arch/x86/kernel/kvm.c

   1 /*
   2  * KVM paravirt_ops implementation
   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 as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  *
  18  * Copyright (C) 2007, Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
  19  * Copyright IBM Corporation, 2007
  20  *   Authors: Anthony Liguori <aliguori@us.ibm.com>
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kvm_para.h>
  26 #include <linux/cpu.h>
  27 #include <linux/mm.h>
  28 #include <linux/highmem.h>
  29 #include <linux/hardirq.h>
  30
  31 #define MMU_QUEUE_SIZE 1024
  32
  33 struct kvm_para_state {
  34         u8 mmu_queue[MMU_QUEUE_SIZE];
  35         int mmu_queue_len;
  36         enum paravirt_lazy_mode mode;
  37 };
  38
  39 static DEFINE_PER_CPU(struct kvm_para_state, para_state);
  40
  41 static struct kvm_para_state *kvm_para_state(void)
  42 {
  43         return &per_cpu(para_state, raw_smp_processor_id());
  44 }
  45
  46 /*
  47  * No need for any "IO delay" on KVM
  48  */
  49 static void kvm_io_delay(void)
  50 {
  51 }
  52
  53 static void kvm_mmu_op(void *buffer, unsigned len)
  54 {
  55         int r;
  56         unsigned long a1, a2;
  57
  58         do {
  59                 a1 = __pa(buffer);
  60                 a2 = 0;   /* on i386 __pa() always returns <4G */
  61                 r = kvm_hypercall3(KVM_HC_MMU_OP, len, a1, a2);
  62                 buffer += r;
  63                 len -= r;
  64         } while (len);
  65 }
  66
  67 static void mmu_queue_flush(struct kvm_para_state *state)
  68 {
  69         if (state->mmu_queue_len) {
  70                 kvm_mmu_op(state->mmu_queue, state->mmu_queue_len);
  71                 state->mmu_queue_len = 0;
  72         }
  73 }
  74
  75 static void kvm_deferred_mmu_op(void *buffer, int len)
  76 {
  77         struct kvm_para_state *state = kvm_para_state();
  78
  79         if (state->mode != PARAVIRT_LAZY_MMU) {
  80                 kvm_mmu_op(buffer, len);
  81                 return;
  82         }
  83         if (state->mmu_queue_len + len > sizeof state->mmu_queue)
  84                 mmu_queue_flush(state);
  85         memcpy(state->mmu_queue + state->mmu_queue_len, buffer, len);
  86         state->mmu_queue_len += len;
  87 }
  88
  89 static void kvm_mmu_write(void *dest, u64 val)
  90 {
  91         __u64 pte_phys;
  92         struct kvm_mmu_op_write_pte wpte;
  93
  94 #ifdef CONFIG_HIGHPTE
  95         struct page *page;
  96         unsigned long dst = (unsigned long) dest;
  97
  98         page = kmap_atomic_to_page(dest);
  99         pte_phys = page_to_pfn(page);
 100         pte_phys <<= PAGE_SHIFT;
 101         pte_phys += (dst & ~(PAGE_MASK));
 102 #else
 103         pte_phys = (unsigned long)__pa(dest);
 104 #endif
 105         wpte.header.op = KVM_MMU_OP_WRITE_PTE;
 106         wpte.pte_val = val;
 107         wpte.pte_phys = pte_phys;
 108
 109         kvm_deferred_mmu_op(&wpte, sizeof wpte);
 110 }
 111
 112 /*
 113  * We only need to hook operations that are MMU writes.  We hook these so that
 114  * we can use lazy MMU mode to batch these operations.  We could probably
 115  * improve the performance of the host code if we used some of the information
 116  * here to simplify processing of batched writes.
 117  */
 118 static void kvm_set_pte(pte_t *ptep, pte_t pte)
 119 {
 120         kvm_mmu_write(ptep, pte_val(pte));
 121 }
 122
 123 static void kvm_set_pte_at(struct mm_struct *mm, unsigned long addr,
 124                            pte_t *ptep, pte_t pte)
 125 {
 126         kvm_mmu_write(ptep, pte_val(pte));
 127 }
 128
 129 static void kvm_set_pmd(pmd_t *pmdp, pmd_t pmd)
 130 {
 131         kvm_mmu_write(pmdp, pmd_val(pmd));
 132 }
 133
 134 #if PAGETABLE_LEVELS >= 3
 135 #ifdef CONFIG_X86_PAE
 136 static void kvm_set_pte_atomic(pte_t *ptep, pte_t pte)
 137 {
 138         kvm_mmu_write(ptep, pte_val(pte));
 139 }
 140
 141 static void kvm_set_pte_present(struct mm_struct *mm, unsigned long addr,
 142                                 pte_t *ptep, pte_t pte)
 143 {
 144         kvm_mmu_write(ptep, pte_val(pte));
 145 }
 146
 147 static void kvm_pte_clear(struct mm_struct *mm,
 148                           unsigned long addr, pte_t *ptep)
 149 {
 150         kvm_mmu_write(ptep, 0);
 151 }
 152
 153 static void kvm_pmd_clear(pmd_t *pmdp)
 154 {
 155         kvm_mmu_write(pmdp, 0);
 156 }
 157 #endif
 158
 159 static void kvm_set_pud(pud_t *pudp, pud_t pud)
 160 {
 161         kvm_mmu_write(pudp, pud_val(pud));
 162 }
 163
 164 #if PAGETABLE_LEVELS == 4
 165 static void kvm_set_pgd(pgd_t *pgdp, pgd_t pgd)
 166 {
 167         kvm_mmu_write(pgdp, pgd_val(pgd));
 168 }
 169 #endif
 170 #endif /* PAGETABLE_LEVELS >= 3 */
 171
 172 static void kvm_flush_tlb(void)
 173 {
 174         struct kvm_mmu_op_flush_tlb ftlb = {
 175                 .header.op = KVM_MMU_OP_FLUSH_TLB,
 176         };
 177
 178         kvm_deferred_mmu_op(&ftlb, sizeof ftlb);
 179 }
 180
 181 static void kvm_release_pt(u32 pfn)
 182 {
 183         struct kvm_mmu_op_release_pt rpt = {
 184                 .header.op = KVM_MMU_OP_RELEASE_PT,
 185                 .pt_phys = (u64)pfn << PAGE_SHIFT,
 186         };
 187
 188         kvm_mmu_op(&rpt, sizeof rpt);
 189 }
 190
 191 static void kvm_enter_lazy_mmu(void)
 192 {
 193         struct kvm_para_state *state = kvm_para_state();
 194
 195         paravirt_enter_lazy_mmu();
 196         state->mode = paravirt_get_lazy_mode();
 197 }
 198
 199 static void kvm_leave_lazy_mmu(void)
 200 {
 201         struct kvm_para_state *state = kvm_para_state();
 202
 203         mmu_queue_flush(state);
 204         paravirt_leave_lazy(paravirt_get_lazy_mode());
 205         state->mode = paravirt_get_lazy_mode();
 206 }
 207
 208 static void paravirt_ops_setup(void)
 209 {
 210         pv_info.name = "KVM";
 211         pv_info.paravirt_enabled = 1;
 212
 213         if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
 214                 pv_cpu_ops.io_delay = kvm_io_delay;
 215
 216         if (kvm_para_has_feature(KVM_FEATURE_MMU_OP)) {
 217                 pv_mmu_ops.set_pte = kvm_set_pte;
 218                 pv_mmu_ops.set_pte_at = kvm_set_pte_at;
 219                 pv_mmu_ops.set_pmd = kvm_set_pmd;
 220 #if PAGETABLE_LEVELS >= 3
 221 #ifdef CONFIG_X86_PAE
 222                 pv_mmu_ops.set_pte_atomic = kvm_set_pte_atomic;
 223                 pv_mmu_ops.set_pte_present = kvm_set_pte_present;
 224                 pv_mmu_ops.pte_clear = kvm_pte_clear;
 225                 pv_mmu_ops.pmd_clear = kvm_pmd_clear;
 226 #endif
 227                 pv_mmu_ops.set_pud = kvm_set_pud;
 228 #if PAGETABLE_LEVELS == 4
 229                 pv_mmu_ops.set_pgd = kvm_set_pgd;
 230 #endif
 231 #endif
 232                 pv_mmu_ops.flush_tlb_user = kvm_flush_tlb;
 233                 pv_mmu_ops.release_pte = kvm_release_pt;
 234                 pv_mmu_ops.release_pmd = kvm_release_pt;
 235                 pv_mmu_ops.release_pud = kvm_release_pt;
 236
 237                 pv_mmu_ops.lazy_mode.enter = kvm_enter_lazy_mmu;
 238                 pv_mmu_ops.lazy_mode.leave = kvm_leave_lazy_mmu;
 239         }
 240 }
 241
 242 void __init kvm_guest_init(void)
 243 {
 244         if (!kvm_para_available())
 245                 return;
 246
 247         paravirt_ops_setup();
 248 }