arch/powerpc/kvm/book3s_hv_builtin.c

   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/init.h>
  16 #include <linux/memblock.h>
  17 #include <linux/sizes.h>
  18 #include <linux/cma.h>
  19
  20 #include <asm/cputable.h>
  21 #include <asm/kvm_ppc.h>
  22 #include <asm/kvm_book3s.h>
  23
  24 #define KVM_CMA_CHUNK_ORDER     18
  25
  26 /*
  27  * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
  28  * should be power of 2.
  29  */
  30 #define HPT_ALIGN_PAGES         ((1 << 18) >> PAGE_SHIFT) /* 256k */
  31 /*
  32  * By default we reserve 5% of memory for hash pagetable allocation.
  33  */
  34 static unsigned long kvm_cma_resv_ratio = 5;
  35 /*
  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.
  43  */
  44 unsigned long kvm_rma_pages = (1 << 27) >> PAGE_SHIFT;  /* 128MB */
  45 EXPORT_SYMBOL_GPL(kvm_rma_pages);
  46
  47 static struct cma *kvm_cma;
  48
  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)
  52 {
  53         switch (rma_size) {
  54         case 32ul << 20:        /* 32 MB */
  55                 if (cpu_has_feature(CPU_FTR_ARCH_206))
  56                         return 8;       /* only supported on POWER7 */
  57                 return -1;
  58         case 64ul << 20:        /* 64 MB */
  59                 return 3;
  60         case 128ul << 20:       /* 128 MB */
  61                 return 7;
  62         case 256ul << 20:       /* 256 MB */
  63                 return 4;
  64         case 1ul << 30:         /* 1 GB */
  65                 return 2;
  66         case 16ul << 30:        /* 16 GB */
  67                 return 1;
  68         case 256ul << 30:       /* 256 GB */
  69                 return 0;
  70         default:
  71                 return -1;
  72         }
  73 }
  74
  75 static int __init early_parse_rma_size(char *p)
  76 {
  77         unsigned long kvm_rma_size;
  78
  79         pr_debug("%s(%s)\n", __func__, p);
  80         if (!p)
  81                 return -EINVAL;
  82         kvm_rma_size = memparse(p, &p);
  83         /*
  84          * Check that the requested size is one supported in hardware
  85          */
  86         if (lpcr_rmls(kvm_rma_size) < 0) {
  87                 pr_err("RMA size of 0x%lx not supported\n", kvm_rma_size);
  88                 return -EINVAL;
  89         }
  90         kvm_rma_pages = kvm_rma_size >> PAGE_SHIFT;
  91         return 0;
  92 }
  93 early_param("kvm_rma_size", early_parse_rma_size);
  94
  95 struct kvm_rma_info *kvm_alloc_rma()
  96 {
  97         struct page *page;
  98         struct kvm_rma_info *ri;
  99
 100         ri = kmalloc(sizeof(struct kvm_rma_info), GFP_KERNEL);
 101         if (!ri)
 102                 return NULL;
 103         page = cma_alloc(kvm_cma, kvm_rma_pages, order_base_2(kvm_rma_pages));
 104         if (!page)
 105                 goto err_out;
 106         atomic_set(&ri->use_count, 1);
 107         ri->base_pfn = page_to_pfn(page);
 108         return ri;
 109 err_out:
 110         kfree(ri);
 111         return NULL;
 112 }
 113 EXPORT_SYMBOL_GPL(kvm_alloc_rma);
 114
 115 void kvm_release_rma(struct kvm_rma_info *ri)
 116 {
 117         if (atomic_dec_and_test(&ri->use_count)) {
 118                 cma_release(kvm_cma, pfn_to_page(ri->base_pfn), kvm_rma_pages);
 119                 kfree(ri);
 120         }
 121 }
 122 EXPORT_SYMBOL_GPL(kvm_release_rma);
 123
 124 static int __init early_parse_kvm_cma_resv(char *p)
 125 {
 126         pr_debug("%s(%s)\n", __func__, p);
 127         if (!p)
 128                 return -EINVAL;
 129         return kstrtoul(p, 0, &kvm_cma_resv_ratio);
 130 }
 131 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
 132
 133 struct page *kvm_alloc_hpt(unsigned long nr_pages)
 134 {
 135         unsigned long align_pages = HPT_ALIGN_PAGES;
 136
 137         VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT);
 138
 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));
 143 }
 144 EXPORT_SYMBOL_GPL(kvm_alloc_hpt);
 145
 146 void kvm_release_hpt(struct page *page, unsigned long nr_pages)
 147 {
 148         cma_release(kvm_cma, page, nr_pages);
 149 }
 150 EXPORT_SYMBOL_GPL(kvm_release_hpt);
 151
 152 /**
 153  * kvm_cma_reserve() - reserve area for kvm hash pagetable
 154  *
 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
 158  * memory.
 159  */
 160 void __init kvm_cma_reserve(void)
 161 {
 162         unsigned long align_size;
 163         struct memblock_region *reg;
 164         phys_addr_t selected_size = 0;
 165
 166         /*
 167          * We need CMA reservation only when we are in HV mode
 168          */
 169         if (!cpu_has_feature(CPU_FTR_HVMODE))
 170                 return;
 171         /*
 172          * We cannot use memblock_phys_mem_size() here, because
 173          * memblock_analyze() has not been called yet.
 174          */
 175         for_each_memblock(memory, reg)
 176                 selected_size += memblock_region_memory_end_pfn(reg) -
 177                                  memblock_region_memory_base_pfn(reg);
 178
 179         selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT;
 180         if (selected_size) {
 181                 pr_debug("%s: reserving %ld MiB for global area\n", __func__,
 182                          (unsigned long)selected_size / SZ_1M);
 183                 /*
 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.
 186                  */
 187                 if (!cpu_has_feature(CPU_FTR_ARCH_206))
 188                         align_size = __rounddown_pow_of_two(selected_size);
 189                 else
 190                         align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
 191
 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);
 195         }
 196 }
 197
 198 /*
 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.
 201  *
 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().
 204  */
 205 static atomic_t hv_vm_count;
 206
 207 void kvm_hv_vm_activated(void)
 208 {
 209         get_online_cpus();
 210         atomic_inc(&hv_vm_count);
 211         put_online_cpus();
 212 }
 213 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated);
 214
 215 void kvm_hv_vm_deactivated(void)
 216 {
 217         get_online_cpus();
 218         atomic_dec(&hv_vm_count);
 219         put_online_cpus();
 220 }
 221 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated);
 222
 223 bool kvm_hv_mode_active(void)
 224 {
 225         return atomic_read(&hv_vm_count) != 0;
 226 }
 227
 228 extern int hcall_real_table[], hcall_real_table_end[];
 229
 230 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd)
 231 {
 232         cmd /= 4;
 233         if (cmd < hcall_real_table_end - hcall_real_table &&
 234             hcall_real_table[cmd])
 235                 return 1;
 236
 237         return 0;
 238 }
 239 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode);