2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
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.
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/init.h>
16 #include <linux/memblock.h>
17 #include <linux/sizes.h>
18 #include <linux/cma.h>
20 #include <asm/cputable.h>
21 #include <asm/kvm_ppc.h>
22 #include <asm/kvm_book3s.h>
24 #define KVM_CMA_CHUNK_ORDER 18
27 * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
28 * should be power of 2.
30 #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */
32 * By default we reserve 5% of memory for hash pagetable allocation.
34 static unsigned long kvm_cma_resv_ratio
= 5;
36 * We allocate RMAs (real mode areas) for KVM guests from the KVM CMA area.
37 * Each RMA has to be physically contiguous and of a size that the
38 * hardware supports. PPC970 and POWER7 support 64MB, 128MB and 256MB,
39 * and other larger sizes. Since we are unlikely to be allocate that
40 * much physically contiguous memory after the system is up and running,
41 * we preallocate a set of RMAs in early boot using CMA.
42 * should be power of 2.
44 unsigned long kvm_rma_pages
= (1 << 27) >> PAGE_SHIFT
; /* 128MB */
45 EXPORT_SYMBOL_GPL(kvm_rma_pages
);
47 static struct cma
*kvm_cma
;
49 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
50 Assumes POWER7 or PPC970. */
51 static inline int lpcr_rmls(unsigned long rma_size
)
54 case 32ul << 20: /* 32 MB */
55 if (cpu_has_feature(CPU_FTR_ARCH_206
))
56 return 8; /* only supported on POWER7 */
58 case 64ul << 20: /* 64 MB */
60 case 128ul << 20: /* 128 MB */
62 case 256ul << 20: /* 256 MB */
64 case 1ul << 30: /* 1 GB */
66 case 16ul << 30: /* 16 GB */
68 case 256ul << 30: /* 256 GB */
75 static int __init
early_parse_rma_size(char *p
)
77 unsigned long kvm_rma_size
;
79 pr_debug("%s(%s)\n", __func__
, p
);
82 kvm_rma_size
= memparse(p
, &p
);
84 * Check that the requested size is one supported in hardware
86 if (lpcr_rmls(kvm_rma_size
) < 0) {
87 pr_err("RMA size of 0x%lx not supported\n", kvm_rma_size
);
90 kvm_rma_pages
= kvm_rma_size
>> PAGE_SHIFT
;
93 early_param("kvm_rma_size", early_parse_rma_size
);
95 struct kvm_rma_info
*kvm_alloc_rma()
98 struct kvm_rma_info
*ri
;
100 ri
= kmalloc(sizeof(struct kvm_rma_info
), GFP_KERNEL
);
103 page
= cma_alloc(kvm_cma
, kvm_rma_pages
, order_base_2(kvm_rma_pages
));
106 atomic_set(&ri
->use_count
, 1);
107 ri
->base_pfn
= page_to_pfn(page
);
113 EXPORT_SYMBOL_GPL(kvm_alloc_rma
);
115 void kvm_release_rma(struct kvm_rma_info
*ri
)
117 if (atomic_dec_and_test(&ri
->use_count
)) {
118 cma_release(kvm_cma
, pfn_to_page(ri
->base_pfn
), kvm_rma_pages
);
122 EXPORT_SYMBOL_GPL(kvm_release_rma
);
124 static int __init
early_parse_kvm_cma_resv(char *p
)
126 pr_debug("%s(%s)\n", __func__
, p
);
129 return kstrtoul(p
, 0, &kvm_cma_resv_ratio
);
131 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv
);
133 struct page
*kvm_alloc_hpt(unsigned long nr_pages
)
135 unsigned long align_pages
= HPT_ALIGN_PAGES
;
137 VM_BUG_ON(order_base_2(nr_pages
) < KVM_CMA_CHUNK_ORDER
- PAGE_SHIFT
);
139 /* Old CPUs require HPT aligned on a multiple of its size */
140 if (!cpu_has_feature(CPU_FTR_ARCH_206
))
141 align_pages
= nr_pages
;
142 return cma_alloc(kvm_cma
, nr_pages
, order_base_2(align_pages
));
144 EXPORT_SYMBOL_GPL(kvm_alloc_hpt
);
146 void kvm_release_hpt(struct page
*page
, unsigned long nr_pages
)
148 cma_release(kvm_cma
, page
, nr_pages
);
150 EXPORT_SYMBOL_GPL(kvm_release_hpt
);
153 * kvm_cma_reserve() - reserve area for kvm hash pagetable
155 * This function reserves memory from early allocator. It should be
156 * called by arch specific code once the memblock allocator
157 * has been activated and all other subsystems have already allocated/reserved
160 void __init
kvm_cma_reserve(void)
162 unsigned long align_size
;
163 struct memblock_region
*reg
;
164 phys_addr_t selected_size
= 0;
167 * We need CMA reservation only when we are in HV mode
169 if (!cpu_has_feature(CPU_FTR_HVMODE
))
172 * We cannot use memblock_phys_mem_size() here, because
173 * memblock_analyze() has not been called yet.
175 for_each_memblock(memory
, reg
)
176 selected_size
+= memblock_region_memory_end_pfn(reg
) -
177 memblock_region_memory_base_pfn(reg
);
179 selected_size
= (selected_size
* kvm_cma_resv_ratio
/ 100) << PAGE_SHIFT
;
181 pr_debug("%s: reserving %ld MiB for global area\n", __func__
,
182 (unsigned long)selected_size
/ SZ_1M
);
184 * Old CPUs require HPT aligned on a multiple of its size. So for them
185 * make the alignment as max size we could request.
187 if (!cpu_has_feature(CPU_FTR_ARCH_206
))
188 align_size
= __rounddown_pow_of_two(selected_size
);
190 align_size
= HPT_ALIGN_PAGES
<< PAGE_SHIFT
;
192 align_size
= max(kvm_rma_pages
<< PAGE_SHIFT
, align_size
);
193 cma_declare_contiguous(0, selected_size
, 0, align_size
,
194 KVM_CMA_CHUNK_ORDER
- PAGE_SHIFT
, false, &kvm_cma
);
199 * When running HV mode KVM we need to block certain operations while KVM VMs
200 * exist in the system. We use a counter of VMs to track this.
202 * One of the operations we need to block is onlining of secondaries, so we
203 * protect hv_vm_count with get/put_online_cpus().
205 static atomic_t hv_vm_count
;
207 void kvm_hv_vm_activated(void)
210 atomic_inc(&hv_vm_count
);
213 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated
);
215 void kvm_hv_vm_deactivated(void)
218 atomic_dec(&hv_vm_count
);
221 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated
);
223 bool kvm_hv_mode_active(void)
225 return atomic_read(&hv_vm_count
) != 0;
228 extern int hcall_real_table
[], hcall_real_table_end
[];
230 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd
)
233 if (cmd
< hcall_real_table_end
- hcall_real_table
&&
234 hcall_real_table
[cmd
])
239 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode
);