powerpc/kvm: Contiguous memory allocator based hash page table allocation

Powerpc architecture uses a hash based page table mechanism for mapping virtual
addresses to physical address. The architecture require this hash page table to
be physically contiguous. With KVM on Powerpc currently we use early reservation
mechanism for allocating guest hash page table. This implies that we need to
reserve a big memory region to ensure we can create large number of guest
simultaneously with KVM on Power. Another disadvantage is that the reserved memory
is not available to rest of the subsystems and and that implies we limit the total
available memory in the host.

This patch series switch the guest hash page table allocation to use
contiguous memory allocator.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>

6 files changed, 332 insertions, 47 deletions

diff --git a/arch/powerpc/kvm/Kconfig b/arch/powerpc/kvm/Kconfig
index eb643f862579..ffaef2cb101a 100644
--- a/arch/powerpc/kvm/Kconfig
+++ b/arch/powerpc/kvm/Kconfig
@@ -72,6 +72,7 @@ config KVM_BOOK3S_64_HV
 	bool "KVM support for POWER7 and PPC970 using hypervisor mode in host"
 	depends on KVM_BOOK3S_64
 	select MMU_NOTIFIER
+	select CMA
 	---help---
 	  Support running unmodified book3s_64 guest kernels in
 	  virtual machines on POWER7 and PPC970 processors that have
diff --git a/arch/powerpc/kvm/Makefile b/arch/powerpc/kvm/Makefile
index 008cd856c5b5..6646c952c5e3 100644
--- a/arch/powerpc/kvm/Makefile
+++ b/arch/powerpc/kvm/Makefile
@@ -81,6 +81,7 @@ kvm-book3s_64-builtin-objs-$(CONFIG_KVM_BOOK3S_64_HV) := \
 	book3s_64_vio_hv.o \
 	book3s_hv_ras.o \
 	book3s_hv_builtin.o \
+	book3s_hv_cma.o \
 	$(kvm-book3s_64-builtin-xics-objs-y)
 
 kvm-book3s_64-objs-$(CONFIG_KVM_XICS) += \
diff --git a/arch/powerpc/kvm/book3s_64_mmu_hv.c b/arch/powerpc/kvm/book3s_64_mmu_hv.c
index 5880dfb31074..354f4bb21f5c 100644
--- a/arch/powerpc/kvm/book3s_64_mmu_hv.c
+++ b/arch/powerpc/kvm/book3s_64_mmu_hv.c
@@ -52,8 +52,8 @@ long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp)
 {
 	unsigned long hpt;
 	struct revmap_entry *rev;
-	struct kvmppc_linear_info *li;
-	long order = kvm_hpt_order;
+	struct page *page = NULL;
+	long order = KVM_DEFAULT_HPT_ORDER;
 
 	if (htab_orderp) {
 		order = *htab_orderp;
@@ -61,26 +61,22 @@ long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp)
 			order = PPC_MIN_HPT_ORDER;
 	}
 
+	kvm->arch.hpt_cma_alloc = 0;
 	/*
-	 * If the user wants a different size from default,
 	 * try first to allocate it from the kernel page allocator.
+	 * We keep the CMA reserved for failed allocation.
 	 */
-	hpt = 0;
-	if (order != kvm_hpt_order) {
-		hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_REPEAT|
-				       __GFP_NOWARN, order - PAGE_SHIFT);
-		if (!hpt)
-			--order;
-	}
+	hpt = __get_free_pages(GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT |
+			       __GFP_NOWARN, order - PAGE_SHIFT);
 
 	/* Next try to allocate from the preallocated pool */
 	if (!hpt) {
-		li = kvm_alloc_hpt();
-		if (li) {
-			hpt = (ulong)li->base_virt;
-			kvm->arch.hpt_li = li;
-			order = kvm_hpt_order;
-		}
+		page = kvm_alloc_hpt(1 << (order - PAGE_SHIFT));
+		if (page) {
+			hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
+			kvm->arch.hpt_cma_alloc = 1;
+		} else
+			--order;
 	}
 
 	/* Lastly try successively smaller sizes from the page allocator */
@@ -118,8 +114,8 @@ long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp)
 	return 0;
 
  out_freehpt:
-	if (kvm->arch.hpt_li)
-		kvm_release_hpt(kvm->arch.hpt_li);
+	if (kvm->arch.hpt_cma_alloc)
+		kvm_release_hpt(page, 1 << (order - PAGE_SHIFT));
 	else
 		free_pages(hpt, order - PAGE_SHIFT);
 	return -ENOMEM;
@@ -165,8 +161,9 @@ void kvmppc_free_hpt(struct kvm *kvm)
 {
 	kvmppc_free_lpid(kvm->arch.lpid);
 	vfree(kvm->arch.revmap);
-	if (kvm->arch.hpt_li)
-		kvm_release_hpt(kvm->arch.hpt_li);
+	if (kvm->arch.hpt_cma_alloc)
+		kvm_release_hpt(virt_to_page(kvm->arch.hpt_virt),
+				1 << (kvm->arch.hpt_order - PAGE_SHIFT));
 	else
 		free_pages(kvm->arch.hpt_virt,
 			   kvm->arch.hpt_order - PAGE_SHIFT);
diff --git a/arch/powerpc/kvm/book3s_hv_builtin.c b/arch/powerpc/kvm/book3s_hv_builtin.c
index ec0a9e5de100..4b865c553331 100644
--- a/arch/powerpc/kvm/book3s_hv_builtin.c
+++ b/arch/powerpc/kvm/book3s_hv_builtin.c
@@ -13,20 +13,30 @@
 #include <linux/spinlock.h>
 #include <linux/bootmem.h>
 #include <linux/init.h>
+#include <linux/memblock.h>
+#include <linux/sizes.h>
 
 #include <asm/cputable.h>
 #include <asm/kvm_ppc.h>
 #include <asm/kvm_book3s.h>
 
+#include "book3s_hv_cma.h"
+
 #define KVM_LINEAR_RMA		0
 #define KVM_LINEAR_HPT		1
 
 static void __init kvm_linear_init_one(ulong size, int count, int type);
 static struct kvmppc_linear_info *kvm_alloc_linear(int type);
 static void kvm_release_linear(struct kvmppc_linear_info *ri);
-
-int kvm_hpt_order = KVM_DEFAULT_HPT_ORDER;
-EXPORT_SYMBOL_GPL(kvm_hpt_order);
+/*
+ * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
+ * should be power of 2.
+ */
+#define HPT_ALIGN_PAGES		((1 << 18) >> PAGE_SHIFT) /* 256k */
+/*
+ * By default we reserve 5% of memory for hash pagetable allocation.
+ */
+static unsigned long kvm_cma_resv_ratio = 5;
 
 /*************** RMA *************/
 
@@ -101,36 +111,29 @@ void kvm_release_rma(struct kvmppc_linear_info *ri)
 }
 EXPORT_SYMBOL_GPL(kvm_release_rma);
 
-/*************** HPT *************/
-
-/*
- * This maintains a list of big linear HPT tables that contain the GVA->HPA
- * memory mappings. If we don't reserve those early on, we might not be able
- * to get a big (usually 16MB) linear memory region from the kernel anymore.
- */
-
-static unsigned long kvm_hpt_count;
-
-static int __init early_parse_hpt_count(char *p)
+static int __init early_parse_kvm_cma_resv(char *p)
 {
+	pr_debug("%s(%s)\n", __func__, p);
 	if (!p)
-		return 1;
-
-	kvm_hpt_count = simple_strtoul(p, NULL, 0);
-
-	return 0;
+		return -EINVAL;
+	return kstrtoul(p, 0, &kvm_cma_resv_ratio);
 }
-early_param("kvm_hpt_count", early_parse_hpt_count);
+early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
 
-struct kvmppc_linear_info *kvm_alloc_hpt(void)
+struct page *kvm_alloc_hpt(unsigned long nr_pages)
 {
-	return kvm_alloc_linear(KVM_LINEAR_HPT);
+	unsigned long align_pages = HPT_ALIGN_PAGES;
+
+	/* Old CPUs require HPT aligned on a multiple of its size */
+	if (!cpu_has_feature(CPU_FTR_ARCH_206))
+		align_pages = nr_pages;
+	return kvm_alloc_cma(nr_pages, align_pages);
 }
 EXPORT_SYMBOL_GPL(kvm_alloc_hpt);
 
-void kvm_release_hpt(struct kvmppc_linear_info *li)
+void kvm_release_hpt(struct page *page, unsigned long nr_pages)
 {
-	kvm_release_linear(li);
+	kvm_release_cma(page, nr_pages);
 }
 EXPORT_SYMBOL_GPL(kvm_release_hpt);
 
@@ -211,9 +214,6 @@ static void kvm_release_linear(struct kvmppc_linear_info *ri)
  */
 void __init kvm_linear_init(void)
 {
-	/* HPT */
-	kvm_linear_init_one(1 << kvm_hpt_order, kvm_hpt_count, KVM_LINEAR_HPT);
-
 	/* RMA */
 	/* Only do this on PPC970 in HV mode */
 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
@@ -231,3 +231,40 @@ void __init kvm_linear_init(void)
 
 	kvm_linear_init_one(kvm_rma_size, kvm_rma_count, KVM_LINEAR_RMA);
 }
+
+/**
+ * kvm_cma_reserve() - reserve area for kvm hash pagetable
+ *
+ * This function reserves memory from early allocator. It should be
+ * called by arch specific code once the early allocator (memblock or bootmem)
+ * has been activated and all other subsystems have already allocated/reserved
+ * memory.
+ */
+void __init kvm_cma_reserve(void)
+{
+	unsigned long align_size;
+	struct memblock_region *reg;
+	phys_addr_t selected_size = 0;
+	/*
+	 * We cannot use memblock_phys_mem_size() here, because
+	 * memblock_analyze() has not been called yet.
+	 */
+	for_each_memblock(memory, reg)
+		selected_size += memblock_region_memory_end_pfn(reg) -
+				 memblock_region_memory_base_pfn(reg);
+
+	selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT;
+	if (selected_size) {
+		pr_debug("%s: reserving %ld MiB for global area\n", __func__,
+			 (unsigned long)selected_size / SZ_1M);
+		/*
+		 * Old CPUs require HPT aligned on a multiple of its size. So for them
+		 * make the alignment as max size we could request.
+		 */
+		if (!cpu_has_feature(CPU_FTR_ARCH_206))
+			align_size = __rounddown_pow_of_two(selected_size);
+		else
+			align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
+		kvm_cma_declare_contiguous(selected_size, align_size);
+	}
+}
diff --git a/arch/powerpc/kvm/book3s_hv_cma.c b/arch/powerpc/kvm/book3s_hv_cma.c
new file mode 100644
index 000000000000..e04b269b9c5b
--- /dev/null
+++ b/arch/powerpc/kvm/book3s_hv_cma.c
@@ -0,0 +1,227 @@
+/*
+ * Contiguous Memory Allocator for ppc KVM hash pagetable  based on CMA
+ * for DMA mapping framework
+ *
+ * Copyright IBM Corporation, 2013
+ * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License or (at your optional) any later version of the license.
+ *
+ */
+#define pr_fmt(fmt) "kvm_cma: " fmt
+
+#ifdef CONFIG_CMA_DEBUG
+#ifndef DEBUG
+#  define DEBUG
+#endif
+#endif
+
+#include <linux/memblock.h>
+#include <linux/mutex.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+
+struct kvm_cma {
+	unsigned long	base_pfn;
+	unsigned long	count;
+	unsigned long	*bitmap;
+};
+
+static DEFINE_MUTEX(kvm_cma_mutex);
+static struct kvm_cma kvm_cma_area;
+
+/**
+ * kvm_cma_declare_contiguous() - reserve area for contiguous memory handling
+ *			          for kvm hash pagetable
+ * @size:  Size of the reserved memory.
+ * @alignment:  Alignment for the contiguous memory area
+ *
+ * This function reserves memory for kvm cma area. It should be
+ * called by arch code when early allocator (memblock or bootmem)
+ * is still activate.
+ */
+long __init kvm_cma_declare_contiguous(phys_addr_t size, phys_addr_t alignment)
+{
+	long base_pfn;
+	phys_addr_t addr;
+	struct kvm_cma *cma = &kvm_cma_area;
+
+	pr_debug("%s(size %lx)\n", __func__, (unsigned long)size);
+
+	if (!size)
+		return -EINVAL;
+	/*
+	 * Sanitise input arguments.
+	 * We should be pageblock aligned for CMA.
+	 */
+	alignment = max(alignment, (phys_addr_t)(PAGE_SIZE << pageblock_order));
+	size = ALIGN(size, alignment);
+	/*
+	 * Reserve memory
+	 * Use __memblock_alloc_base() since
+	 * memblock_alloc_base() panic()s.
+	 */
+	addr = __memblock_alloc_base(size, alignment, 0);
+	if (!addr) {
+		base_pfn = -ENOMEM;
+		goto err;
+	} else
+		base_pfn = PFN_DOWN(addr);
+
+	/*
+	 * Each reserved area must be initialised later, when more kernel
+	 * subsystems (like slab allocator) are available.
+	 */
+	cma->base_pfn = base_pfn;
+	cma->count    = size >> PAGE_SHIFT;
+	pr_info("CMA: reserved %ld MiB\n", (unsigned long)size / SZ_1M);
+	return 0;
+err:
+	pr_err("CMA: failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
+	return base_pfn;
+}
+
+/**
+ * kvm_alloc_cma() - allocate pages from contiguous area
+ * @nr_pages: Requested number of pages.
+ * @align_pages: Requested alignment in number of pages
+ *
+ * This function allocates memory buffer for hash pagetable.
+ */
+struct page *kvm_alloc_cma(unsigned long nr_pages, unsigned long align_pages)
+{
+	int ret;
+	struct page *page = NULL;
+	struct kvm_cma *cma = &kvm_cma_area;
+	unsigned long mask, pfn, pageno, start = 0;
+
+
+	if (!cma || !cma->count)
+		return NULL;
+
+	pr_debug("%s(cma %p, count %lu, align pages %lu)\n", __func__,
+		 (void *)cma, nr_pages, align_pages);
+
+	if (!nr_pages)
+		return NULL;
+
+	VM_BUG_ON(!is_power_of_2(align_pages));
+	mask = align_pages - 1;
+
+	mutex_lock(&kvm_cma_mutex);
+	for (;;) {
+		pageno = bitmap_find_next_zero_area(cma->bitmap, cma->count,
+						    start, nr_pages, mask);
+		if (pageno >= cma->count)
+			break;
+
+		pfn = cma->base_pfn + pageno;
+		ret = alloc_contig_range(pfn, pfn + nr_pages, MIGRATE_CMA);
+		if (ret == 0) {
+			bitmap_set(cma->bitmap, pageno, nr_pages);
+			page = pfn_to_page(pfn);
+			memset(pfn_to_kaddr(pfn), 0, nr_pages << PAGE_SHIFT);
+			break;
+		} else if (ret != -EBUSY) {
+			break;
+		}
+		pr_debug("%s(): memory range at %p is busy, retrying\n",
+			 __func__, pfn_to_page(pfn));
+		/* try again with a bit different memory target */
+		start = pageno + mask + 1;
+	}
+	mutex_unlock(&kvm_cma_mutex);
+	pr_debug("%s(): returned %p\n", __func__, page);
+	return page;
+}
+
+/**
+ * kvm_release_cma() - release allocated pages for hash pagetable
+ * @pages: Allocated pages.
+ * @nr_pages: Number of allocated pages.
+ *
+ * This function releases memory allocated by kvm_alloc_cma().
+ * It returns false when provided pages do not belong to contiguous area and
+ * true otherwise.
+ */
+bool kvm_release_cma(struct page *pages, unsigned long nr_pages)
+{
+	unsigned long pfn;
+	struct kvm_cma *cma = &kvm_cma_area;
+
+
+	if (!cma || !pages)
+		return false;
+
+	pr_debug("%s(page %p count %lu)\n", __func__, (void *)pages, nr_pages);
+
+	pfn = page_to_pfn(pages);
+
+	if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
+		return false;
+
+	VM_BUG_ON(pfn + nr_pages > cma->base_pfn + cma->count);
+
+	mutex_lock(&kvm_cma_mutex);
+	bitmap_clear(cma->bitmap, pfn - cma->base_pfn, nr_pages);
+	free_contig_range(pfn, nr_pages);
+	mutex_unlock(&kvm_cma_mutex);
+
+	return true;
+}
+
+static int __init kvm_cma_activate_area(unsigned long base_pfn,
+					unsigned long count)
+{
+	unsigned long pfn = base_pfn;
+	unsigned i = count >> pageblock_order;
+	struct zone *zone;
+
+	WARN_ON_ONCE(!pfn_valid(pfn));
+	zone = page_zone(pfn_to_page(pfn));
+	do {
+		unsigned j;
+		base_pfn = pfn;
+		for (j = pageblock_nr_pages; j; --j, pfn++) {
+			WARN_ON_ONCE(!pfn_valid(pfn));
+			/*
+			 * alloc_contig_range requires the pfn range
+			 * specified to be in the same zone. Make this
+			 * simple by forcing the entire CMA resv range
+			 * to be in the same zone.
+			 */
+			if (page_zone(pfn_to_page(pfn)) != zone)
+				return -EINVAL;
+		}
+		init_cma_reserved_pageblock(pfn_to_page(base_pfn));
+	} while (--i);
+	return 0;
+}
+
+static int __init kvm_cma_init_reserved_areas(void)
+{
+	int bitmap_size, ret;
+	struct kvm_cma *cma = &kvm_cma_area;
+
+	pr_debug("%s()\n", __func__);
+	if (!cma->count)
+		return 0;
+
+	bitmap_size = BITS_TO_LONGS(cma->count) * sizeof(long);
+	cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
+	if (!cma->bitmap)
+		return -ENOMEM;
+
+	ret = kvm_cma_activate_area(cma->base_pfn, cma->count);
+	if (ret)
+		goto error;
+	return 0;
+
+error:
+	kfree(cma->bitmap);
+	return ret;
+}
+core_initcall(kvm_cma_init_reserved_areas);
diff --git a/arch/powerpc/kvm/book3s_hv_cma.h b/arch/powerpc/kvm/book3s_hv_cma.h
new file mode 100644
index 000000000000..788bc3b73104
--- /dev/null
+++ b/arch/powerpc/kvm/book3s_hv_cma.h
@@ -0,0 +1,22 @@
+/*
+ * Contiguous Memory Allocator for ppc KVM hash pagetable  based on CMA
+ * for DMA mapping framework
+ *
+ * Copyright IBM Corporation, 2013
+ * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License or (at your optional) any later version of the license.
+ *
+ */
+
+#ifndef __POWERPC_KVM_CMA_ALLOC_H__
+#define __POWERPC_KVM_CMA_ALLOC_H__
+extern struct page *kvm_alloc_cma(unsigned long nr_pages,
+				  unsigned long align_pages);
+extern bool kvm_release_cma(struct page *pages, unsigned long nr_pages);
+extern long kvm_cma_declare_contiguous(phys_addr_t size,
+				       phys_addr_t alignment) __init;
+#endif