KVM: Reinstate gfn_to_pfn_cache with invalidation support

This can be used in two modes. There is an atomic mode where the cached
mapping is accessed while holding the rwlock, and a mode where the
physical address is used by a vCPU in guest mode.

For the latter case, an invalidation will wake the vCPU with the new
KVM_REQ_GPC_INVALIDATE, and the architecture will need to refresh any
caches it still needs to access before entering guest mode again.

Only one vCPU can be targeted by the wake requests; it's simple enough
to make it wake all vCPUs or even a mask but I don't see a use case for
that additional complexity right now.

Invalidation happens from the invalidate_range_start MMU notifier, which
needs to be able to sleep in order to wake the vCPU and wait for it.

This means that revalidation potentially needs to "wait" for the MMU
operation to complete and the invalidate_range_end notifier to be
invoked. Like the vCPU when it takes a page fault in that period, we
just spin — fixing that in a future patch by implementing an actual
*wait* may be another part of shaving this particularly hirsute yak.

As noted in the comments in the function itself, the only case where
the invalidate_range_start notifier is expected to be called *without*
being able to sleep is when the OOM reaper is killing the process. In
that case, we expect the vCPU threads already to have exited, and thus
there will be nothing to wake, and no reason to wait. So we clear the
KVM_REQUEST_WAIT bit and send the request anyway, then complain loudly
if there actually *was* anything to wake up.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20211210163625.2886-3-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

7 files changed, 395 insertions, 27 deletions

diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
index 97cf5413ac25..f4834c20e4a6 100644
--- a/virt/kvm/Kconfig
+++ b/virt/kvm/Kconfig
@@ -4,6 +4,9 @@
 config HAVE_KVM
        bool
 
+config HAVE_KVM_PFNCACHE
+       bool
+
 config HAVE_KVM_IRQCHIP
        bool
 
diff --git a/virt/kvm/Makefile.kvm b/virt/kvm/Makefile.kvm
index ffdcad3cc97a..2c27d5d0c367 100644
--- a/virt/kvm/Makefile.kvm
+++ b/virt/kvm/Makefile.kvm
@@ -11,3 +11,4 @@ kvm-$(CONFIG_KVM_MMIO) += $(KVM)/coalesced_mmio.o
 kvm-$(CONFIG_KVM_ASYNC_PF) += $(KVM)/async_pf.o
 kvm-$(CONFIG_HAVE_KVM_IRQ_ROUTING) += $(KVM)/irqchip.o
 kvm-$(CONFIG_HAVE_KVM_DIRTY_RING) += $(KVM)/dirty_ring.o
+kvm-$(CONFIG_HAVE_KVM_PFNCACHE) += $(KVM)/pfncache.o
diff --git a/virt/kvm/dirty_ring.c b/virt/kvm/dirty_ring.c
index 8e9874760fb3..222ecc81d7df 100644
--- a/virt/kvm/dirty_ring.c
+++ b/virt/kvm/dirty_ring.c
@@ -9,7 +9,7 @@
 #include <linux/vmalloc.h>
 #include <linux/kvm_dirty_ring.h>
 #include <trace/events/kvm.h>
-#include "mmu_lock.h"
+#include "kvm_mm.h"
 
 int __weak kvm_cpu_dirty_log_size(void)
 {
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index af5b4427b139..6e8e9d36f382 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -59,7 +59,7 @@
 
 #include "coalesced_mmio.h"
 #include "async_pf.h"
-#include "mmu_lock.h"
+#include "kvm_mm.h"
 #include "vfio.h"
 
 #define CREATE_TRACE_POINTS
@@ -711,6 +711,9 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 	kvm->mn_active_invalidate_count++;
 	spin_unlock(&kvm->mn_invalidate_lock);
 
+	gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
+					  hva_range.may_block);
+
 	__kvm_handle_hva_range(kvm, &hva_range);
 
 	return 0;
@@ -1071,6 +1074,9 @@ static struct kvm *kvm_create_vm(unsigned long type)
 	rcuwait_init(&kvm->mn_memslots_update_rcuwait);
 	xa_init(&kvm->vcpu_array);
 
+	INIT_LIST_HEAD(&kvm->gpc_list);
+	spin_lock_init(&kvm->gpc_lock);
+
 	INIT_LIST_HEAD(&kvm->devices);
 
 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
@@ -2539,8 +2545,8 @@ out:
  * 2): @write_fault = false && @writable, @writable will tell the caller
  *     whether the mapping is writable.
  */
-static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
-			bool write_fault, bool *writable)
+kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
+		     bool write_fault, bool *writable)
 {
 	struct vm_area_struct *vma;
 	kvm_pfn_t pfn = 0;
diff --git a/virt/kvm/kvm_mm.h b/virt/kvm/kvm_mm.h
new file mode 100644
index 000000000000..34ca40823260
--- /dev/null
+++ b/virt/kvm/kvm_mm.h
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#ifndef __KVM_MM_H__
+#define __KVM_MM_H__ 1
+
+/*
+ * Architectures can choose whether to use an rwlock or spinlock
+ * for the mmu_lock.  These macros, for use in common code
+ * only, avoids using #ifdefs in places that must deal with
+ * multiple architectures.
+ */
+
+#ifdef KVM_HAVE_MMU_RWLOCK
+#define KVM_MMU_LOCK_INIT(kvm)		rwlock_init(&(kvm)->mmu_lock)
+#define KVM_MMU_LOCK(kvm)		write_lock(&(kvm)->mmu_lock)
+#define KVM_MMU_UNLOCK(kvm)		write_unlock(&(kvm)->mmu_lock)
+#define KVM_MMU_READ_LOCK(kvm)		read_lock(&(kvm)->mmu_lock)
+#define KVM_MMU_READ_UNLOCK(kvm)	read_unlock(&(kvm)->mmu_lock)
+#else
+#define KVM_MMU_LOCK_INIT(kvm)		spin_lock_init(&(kvm)->mmu_lock)
+#define KVM_MMU_LOCK(kvm)		spin_lock(&(kvm)->mmu_lock)
+#define KVM_MMU_UNLOCK(kvm)		spin_unlock(&(kvm)->mmu_lock)
+#define KVM_MMU_READ_LOCK(kvm)		spin_lock(&(kvm)->mmu_lock)
+#define KVM_MMU_READ_UNLOCK(kvm)	spin_unlock(&(kvm)->mmu_lock)
+#endif /* KVM_HAVE_MMU_RWLOCK */
+
+kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
+		     bool write_fault, bool *writable);
+
+#ifdef CONFIG_HAVE_KVM_PFNCACHE
+void gfn_to_pfn_cache_invalidate_start(struct kvm *kvm,
+				       unsigned long start,
+				       unsigned long end,
+				       bool may_block);
+#else
+static inline void gfn_to_pfn_cache_invalidate_start(struct kvm *kvm,
+						     unsigned long start,
+						     unsigned long end,
+						     bool may_block)
+{
+}
+#endif /* HAVE_KVM_PFNCACHE */
+
+#endif /* __KVM_MM_H__ */
diff --git a/virt/kvm/mmu_lock.h b/virt/kvm/mmu_lock.h
deleted file mode 100644
index 9e1308f9734c..000000000000
--- a/virt/kvm/mmu_lock.h
+++ /dev/null
@@ -1,23 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-only
-
-#ifndef KVM_MMU_LOCK_H
-#define KVM_MMU_LOCK_H 1
-
-/*
- * Architectures can choose whether to use an rwlock or spinlock
- * for the mmu_lock.  These macros, for use in common code
- * only, avoids using #ifdefs in places that must deal with
- * multiple architectures.
- */
-
-#ifdef KVM_HAVE_MMU_RWLOCK
-#define KVM_MMU_LOCK_INIT(kvm) rwlock_init(&(kvm)->mmu_lock)
-#define KVM_MMU_LOCK(kvm)      write_lock(&(kvm)->mmu_lock)
-#define KVM_MMU_UNLOCK(kvm)    write_unlock(&(kvm)->mmu_lock)
-#else
-#define KVM_MMU_LOCK_INIT(kvm) spin_lock_init(&(kvm)->mmu_lock)
-#define KVM_MMU_LOCK(kvm)      spin_lock(&(kvm)->mmu_lock)
-#define KVM_MMU_UNLOCK(kvm)    spin_unlock(&(kvm)->mmu_lock)
-#endif /* KVM_HAVE_MMU_RWLOCK */
-
-#endif
diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
new file mode 100644
index 000000000000..ce878f4be4da
--- /dev/null
+++ b/virt/kvm/pfncache.c
@@ -0,0 +1,337 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Kernel-based Virtual Machine driver for Linux
+ *
+ * This module enables kernel and guest-mode vCPU access to guest physical
+ * memory with suitable invalidation mechanisms.
+ *
+ * Copyright © 2021 Amazon.com, Inc. or its affiliates.
+ *
+ * Authors:
+ *   David Woodhouse <dwmw2@infradead.org>
+ */
+
+#include <linux/kvm_host.h>
+#include <linux/kvm.h>
+#include <linux/highmem.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+
+#include "kvm_mm.h"
+
+/*
+ * MMU notifier 'invalidate_range_start' hook.
+ */
+void gfn_to_pfn_cache_invalidate_start(struct kvm *kvm, unsigned long start,
+				       unsigned long end, bool may_block)
+{
+	DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
+	struct gfn_to_pfn_cache *gpc;
+	bool wake_vcpus = false;
+
+	spin_lock(&kvm->gpc_lock);
+	list_for_each_entry(gpc, &kvm->gpc_list, list) {
+		write_lock_irq(&gpc->lock);
+
+		/* Only a single page so no need to care about length */
+		if (gpc->valid && !is_error_noslot_pfn(gpc->pfn) &&
+		    gpc->uhva >= start && gpc->uhva < end) {
+			gpc->valid = false;
+
+			/*
+			 * If a guest vCPU could be using the physical address,
+			 * it needs to be woken.
+			 */
+			if (gpc->guest_uses_pa) {
+				if (!wake_vcpus) {
+					wake_vcpus = true;
+					bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
+				}
+				__set_bit(gpc->vcpu->vcpu_idx, vcpu_bitmap);
+			}
+
+			/*
+			 * We cannot call mark_page_dirty() from here because
+			 * this physical CPU might not have an active vCPU
+			 * with which to do the KVM dirty tracking.
+			 *
+			 * Neither is there any point in telling the kernel MM
+			 * that the underlying page is dirty. A vCPU in guest
+			 * mode might still be writing to it up to the point
+			 * where we wake them a few lines further down anyway.
+			 *
+			 * So all the dirty marking happens on the unmap.
+			 */
+		}
+		write_unlock_irq(&gpc->lock);
+	}
+	spin_unlock(&kvm->gpc_lock);
+
+	if (wake_vcpus) {
+		unsigned int req = KVM_REQ_GPC_INVALIDATE;
+		bool called;
+
+		/*
+		 * If the OOM reaper is active, then all vCPUs should have
+		 * been stopped already, so perform the request without
+		 * KVM_REQUEST_WAIT and be sad if any needed to be woken.
+		 */
+		if (!may_block)
+			req &= ~KVM_REQUEST_WAIT;
+
+		called = kvm_make_vcpus_request_mask(kvm, req, vcpu_bitmap);
+
+		WARN_ON_ONCE(called && !may_block);
+	}
+}
+
+bool kvm_gfn_to_pfn_cache_check(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
+				gpa_t gpa, unsigned long len)
+{
+	struct kvm_memslots *slots = kvm_memslots(kvm);
+
+	if ((gpa & ~PAGE_MASK) + len > PAGE_SIZE)
+		return false;
+
+	if (gpc->gpa != gpa || gpc->generation != slots->generation ||
+	    kvm_is_error_hva(gpc->uhva))
+		return false;
+
+	if (!gpc->valid)
+		return false;
+
+	return true;
+}
+EXPORT_SYMBOL_GPL(kvm_gfn_to_pfn_cache_check);
+
+static void __release_gpc(struct kvm *kvm, kvm_pfn_t pfn, void *khva,
+			  gpa_t gpa, bool dirty)
+{
+	/* Unmap the old page if it was mapped before, and release it */
+	if (!is_error_noslot_pfn(pfn)) {
+		if (khva) {
+			if (pfn_valid(pfn))
+				kunmap(pfn_to_page(pfn));
+#ifdef CONFIG_HAS_IOMEM
+			else
+				memunmap(khva);
+#endif
+		}
+
+		kvm_release_pfn(pfn, dirty);
+		if (dirty)
+			mark_page_dirty(kvm, gpa);
+	}
+}
+
+static kvm_pfn_t hva_to_pfn_retry(struct kvm *kvm, unsigned long uhva)
+{
+	unsigned long mmu_seq;
+	kvm_pfn_t new_pfn;
+	int retry;
+
+	do {
+		mmu_seq = kvm->mmu_notifier_seq;
+		smp_rmb();
+
+		/* We always request a writeable mapping */
+		new_pfn = hva_to_pfn(uhva, false, NULL, true, NULL);
+		if (is_error_noslot_pfn(new_pfn))
+			break;
+
+		KVM_MMU_READ_LOCK(kvm);
+		retry = mmu_notifier_retry_hva(kvm, mmu_seq, uhva);
+		KVM_MMU_READ_UNLOCK(kvm);
+		if (!retry)
+			break;
+
+		cond_resched();
+	} while (1);
+
+	return new_pfn;
+}
+
+int kvm_gfn_to_pfn_cache_refresh(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
+				 gpa_t gpa, unsigned long len, bool dirty)
+{
+	struct kvm_memslots *slots = kvm_memslots(kvm);
+	unsigned long page_offset = gpa & ~PAGE_MASK;
+	kvm_pfn_t old_pfn, new_pfn;
+	unsigned long old_uhva;
+	gpa_t old_gpa;
+	void *old_khva;
+	bool old_valid, old_dirty;
+	int ret = 0;
+
+	/*
+	 * If must fit within a single page. The 'len' argument is
+	 * only to enforce that.
+	 */
+	if (page_offset + len > PAGE_SIZE)
+		return -EINVAL;
+
+	write_lock_irq(&gpc->lock);
+
+	old_gpa = gpc->gpa;
+	old_pfn = gpc->pfn;
+	old_khva = gpc->khva - offset_in_page(gpc->khva);
+	old_uhva = gpc->uhva;
+	old_valid = gpc->valid;
+	old_dirty = gpc->dirty;
+
+	/* If the userspace HVA is invalid, refresh that first */
+	if (gpc->gpa != gpa || gpc->generation != slots->generation ||
+	    kvm_is_error_hva(gpc->uhva)) {
+		gfn_t gfn = gpa_to_gfn(gpa);
+
+		gpc->dirty = false;
+		gpc->gpa = gpa;
+		gpc->generation = slots->generation;
+		gpc->memslot = __gfn_to_memslot(slots, gfn);
+		gpc->uhva = gfn_to_hva_memslot(gpc->memslot, gfn);
+
+		if (kvm_is_error_hva(gpc->uhva)) {
+			ret = -EFAULT;
+			goto out;
+		}
+
+		gpc->uhva += page_offset;
+	}
+
+	/*
+	 * If the userspace HVA changed or the PFN was already invalid,
+	 * drop the lock and do the HVA to PFN lookup again.
+	 */
+	if (!old_valid || old_uhva != gpc->uhva) {
+		unsigned long uhva = gpc->uhva;
+		void *new_khva = NULL;
+
+		/* Placeholders for "hva is valid but not yet mapped" */
+		gpc->pfn = KVM_PFN_ERR_FAULT;
+		gpc->khva = NULL;
+		gpc->valid = true;
+
+		write_unlock_irq(&gpc->lock);
+
+		new_pfn = hva_to_pfn_retry(kvm, uhva);
+		if (is_error_noslot_pfn(new_pfn)) {
+			ret = -EFAULT;
+			goto map_done;
+		}
+
+		if (gpc->kernel_map) {
+			if (new_pfn == old_pfn) {
+				new_khva = old_khva;
+				old_pfn = KVM_PFN_ERR_FAULT;
+				old_khva = NULL;
+			} else if (pfn_valid(new_pfn)) {
+				new_khva = kmap(pfn_to_page(new_pfn));
+#ifdef CONFIG_HAS_IOMEM
+			} else {
+				new_khva = memremap(pfn_to_hpa(new_pfn), PAGE_SIZE, MEMREMAP_WB);
+#endif
+			}
+			if (new_khva)
+				new_khva += page_offset;
+			else
+				ret = -EFAULT;
+		}
+
+	map_done:
+		write_lock_irq(&gpc->lock);
+		if (ret) {
+			gpc->valid = false;
+			gpc->pfn = KVM_PFN_ERR_FAULT;
+			gpc->khva = NULL;
+		} else {
+			/* At this point, gpc->valid may already have been cleared */
+			gpc->pfn = new_pfn;
+			gpc->khva = new_khva;
+		}
+	} else {
+		/* If the HVA→PFN mapping was already valid, don't unmap it. */
+		old_pfn = KVM_PFN_ERR_FAULT;
+		old_khva = NULL;
+	}
+
+ out:
+	if (ret)
+		gpc->dirty = false;
+	else
+		gpc->dirty = dirty;
+
+	write_unlock_irq(&gpc->lock);
+
+	__release_gpc(kvm, old_pfn, old_khva, old_gpa, old_dirty);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(kvm_gfn_to_pfn_cache_refresh);
+
+void kvm_gfn_to_pfn_cache_unmap(struct kvm *kvm, struct gfn_to_pfn_cache *gpc)
+{
+	void *old_khva;
+	kvm_pfn_t old_pfn;
+	bool old_dirty;
+	gpa_t old_gpa;
+
+	write_lock_irq(&gpc->lock);
+
+	gpc->valid = false;
+
+	old_khva = gpc->khva - offset_in_page(gpc->khva);
+	old_dirty = gpc->dirty;
+	old_gpa = gpc->gpa;
+	old_pfn = gpc->pfn;
+
+	/*
+	 * We can leave the GPA → uHVA map cache intact but the PFN
+	 * lookup will need to be redone even for the same page.
+	 */
+	gpc->khva = NULL;
+	gpc->pfn = KVM_PFN_ERR_FAULT;
+
+	write_unlock_irq(&gpc->lock);
+
+	__release_gpc(kvm, old_pfn, old_khva, old_gpa, old_dirty);
+}
+EXPORT_SYMBOL_GPL(kvm_gfn_to_pfn_cache_unmap);
+
+
+int kvm_gfn_to_pfn_cache_init(struct kvm *kvm, struct gfn_to_pfn_cache *gpc,
+			      struct kvm_vcpu *vcpu, bool guest_uses_pa,
+			      bool kernel_map, gpa_t gpa, unsigned long len,
+			      bool dirty)
+{
+	if (!gpc->active) {
+		rwlock_init(&gpc->lock);
+
+		gpc->khva = NULL;
+		gpc->pfn = KVM_PFN_ERR_FAULT;
+		gpc->uhva = KVM_HVA_ERR_BAD;
+		gpc->vcpu = vcpu;
+		gpc->kernel_map = kernel_map;
+		gpc->guest_uses_pa = guest_uses_pa;
+		gpc->valid = false;
+		gpc->active = true;
+
+		spin_lock(&kvm->gpc_lock);
+		list_add(&gpc->list, &kvm->gpc_list);
+		spin_unlock(&kvm->gpc_lock);
+	}
+	return kvm_gfn_to_pfn_cache_refresh(kvm, gpc, gpa, len, dirty);
+}
+EXPORT_SYMBOL_GPL(kvm_gfn_to_pfn_cache_init);
+
+void kvm_gfn_to_pfn_cache_destroy(struct kvm *kvm, struct gfn_to_pfn_cache *gpc)
+{
+	if (gpc->active) {
+		spin_lock(&kvm->gpc_lock);
+		list_del(&gpc->list);
+		spin_unlock(&kvm->gpc_lock);
+
+		kvm_gfn_to_pfn_cache_unmap(kvm, gpc);
+		gpc->active = false;
+	}
+}
+EXPORT_SYMBOL_GPL(kvm_gfn_to_pfn_cache_destroy);