1 files changed, 102 insertions, 3 deletions

diff --git a/arch/powerpc/kvm/book3s_hv_builtin.c b/arch/powerpc/kvm/book3s_hv_builtin.c
index a71e2fc..18afe65 100644
--- a/arch/powerpc/kvm/book3s_hv_builtin.c
+++ b/arch/powerpc/kvm/book3s_hv_builtin.c
@@ -53,11 +53,109 @@ EXPORT_SYMBOL_GPL(__xive_vm_h_eoi);
 
 /*
  * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
- * should be power of 2.
+ * only needs to be 256kB.
  */
-#define HPT_ALIGN_PAGES		((1 << 18) >> PAGE_SHIFT) /* 256k */
+#define HPT_ALIGN_ORDER		18		/* 256k */
+#define HPT_ALIGN_PAGES		((1 << HPT_ALIGN_ORDER) >> PAGE_SHIFT)
+
+#define KVM_RESV_CHUNK_ORDER	HPT_ALIGN_ORDER
+
 /*
- * By default we reserve 5% of memory for hash pagetable allocation.
+ * By default we reserve 2% of memory exclusively for guest HPT
+ * allocations, plus another 3% in the CMA zone which can be used
+ * either for HPTs or for movable page allocations.
+ * Each guest's HPT will be sized at between 1/128 and 1/64 of its
+ * memory, i.e. up to 1.56%, and allowing for about a 3x memory
+ * overcommit factor gets us to about 5%.
+ */
+static unsigned long kvm_hpt_resv_ratio = 2;
+
+static int __init early_parse_kvm_hpt_resv(char *p)
+{
+	pr_debug("%s(%s)\n", __func__, p);
+	if (!p)
+		return -EINVAL;
+	return kstrtoul(p, 0, &kvm_hpt_resv_ratio);
+}
+early_param("kvm_hpt_resv_ratio", early_parse_kvm_hpt_resv);
+
+static unsigned long kvm_resv_addr;
+static unsigned long *kvm_resv_bitmap;
+static unsigned long kvm_resv_chunks;
+static DEFINE_MUTEX(kvm_resv_lock);
+
+void kvm_resv_hpt_init(void)
+{
+	unsigned long align = 1ul << KVM_RESV_CHUNK_ORDER;
+	unsigned long size, bm_size;
+	unsigned long addr, bm;
+	unsigned long *bmp;
+
+	if (!cpu_has_feature(CPU_FTR_HVMODE))
+		return;
+
+	size = memblock_phys_mem_size() * kvm_hpt_resv_ratio / 100;
+	size = ALIGN(size, align);
+	if (!size)
+		return;
+
+	pr_info("KVM: Allocating %lu MiB for hashed page tables\n",
+		size >> 20);
+
+	addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
+	if (!addr) {
+		pr_err("KVM: Allocation of reserved memory for HPTs failed\n");
+		return;
+	}
+	pr_info("KVM: %lu MiB reserved for HPTs at %lx\n", size >> 20, addr);
+
+	bm_size = BITS_TO_LONGS(size >> KVM_RESV_CHUNK_ORDER) * sizeof(long);
+	bm = __memblock_alloc_base(bm_size, sizeof(long),
+				   MEMBLOCK_ALLOC_ACCESSIBLE);
+	if (!bm) {
+		pr_err("KVM: Allocation of reserved memory bitmap failed\n");
+		return;
+	}
+	bmp = __va(bm);
+	memset(bmp, 0, bm_size);
+
+	kvm_resv_addr = (unsigned long) __va(addr);
+	kvm_resv_chunks = size >> KVM_RESV_CHUNK_ORDER;
+	kvm_resv_bitmap = bmp;
+}
+
+unsigned long kvmhv_alloc_resv_hpt(u32 order)
+{
+	unsigned long nr_chunks = 1ul << (order - KVM_RESV_CHUNK_ORDER);
+	unsigned long chunk;
+
+	mutex_lock(&kvm_resv_lock);
+	chunk = bitmap_find_next_zero_area(kvm_resv_bitmap, kvm_resv_chunks,
+					   0, nr_chunks, 0);
+	if (chunk < kvm_resv_chunks)
+		bitmap_set(kvm_resv_bitmap, chunk, nr_chunks);
+	mutex_unlock(&kvm_resv_lock);
+
+	if (chunk < kvm_resv_chunks)
+		return kvm_resv_addr + (chunk << KVM_RESV_CHUNK_ORDER);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(kvmhv_alloc_resv_hpt);
+
+void kvmhv_release_resv_hpt(unsigned long addr, u32 order)
+{
+	unsigned long nr_chunks = 1ul << (order - KVM_RESV_CHUNK_ORDER);
+	unsigned long chunk = (addr - kvm_resv_addr) >> KVM_RESV_CHUNK_ORDER;
+
+	mutex_lock(&kvm_resv_lock);
+	if (chunk + nr_chunks <= kvm_resv_chunks)
+		bitmap_clear(kvm_resv_bitmap, chunk, nr_chunks);
+	mutex_unlock(&kvm_resv_lock);
+}
+EXPORT_SYMBOL_GPL(kvmhv_release_resv_hpt);
+
+/*
+ * By default we reserve 3% of memory for the CMA zone.
  */
 static unsigned long kvm_cma_resv_ratio = 5;
 
@@ -106,6 +204,7 @@ void __init kvm_cma_reserve(void)
 	 */
 	if (!cpu_has_feature(CPU_FTR_HVMODE))
 		return;
+
 	/*
 	 * We cannot use memblock_phys_mem_size() here, because
 	 * memblock_analyze() has not been called yet.