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/cpu.h>
10 #include <linux/kvm_host.h>
11 #include <linux/preempt.h>
12 #include <linux/export.h>
13 #include <linux/sched.h>
14 #include <linux/spinlock.h>
15 #include <linux/bootmem.h>
16 #include <linux/init.h>
17 #include <linux/memblock.h>
18 #include <linux/sizes.h>
19 #include <linux/cma.h>
20 
21 #include <asm/cputable.h>
22 #include <asm/kvm_ppc.h>
23 #include <asm/kvm_book3s.h>
24 
25 #define KVM_CMA_CHUNK_ORDER	18
26 
27 /*
28  * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
29  * should be power of 2.
30  */
31 #define HPT_ALIGN_PAGES		((1 << 18) >> PAGE_SHIFT) /* 256k */
32 /*
33  * By default we reserve 5% of memory for hash pagetable allocation.
34  */
35 static unsigned long kvm_cma_resv_ratio = 5;
36 /*
37  * We allocate RMAs (real mode areas) for KVM guests from the KVM CMA area.
38  * Each RMA has to be physically contiguous and of a size that the
39  * hardware supports.  PPC970 and POWER7 support 64MB, 128MB and 256MB,
40  * and other larger sizes.  Since we are unlikely to be allocate that
41  * much physically contiguous memory after the system is up and running,
42  * we preallocate a set of RMAs in early boot using CMA.
43  * should be power of 2.
44  */
45 unsigned long kvm_rma_pages = (1 << 27) >> PAGE_SHIFT;	/* 128MB */
46 EXPORT_SYMBOL_GPL(kvm_rma_pages);
47 
48 static struct cma *kvm_cma;
49 
50 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
51    Assumes POWER7 or PPC970. */
52 static inline int lpcr_rmls(unsigned long rma_size)
53 {
54 	switch (rma_size) {
55 	case 32ul << 20:	/* 32 MB */
56 		if (cpu_has_feature(CPU_FTR_ARCH_206))
57 			return 8;	/* only supported on POWER7 */
58 		return -1;
59 	case 64ul << 20:	/* 64 MB */
60 		return 3;
61 	case 128ul << 20:	/* 128 MB */
62 		return 7;
63 	case 256ul << 20:	/* 256 MB */
64 		return 4;
65 	case 1ul << 30:		/* 1 GB */
66 		return 2;
67 	case 16ul << 30:	/* 16 GB */
68 		return 1;
69 	case 256ul << 30:	/* 256 GB */
70 		return 0;
71 	default:
72 		return -1;
73 	}
74 }
75 
76 static int __init early_parse_rma_size(char *p)
77 {
78 	unsigned long kvm_rma_size;
79 
80 	pr_debug("%s(%s)\n", __func__, p);
81 	if (!p)
82 		return -EINVAL;
83 	kvm_rma_size = memparse(p, &p);
84 	/*
85 	 * Check that the requested size is one supported in hardware
86 	 */
87 	if (lpcr_rmls(kvm_rma_size) < 0) {
88 		pr_err("RMA size of 0x%lx not supported\n", kvm_rma_size);
89 		return -EINVAL;
90 	}
91 	kvm_rma_pages = kvm_rma_size >> PAGE_SHIFT;
92 	return 0;
93 }
94 early_param("kvm_rma_size", early_parse_rma_size);
95 
96 struct kvm_rma_info *kvm_alloc_rma()
97 {
98 	struct page *page;
99 	struct kvm_rma_info *ri;
100 
101 	ri = kmalloc(sizeof(struct kvm_rma_info), GFP_KERNEL);
102 	if (!ri)
103 		return NULL;
104 	page = cma_alloc(kvm_cma, kvm_rma_pages, get_order(kvm_rma_pages));
105 	if (!page)
106 		goto err_out;
107 	atomic_set(&ri->use_count, 1);
108 	ri->base_pfn = page_to_pfn(page);
109 	return ri;
110 err_out:
111 	kfree(ri);
112 	return NULL;
113 }
114 EXPORT_SYMBOL_GPL(kvm_alloc_rma);
115 
116 void kvm_release_rma(struct kvm_rma_info *ri)
117 {
118 	if (atomic_dec_and_test(&ri->use_count)) {
119 		cma_release(kvm_cma, pfn_to_page(ri->base_pfn), kvm_rma_pages);
120 		kfree(ri);
121 	}
122 }
123 EXPORT_SYMBOL_GPL(kvm_release_rma);
124 
125 static int __init early_parse_kvm_cma_resv(char *p)
126 {
127 	pr_debug("%s(%s)\n", __func__, p);
128 	if (!p)
129 		return -EINVAL;
130 	return kstrtoul(p, 0, &kvm_cma_resv_ratio);
131 }
132 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
133 
134 struct page *kvm_alloc_hpt(unsigned long nr_pages)
135 {
136 	unsigned long align_pages = HPT_ALIGN_PAGES;
137 
138 	VM_BUG_ON(get_order(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT);
139 
140 	/* Old CPUs require HPT aligned on a multiple of its size */
141 	if (!cpu_has_feature(CPU_FTR_ARCH_206))
142 		align_pages = nr_pages;
143 	return cma_alloc(kvm_cma, nr_pages, get_order(align_pages));
144 }
145 EXPORT_SYMBOL_GPL(kvm_alloc_hpt);
146 
147 void kvm_release_hpt(struct page *page, unsigned long nr_pages)
148 {
149 	cma_release(kvm_cma, page, nr_pages);
150 }
151 EXPORT_SYMBOL_GPL(kvm_release_hpt);
152 
153 /**
154  * kvm_cma_reserve() - reserve area for kvm hash pagetable
155  *
156  * This function reserves memory from early allocator. It should be
157  * called by arch specific code once the early allocator (memblock or bootmem)
158  * has been activated and all other subsystems have already allocated/reserved
159  * memory.
160  */
161 void __init kvm_cma_reserve(void)
162 {
163 	unsigned long align_size;
164 	struct memblock_region *reg;
165 	phys_addr_t selected_size = 0;
166 	/*
167 	 * We cannot use memblock_phys_mem_size() here, because
168 	 * memblock_analyze() has not been called yet.
169 	 */
170 	for_each_memblock(memory, reg)
171 		selected_size += memblock_region_memory_end_pfn(reg) -
172 				 memblock_region_memory_base_pfn(reg);
173 
174 	selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT;
175 	if (selected_size) {
176 		pr_debug("%s: reserving %ld MiB for global area\n", __func__,
177 			 (unsigned long)selected_size / SZ_1M);
178 		/*
179 		 * Old CPUs require HPT aligned on a multiple of its size. So for them
180 		 * make the alignment as max size we could request.
181 		 */
182 		if (!cpu_has_feature(CPU_FTR_ARCH_206))
183 			align_size = __rounddown_pow_of_two(selected_size);
184 		else
185 			align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
186 
187 		align_size = max(kvm_rma_pages << PAGE_SHIFT, align_size);
188 		cma_declare_contiguous(0, selected_size, 0, align_size,
189 			KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, false, &kvm_cma);
190 	}
191 }
192 
193 /*
194  * When running HV mode KVM we need to block certain operations while KVM VMs
195  * exist in the system. We use a counter of VMs to track this.
196  *
197  * One of the operations we need to block is onlining of secondaries, so we
198  * protect hv_vm_count with get/put_online_cpus().
199  */
200 static atomic_t hv_vm_count;
201 
202 void kvm_hv_vm_activated(void)
203 {
204 	get_online_cpus();
205 	atomic_inc(&hv_vm_count);
206 	put_online_cpus();
207 }
208 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated);
209 
210 void kvm_hv_vm_deactivated(void)
211 {
212 	get_online_cpus();
213 	atomic_dec(&hv_vm_count);
214 	put_online_cpus();
215 }
216 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated);
217 
218 bool kvm_hv_mode_active(void)
219 {
220 	return atomic_read(&hv_vm_count) != 0;
221 }
222 
223 extern int hcall_real_table[], hcall_real_table_end[];
224 
225 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd)
226 {
227 	cmd /= 4;
228 	if (cmd < hcall_real_table_end - hcall_real_table &&
229 	    hcall_real_table[cmd])
230 		return 1;
231 
232 	return 0;
233 }
234 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode);
235