_irqrestore(&gpc->lock, flags); + if (locked) + read_unlock(&gpc->lock); + + local_irq_restore(flags); return kick_vcpu; }

From: David Woodhouse dwmw@amazon.co.uk

This function can race with kvm_gpc_deactivate(), which does not take the ->refresh_lock. This means kvm_gpc_deactivate() can wipe the ->pfn and ->khva fields, and unmap the latter, while hva_to_pfn_retry() has temporarily dropped its write lock on gpc->lock.

Then if hva_to_pfn_retry() determines that the PFN hasn't changed and that the original pfn and khva can be reused, they get assigned back to gpc->pfn and gpc->khva even though the khva was already unmapped by kvm_gpc_deactivate(). This leaves the cache in an apparently valid state but with ->khva pointing to an address which has been unmapped. Which in turn leads to oopses in e.g. __kvm_xen_has_interrupt() and set_shinfo_evtchn_pending() when they dereference said khva.

It may be possible to fix this just by making kvm_gpc_deactivate() take the ->refresh_lock, but that still leaves ->refresh_lock being basically redundant with the write lock on ->lock, which frankly makes my skin itch, with the way that pfn_to_hva_retry() operates on fields in the gpc without holding a write lock on ->lock.

Instead, fix it by cleaning up the semantics of hva_to_pfn_retry(). It now no longer does locking gymnastics because it no longer operates on the gpc object at all. I's called with a uhva and simply returns the corresponding pfn (pinned), and a mapped khva for it.

Its caller __kvm_gpc_refresh() now sets gpc->uhva and clears gpc->valid before dropping ->lock, calling hva_to_pfn_retry() and retaking ->lock for write.

If hva_to_pfn_retry() fails, *or* if the ->uhva or ->active fields in the gpc changed while the lock was dropped, the new mapping is discarded and the gpc is not modified. On success with an unchanged gpc, the new mapping is installed and the current ->pfn and ->uhva are taken into the local old_pfn and old_khva variables to be unmapped once the locks are all released.

This simplification means that ->refresh_lock is no longer needed for correctness, but it does still provide a minor optimisation because it will prevent two concurrent __kvm_gpc_refresh() calls from mapping a given PFN, only for one of them to lose the race and discard its mapping.

The optimisation in hva_to_pfn_retry() where it attempts to use the old mapping if the pfn doesn't change is dropped, since it makes the pinning more complex. It's a pointless optimisation anyway, since the odds of the pfn ending up the same when the uhva has changed (i.e. the odds of the two userspace addresses both pointing to the same underlying physical page) are negligible,

The 'hva_changed' local variable in __kvm_gpc_refresh() is also removed, since it's simpler just to clear gpc->valid if the uhva changed. Likewise the unmap_old variable is dropped because it's just as easy to check the old_pfn variable for KVM_PFN_ERR_FAULT.

I remain slightly confused because although this is clearly a race in the gfn_to_pfn_cache code, I don't quite know how the Xen support code actually managed to trigger it. We've seen oopses from dereferencing a valid-looking ->khva in both __kvm_xen_has_interrupt() (the vcpu_info) and in set_shinfo_evtchn_pending() (the shared_info). But surely the race shouldn't happen for the vcpu_info gpc because all calls to both refresh and deactivate hold the vcpu mutex, and it shouldn't happen for the shared_info gpc because all calls to both will hold the kvm->arch.xen.xen_lock mutex.

Signed-off-by: David Woodhouse dwmw@amazon.co.uk Reviewed-by: Paul Durrant paul@xen.org --- Cc: Sean Christopherson seanjc@google.com Cc: Paolo Bonzini pbonzini@redhat.com

v12: - New in this version. --- virt/kvm/pfncache.c | 184 +++++++++++++++++++++----------------------- 1 file changed, 88 insertions(+), 96 deletions(-)

diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c index 70394d7c9a38..4863f9f3d369 100644 --- a/virt/kvm/pfncache.c +++ b/virt/kvm/pfncache.c @@ -135,110 +135,67 @@ static inline bool mmu_notifier_retry_cache(struct kvm *kvm, unsigned long mmu_s return kvm->mmu_invalidate_seq != mmu_seq; }

-static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc) +/* + * Given a user virtual address, obtain a pinned host PFN and kernel mapping + * for it. The caller will release the PFN after installing it into the GPC + * so that the MMU notifier invalidation mechanism is active. + */ +static kvm_pfn_t hva_to_pfn_retry(struct kvm *kvm, unsigned long uhva, + kvm_pfn_t *pfn, void **khva) { /* Note, the new page offset may be different than the old! */ - void *old_khva = (void *)PAGE_ALIGN_DOWN((uintptr_t)gpc->khva); kvm_pfn_t new_pfn = KVM_PFN_ERR_FAULT; void *new_khva = NULL; unsigned long mmu_seq;

- lockdep_assert_held(&gpc->refresh_lock); - - lockdep_assert_held_write(&gpc->lock); - - /* - * Invalidate the cache prior to dropping gpc->lock, the gpa=>uhva - * assets have already been updated and so a concurrent check() from a - * different task may not fail the gpa/uhva/generation checks. - */ - gpc->valid = false; - - do { - mmu_seq = gpc->kvm->mmu_invalidate_seq; + for (;;) { + mmu_seq = kvm->mmu_invalidate_seq; smp_rmb();

- write_unlock_irq(&gpc->lock); - - /* - * If the previous iteration "failed" due to an mmu_notifier - * event, release the pfn and unmap the kernel virtual address - * from the previous attempt. Unmapping might sleep, so this - * needs to be done after dropping the lock. Opportunistically - * check for resched while the lock isn't held. - */ - if (new_pfn != KVM_PFN_ERR_FAULT) { - /* - * Keep the mapping if the previous iteration reused - * the existing mapping and didn't create a new one. - */ - if (new_khva != old_khva) - gpc_unmap(new_pfn, new_khva); - - kvm_release_pfn_clean(new_pfn); - - cond_resched(); - } - /* We always request a writeable mapping */ - new_pfn = hva_to_pfn(gpc->uhva, false, false, NULL, true, NULL); + new_pfn = hva_to_pfn(uhva, false, false, NULL, true, NULL); if (is_error_noslot_pfn(new_pfn)) - goto out_error; + return -EFAULT;

/* - * Obtain a new kernel mapping if KVM itself will access the - * pfn. Note, kmap() and memremap() can both sleep, so this - * too must be done outside of gpc->lock! + * Always obtain a new kernel mapping. Trying to reuse an + * existing one is more complex than it's worth. */ - if (new_pfn == gpc->pfn) - new_khva = old_khva; - else - new_khva = gpc_map(new_pfn); - + new_khva = gpc_map(new_pfn); if (!new_khva) { kvm_release_pfn_clean(new_pfn); - goto out_error; + return -EFAULT; }

- write_lock_irq(&gpc->lock); + if (!mmu_notifier_retry_cache(kvm, mmu_seq)) + break;

/* - * Other tasks must wait for _this_ refresh to complete before - * attempting to refresh. + * If this iteration "failed" due to an mmu_notifier event, + * release the pfn and unmap the kernel virtual address, and + * loop around again. */ - WARN_ON_ONCE(gpc->valid); - } while (mmu_notifier_retry_cache(gpc->kvm, mmu_seq)); - - gpc->valid = true; - gpc->pfn = new_pfn; - gpc->khva = new_khva + offset_in_page(gpc->uhva); + if (new_pfn != KVM_PFN_ERR_FAULT) { + gpc_unmap(new_pfn, new_khva); + kvm_release_pfn_clean(new_pfn); + } + }

- /* - * Put the reference to the _new_ pfn. The pfn is now tracked by the - * cache and can be safely migrated, swapped, etc... as the cache will - * invalidate any mappings in response to relevant mmu_notifier events. - */ - kvm_release_pfn_clean(new_pfn); + *pfn = new_pfn; + *khva = new_khva;

return 0; - -out_error: - write_lock_irq(&gpc->lock); - - return -EFAULT; }

-static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long uhva, - unsigned long len) +static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, + unsigned long uhva, unsigned long len) { struct kvm_memslots *slots = kvm_memslots(gpc->kvm); unsigned long page_offset = (gpa != KVM_XEN_INVALID_GPA) ? offset_in_page(gpa) : offset_in_page(uhva); - bool unmap_old = false; unsigned long old_uhva; - kvm_pfn_t old_pfn; - bool hva_change = false; + kvm_pfn_t old_pfn = KVM_PFN_ERR_FAULT; void *old_khva; int ret;

@@ -251,8 +208,9 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l

/* * If another task is refreshing the cache, wait for it to complete. - * There is no guarantee that concurrent refreshes will see the same - * gpa, memslots generation, etc..., so they must be fully serialized. + * This is purely an optimisation, to avoid concurrent mappings from + * hva_to_pfn_retry(), all but one of which will be discarded after + * losing a race to install them in the GPC. */ mutex_lock(&gpc->refresh_lock);

@@ -272,7 +230,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l gpc->uhva = PAGE_ALIGN_DOWN(uhva);

if (gpc->uhva != old_uhva) - hva_change = true; + gpc->valid = false; } else if (gpc->gpa != gpa || gpc->generation != slots->generation || kvm_is_error_hva(gpc->uhva)) { @@ -285,7 +243,11 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l

if (kvm_is_error_hva(gpc->uhva)) { ret = -EFAULT; - goto out; + + gpc->valid = false; + gpc->pfn = KVM_PFN_ERR_FAULT; + gpc->khva = NULL; + goto out_unlock; }

/* @@ -293,7 +255,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l * HVA may still be the same. */ if (gpc->uhva != old_uhva) - hva_change = true; + gpc->valid = false; } else { gpc->uhva = old_uhva; } @@ -305,9 +267,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l * If the userspace HVA changed or the PFN was already invalid, * drop the lock and do the HVA to PFN lookup again. */ - if (!gpc->valid || hva_change) { - ret = hva_to_pfn_retry(gpc); - } else { + if (gpc->valid) { /* * If the HVA→PFN mapping was already valid, don't unmap it. * But do update gpc->khva because the offset within the page @@ -315,30 +275,62 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l */ gpc->khva = old_khva + page_offset; ret = 0; - goto out_unlock; - }

- out: - /* - * Invalidate the cache and purge the pfn/khva if the refresh failed. - * Some/all of the uhva, gpa, and memslot generation info may still be - * valid, leave it as is. - */ - if (ret) { + /* old_pfn must not be unmapped because it was reused. */ + old_pfn = KVM_PFN_ERR_FAULT; + } else { + kvm_pfn_t new_pfn = KVM_PFN_ERR_FAULT; + unsigned long new_uhva = gpc->uhva; + void *new_khva = NULL; + + /* + * Invalidate the cache prior to dropping gpc->lock; the + * gpa=>uhva assets have already been updated and so a + * concurrent check() from a different task may not fail + * the gpa/uhva/generation checks as it should. + */ gpc->valid = false; - gpc->pfn = KVM_PFN_ERR_FAULT; - gpc->khva = NULL; - }

- /* Detect a pfn change before dropping the lock! */ - unmap_old = (old_pfn != gpc->pfn); + write_unlock_irq(&gpc->lock); + + ret = hva_to_pfn_retry(gpc->kvm, new_uhva, &new_pfn, &new_khva); + + write_lock_irq(&gpc->lock); + + WARN_ON_ONCE(gpc->valid); + + if (ret || !gpc->active || gpc->uhva != new_uhva) { + /* + * On failure or if another change occurred while the + * lock was dropped, just purge the new mapping. + */ + old_pfn = new_pfn; + old_khva = new_khva; + } else { + old_pfn = gpc->pfn; + old_khva = gpc->khva; + + gpc->pfn = new_pfn; + gpc->khva = new_khva + offset_in_page(gpc->uhva); + gpc->valid = true; + } + + /* + * Put the reference to the _new_ pfn. On success, the + * pfn is now tracked by the cache and can safely be + * migrated, swapped, etc. as the cache will invalidate + * any mappings in response to relevant mmu_notifier + * events. + */ + kvm_release_pfn_clean(new_pfn); + }

out_unlock: write_unlock_irq(&gpc->lock);

mutex_unlock(&gpc->refresh_lock);

- if (unmap_old) + if (old_pfn != KVM_PFN_ERR_FAULT) gpc_unmap(old_pfn, old_khva);

return ret;