1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License, version 2, as
6  * published by the Free Software Foundation.
7  */
8 
9 #include <linux/kvm_host.h>
10 #include <linux/preempt.h>
11 #include <linux/export.h>
12 #include <linux/sched.h>
13 #include <linux/spinlock.h>
14 #include <linux/bootmem.h>
15 #include <linux/init.h>
16 #include <linux/memblock.h>
17 #include <linux/sizes.h>
18 
19 #include <asm/cputable.h>
20 #include <asm/kvm_ppc.h>
21 #include <asm/kvm_book3s.h>
22 
23 #include "book3s_hv_cma.h"
24 
25 #define KVM_LINEAR_RMA		0
26 #define KVM_LINEAR_HPT		1
27 
28 static void __init kvm_linear_init_one(ulong size, int count, int type);
29 static struct kvmppc_linear_info *kvm_alloc_linear(int type);
30 static void kvm_release_linear(struct kvmppc_linear_info *ri);
31 /*
32  * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
33  * should be power of 2.
34  */
35 #define HPT_ALIGN_PAGES		((1 << 18) >> PAGE_SHIFT) /* 256k */
36 /*
37  * By default we reserve 5% of memory for hash pagetable allocation.
38  */
39 static unsigned long kvm_cma_resv_ratio = 5;
40 
41 /*************** RMA *************/
42 
43 /*
44  * This maintains a list of RMAs (real mode areas) for KVM guests to use.
45  * Each RMA has to be physically contiguous and of a size that the
46  * hardware supports.  PPC970 and POWER7 support 64MB, 128MB and 256MB,
47  * and other larger sizes.  Since we are unlikely to be allocate that
48  * much physically contiguous memory after the system is up and running,
49  * we preallocate a set of RMAs in early boot for KVM to use.
50  */
51 static unsigned long kvm_rma_size = 64 << 20;	/* 64MB */
52 static unsigned long kvm_rma_count;
53 
54 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
55    Assumes POWER7 or PPC970. */
56 static inline int lpcr_rmls(unsigned long rma_size)
57 {
58 	switch (rma_size) {
59 	case 32ul << 20:	/* 32 MB */
60 		if (cpu_has_feature(CPU_FTR_ARCH_206))
61 			return 8;	/* only supported on POWER7 */
62 		return -1;
63 	case 64ul << 20:	/* 64 MB */
64 		return 3;
65 	case 128ul << 20:	/* 128 MB */
66 		return 7;
67 	case 256ul << 20:	/* 256 MB */
68 		return 4;
69 	case 1ul << 30:		/* 1 GB */
70 		return 2;
71 	case 16ul << 30:	/* 16 GB */
72 		return 1;
73 	case 256ul << 30:	/* 256 GB */
74 		return 0;
75 	default:
76 		return -1;
77 	}
78 }
79 
80 static int __init early_parse_rma_size(char *p)
81 {
82 	if (!p)
83 		return 1;
84 
85 	kvm_rma_size = memparse(p, &p);
86 
87 	return 0;
88 }
89 early_param("kvm_rma_size", early_parse_rma_size);
90 
91 static int __init early_parse_rma_count(char *p)
92 {
93 	if (!p)
94 		return 1;
95 
96 	kvm_rma_count = simple_strtoul(p, NULL, 0);
97 
98 	return 0;
99 }
100 early_param("kvm_rma_count", early_parse_rma_count);
101 
102 struct kvmppc_linear_info *kvm_alloc_rma(void)
103 {
104 	return kvm_alloc_linear(KVM_LINEAR_RMA);
105 }
106 EXPORT_SYMBOL_GPL(kvm_alloc_rma);
107 
108 void kvm_release_rma(struct kvmppc_linear_info *ri)
109 {
110 	kvm_release_linear(ri);
111 }
112 EXPORT_SYMBOL_GPL(kvm_release_rma);
113 
114 static int __init early_parse_kvm_cma_resv(char *p)
115 {
116 	pr_debug("%s(%s)\n", __func__, p);
117 	if (!p)
118 		return -EINVAL;
119 	return kstrtoul(p, 0, &kvm_cma_resv_ratio);
120 }
121 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
122 
123 struct page *kvm_alloc_hpt(unsigned long nr_pages)
124 {
125 	unsigned long align_pages = HPT_ALIGN_PAGES;
126 
127 	/* Old CPUs require HPT aligned on a multiple of its size */
128 	if (!cpu_has_feature(CPU_FTR_ARCH_206))
129 		align_pages = nr_pages;
130 	return kvm_alloc_cma(nr_pages, align_pages);
131 }
132 EXPORT_SYMBOL_GPL(kvm_alloc_hpt);
133 
134 void kvm_release_hpt(struct page *page, unsigned long nr_pages)
135 {
136 	kvm_release_cma(page, nr_pages);
137 }
138 EXPORT_SYMBOL_GPL(kvm_release_hpt);
139 
140 /*************** generic *************/
141 
142 static LIST_HEAD(free_linears);
143 static DEFINE_SPINLOCK(linear_lock);
144 
145 static void __init kvm_linear_init_one(ulong size, int count, int type)
146 {
147 	unsigned long i;
148 	unsigned long j, npages;
149 	void *linear;
150 	struct page *pg;
151 	const char *typestr;
152 	struct kvmppc_linear_info *linear_info;
153 
154 	if (!count)
155 		return;
156 
157 	typestr = (type == KVM_LINEAR_RMA) ? "RMA" : "HPT";
158 
159 	npages = size >> PAGE_SHIFT;
160 	linear_info = alloc_bootmem(count * sizeof(struct kvmppc_linear_info));
161 	for (i = 0; i < count; ++i) {
162 		linear = alloc_bootmem_align(size, size);
163 		pr_debug("Allocated KVM %s at %p (%ld MB)\n", typestr, linear,
164 			 size >> 20);
165 		linear_info[i].base_virt = linear;
166 		linear_info[i].base_pfn = __pa(linear) >> PAGE_SHIFT;
167 		linear_info[i].npages = npages;
168 		linear_info[i].type = type;
169 		list_add_tail(&linear_info[i].list, &free_linears);
170 		atomic_set(&linear_info[i].use_count, 0);
171 
172 		pg = pfn_to_page(linear_info[i].base_pfn);
173 		for (j = 0; j < npages; ++j) {
174 			atomic_inc(&pg->_count);
175 			++pg;
176 		}
177 	}
178 }
179 
180 static struct kvmppc_linear_info *kvm_alloc_linear(int type)
181 {
182 	struct kvmppc_linear_info *ri, *ret;
183 
184 	ret = NULL;
185 	spin_lock(&linear_lock);
186 	list_for_each_entry(ri, &free_linears, list) {
187 		if (ri->type != type)
188 			continue;
189 
190 		list_del(&ri->list);
191 		atomic_inc(&ri->use_count);
192 		memset(ri->base_virt, 0, ri->npages << PAGE_SHIFT);
193 		ret = ri;
194 		break;
195 	}
196 	spin_unlock(&linear_lock);
197 	return ret;
198 }
199 
200 static void kvm_release_linear(struct kvmppc_linear_info *ri)
201 {
202 	if (atomic_dec_and_test(&ri->use_count)) {
203 		spin_lock(&linear_lock);
204 		list_add_tail(&ri->list, &free_linears);
205 		spin_unlock(&linear_lock);
206 
207 	}
208 }
209 
210 /*
211  * Called at boot time while the bootmem allocator is active,
212  * to allocate contiguous physical memory for the hash page
213  * tables for guests.
214  */
215 void __init kvm_linear_init(void)
216 {
217 	/* RMA */
218 	/* Only do this on PPC970 in HV mode */
219 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
220 	    !cpu_has_feature(CPU_FTR_ARCH_201))
221 		return;
222 
223 	if (!kvm_rma_size || !kvm_rma_count)
224 		return;
225 
226 	/* Check that the requested size is one supported in hardware */
227 	if (lpcr_rmls(kvm_rma_size) < 0) {
228 		pr_err("RMA size of 0x%lx not supported\n", kvm_rma_size);
229 		return;
230 	}
231 
232 	kvm_linear_init_one(kvm_rma_size, kvm_rma_count, KVM_LINEAR_RMA);
233 }
234 
235 /**
236  * kvm_cma_reserve() - reserve area for kvm hash pagetable
237  *
238  * This function reserves memory from early allocator. It should be
239  * called by arch specific code once the early allocator (memblock or bootmem)
240  * has been activated and all other subsystems have already allocated/reserved
241  * memory.
242  */
243 void __init kvm_cma_reserve(void)
244 {
245 	unsigned long align_size;
246 	struct memblock_region *reg;
247 	phys_addr_t selected_size = 0;
248 	/*
249 	 * We cannot use memblock_phys_mem_size() here, because
250 	 * memblock_analyze() has not been called yet.
251 	 */
252 	for_each_memblock(memory, reg)
253 		selected_size += memblock_region_memory_end_pfn(reg) -
254 				 memblock_region_memory_base_pfn(reg);
255 
256 	selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT;
257 	if (selected_size) {
258 		pr_debug("%s: reserving %ld MiB for global area\n", __func__,
259 			 (unsigned long)selected_size / SZ_1M);
260 		/*
261 		 * Old CPUs require HPT aligned on a multiple of its size. So for them
262 		 * make the alignment as max size we could request.
263 		 */
264 		if (!cpu_has_feature(CPU_FTR_ARCH_206))
265 			align_size = __rounddown_pow_of_two(selected_size);
266 		else
267 			align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
268 		kvm_cma_declare_contiguous(selected_size, align_size);
269 	}
270 }
271