arm64: KVM: Fix stage-2 PGD allocation to have per-page refcounting

We're using __get_free_pages with to allocate the guest's stage-2
PGD. The standard behaviour of this function is to return a set of
pages where only the head page has a valid refcount.

This behaviour gets us into trouble when we're trying to increment
the refcount on a non-head page:

page:ffff7c00cfb693c0 count:0 mapcount:0 mapping:          (null) index:0x0
flags: 0x4000000000000000()
page dumped because: VM_BUG_ON_PAGE((*({ __attribute__((unused)) typeof((&page->_count)->counter) __var = ( typeof((&page->_count)->counter)) 0; (volatile typeof((&page->_count)->counter) *)&((&page->_count)->counter); })) <= 0)
BUG: failure at include/linux/mm.h:548/get_page()!
Kernel panic - not syncing: BUG!
CPU: 1 PID: 1695 Comm: kvm-vcpu-0 Not tainted 4.0.0-rc1+ #3825
Hardware name: APM X-Gene Mustang board (DT)
Call trace:
[<ffff80000008a09c>] dump_backtrace+0x0/0x13c
[<ffff80000008a1e8>] show_stack+0x10/0x1c
[<ffff800000691da8>] dump_stack+0x74/0x94
[<ffff800000690d78>] panic+0x100/0x240
[<ffff8000000a0bc4>] stage2_get_pmd+0x17c/0x2bc
[<ffff8000000a1dc4>] kvm_handle_guest_abort+0x4b4/0x6b0
[<ffff8000000a420c>] handle_exit+0x58/0x180
[<ffff80000009e7a4>] kvm_arch_vcpu_ioctl_run+0x114/0x45c
[<ffff800000099df4>] kvm_vcpu_ioctl+0x2e0/0x754
[<ffff8000001c0a18>] do_vfs_ioctl+0x424/0x5c8
[<ffff8000001c0bfc>] SyS_ioctl+0x40/0x78
CPU0: stopping

A possible approach for this is to split the compound page using
split_page() at allocation time, and change the teardown path to
free one page at a time.  It turns out that alloc_pages_exact() and
free_pages_exact() does exactly that.

While we're at it, the PGD allocation code is reworked to reduce
duplication.

This has been tested on an X-Gene platform with a 4kB/48bit-VA host
kernel, and kvmtool hacked to place memory in the second page of
the hardware PGD (PUD for the host kernel). Also regression-tested
on a Cubietruck (Cortex-A7).

 [ Reworked to use alloc_pages_exact() and free_pages_exact() and to
   return pointers directly instead of by reference as arguments
    - Christoffer ]
Reported-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

arch/arm/include/asm/kvm_mmu.h

@@ -162,16 +162,14 @@ static inline bool kvm_page_empty(void *ptr)
 
 #define KVM_PREALLOC_LEVEL	0
 
-static inline int kvm_prealloc_hwpgd(struct kvm *kvm, pgd_t *pgd)
+static inline void *kvm_get_hwpgd(struct kvm *kvm)
 {
-	return 0;
+	return kvm->arch.pgd;
 }
 
-static inline void kvm_free_hwpgd(struct kvm *kvm) { }
-
-static inline void *kvm_get_hwpgd(struct kvm *kvm)
+static inline unsigned int kvm_get_hwpgd_size(void)
 {
-	return kvm->arch.pgd;
+	return PTRS_PER_S2_PGD * sizeof(pgd_t);
 }
 
 struct kvm;

arch/arm/kvm/mmu.c

@@ -632,6 +632,20 @@ int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr)
 				     __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE);
 }
 
+/* Free the HW pgd, one page at a time */
+static void kvm_free_hwpgd(void *hwpgd)
+{
+	free_pages_exact(hwpgd, kvm_get_hwpgd_size());
+}
+
+/* Allocate the HW PGD, making sure that each page gets its own refcount */
+static void *kvm_alloc_hwpgd(void)
+{
+	unsigned int size = kvm_get_hwpgd_size();
+
+	return alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
+}
+
 /**
  * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation.
  * @kvm:	The KVM struct pointer for the VM.
@@ -645,15 +659,31 @@ int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr)
  */
 int kvm_alloc_stage2_pgd(struct kvm *kvm)
 {
-	int ret;
 	pgd_t *pgd;
+	void *hwpgd;
 
 	if (kvm->arch.pgd != NULL) {
 		kvm_err("kvm_arch already initialized?\n");
 		return -EINVAL;
 	}
 
+	hwpgd = kvm_alloc_hwpgd();
+	if (!hwpgd)
+		return -ENOMEM;
+
+	/* When the kernel uses more levels of page tables than the
+	 * guest, we allocate a fake PGD and pre-populate it to point
+	 * to the next-level page table, which will be the real
+	 * initial page table pointed to by the VTTBR.
+	 *
+	 * When KVM_PREALLOC_LEVEL==2, we allocate a single page for
+	 * the PMD and the kernel will use folded pud.
+	 * When KVM_PREALLOC_LEVEL==1, we allocate 2 consecutive PUD
+	 * pages.
+	 */
 	if (KVM_PREALLOC_LEVEL > 0) {
+		int i;
+
 		/*
 		 * Allocate fake pgd for the page table manipulation macros to
 		 * work.  This is not used by the hardware and we have no
@@ -661,30 +691,32 @@ int kvm_alloc_stage2_pgd(struct kvm *kvm)
 		 */
 		pgd = (pgd_t *)kmalloc(PTRS_PER_S2_PGD * sizeof(pgd_t),
 				       GFP_KERNEL | __GFP_ZERO);
+
+		if (!pgd) {
+			kvm_free_hwpgd(hwpgd);
+			return -ENOMEM;
+		}
+
+		/* Plug the HW PGD into the fake one. */
+		for (i = 0; i < PTRS_PER_S2_PGD; i++) {
+			if (KVM_PREALLOC_LEVEL == 1)
+				pgd_populate(NULL, pgd + i,
+					     (pud_t *)hwpgd + i * PTRS_PER_PUD);
+			else if (KVM_PREALLOC_LEVEL == 2)
+				pud_populate(NULL, pud_offset(pgd, 0) + i,
+					     (pmd_t *)hwpgd + i * PTRS_PER_PMD);
+		}
 	} else {
 		/*
 		 * Allocate actual first-level Stage-2 page table used by the
 		 * hardware for Stage-2 page table walks.
 		 */
-		pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, S2_PGD_ORDER);
+		pgd = (pgd_t *)hwpgd;
 	}
 
-	if (!pgd)
-		return -ENOMEM;
-
-	ret = kvm_prealloc_hwpgd(kvm, pgd);
-	if (ret)
-		goto out_err;
-
 	kvm_clean_pgd(pgd);
 	kvm->arch.pgd = pgd;
 	return 0;
-out_err:
-	if (KVM_PREALLOC_LEVEL > 0)
-		kfree(pgd);
-	else
-		free_pages((unsigned long)pgd, S2_PGD_ORDER);
-	return ret;
 }
 
 /**
@@ -785,11 +817,10 @@ void kvm_free_stage2_pgd(struct kvm *kvm)
 		return;
 
 	unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
-	kvm_free_hwpgd(kvm);
+	kvm_free_hwpgd(kvm_get_hwpgd(kvm));
 	if (KVM_PREALLOC_LEVEL > 0)
 		kfree(kvm->arch.pgd);
-	else
-		free_pages((unsigned long)kvm->arch.pgd, S2_PGD_ORDER);
+
 	kvm->arch.pgd = NULL;
 }
 

arch/arm64/include/asm/kvm_mmu.h

@@ -171,43 +171,6 @@ static inline bool kvm_s2pmd_readonly(pmd_t *pmd)
 #define KVM_PREALLOC_LEVEL	(0)
 #endif
 
-/**
- * kvm_prealloc_hwpgd - allocate inital table for VTTBR
- * @kvm:	The KVM struct pointer for the VM.
- * @pgd:	The kernel pseudo pgd
- *
- * When the kernel uses more levels of page tables than the guest, we allocate
- * a fake PGD and pre-populate it to point to the next-level page table, which
- * will be the real initial page table pointed to by the VTTBR.
- *
- * When KVM_PREALLOC_LEVEL==2, we allocate a single page for the PMD and
- * the kernel will use folded pud.  When KVM_PREALLOC_LEVEL==1, we
- * allocate 2 consecutive PUD pages.
- */
-static inline int kvm_prealloc_hwpgd(struct kvm *kvm, pgd_t *pgd)
-{
-	unsigned int i;
-	unsigned long hwpgd;
-
-	if (KVM_PREALLOC_LEVEL == 0)
-		return 0;
-
-	hwpgd = __get_free_pages(GFP_KERNEL | __GFP_ZERO, PTRS_PER_S2_PGD_SHIFT);
-	if (!hwpgd)
-		return -ENOMEM;
-
-	for (i = 0; i < PTRS_PER_S2_PGD; i++) {
-		if (KVM_PREALLOC_LEVEL == 1)
-			pgd_populate(NULL, pgd + i,
-				     (pud_t *)hwpgd + i * PTRS_PER_PUD);
-		else if (KVM_PREALLOC_LEVEL == 2)
-			pud_populate(NULL, pud_offset(pgd, 0) + i,
-				     (pmd_t *)hwpgd + i * PTRS_PER_PMD);
-	}
-
-	return 0;
-}
-
 static inline void *kvm_get_hwpgd(struct kvm *kvm)
 {
 	pgd_t *pgd = kvm->arch.pgd;
@@ -224,12 +187,11 @@ static inline void *kvm_get_hwpgd(struct kvm *kvm)
 	return pmd_offset(pud, 0);
 }
 
-static inline void kvm_free_hwpgd(struct kvm *kvm)
+static inline unsigned int kvm_get_hwpgd_size(void)
 {
-	if (KVM_PREALLOC_LEVEL > 0) {
-		unsigned long hwpgd = (unsigned long)kvm_get_hwpgd(kvm);
-		free_pages(hwpgd, PTRS_PER_S2_PGD_SHIFT);
-	}
+	if (KVM_PREALLOC_LEVEL > 0)
+		return PTRS_PER_S2_PGD * PAGE_SIZE;
+	return PTRS_PER_S2_PGD * sizeof(pgd_t);
 }
 
 static inline bool kvm_page_empty(void *ptr)