arch/x86/mm/mem_encrypt.c

   1 /*
   2  * AMD Memory Encryption Support
   3  *
   4  * Copyright (C) 2016 Advanced Micro Devices, Inc.
   5  *
   6  * Author: Tom Lendacky <thomas.lendacky@amd.com>
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12
  13 #define DISABLE_BRANCH_PROFILING
  14
  15 #include <linux/linkage.h>
  16 #include <linux/init.h>
  17 #include <linux/mm.h>
  18 #include <linux/dma-direct.h>
  19 #include <linux/swiotlb.h>
  20 #include <linux/mem_encrypt.h>
  21
  22 #include <asm/tlbflush.h>
  23 #include <asm/fixmap.h>
  24 #include <asm/setup.h>
  25 #include <asm/bootparam.h>
  26 #include <asm/set_memory.h>
  27 #include <asm/cacheflush.h>
  28 #include <asm/processor-flags.h>
  29 #include <asm/msr.h>
  30 #include <asm/cmdline.h>
  31
  32 #include "mm_internal.h"
  33
  34 /*
  35  * Since SME related variables are set early in the boot process they must
  36  * reside in the .data section so as not to be zeroed out when the .bss
  37  * section is later cleared.
  38  */
  39 u64 sme_me_mask __section(.data) = 0;
  40 EXPORT_SYMBOL(sme_me_mask);
  41 DEFINE_STATIC_KEY_FALSE(sev_enable_key);
  42 EXPORT_SYMBOL_GPL(sev_enable_key);
  43
  44 bool sev_enabled __section(.data);
  45
  46 /* Buffer used for early in-place encryption by BSP, no locking needed */
  47 static char sme_early_buffer[PAGE_SIZE] __aligned(PAGE_SIZE);
  48
  49 /*
  50  * This routine does not change the underlying encryption setting of the
  51  * page(s) that map this memory. It assumes that eventually the memory is
  52  * meant to be accessed as either encrypted or decrypted but the contents
  53  * are currently not in the desired state.
  54  *
  55  * This routine follows the steps outlined in the AMD64 Architecture
  56  * Programmer's Manual Volume 2, Section 7.10.8 Encrypt-in-Place.
  57  */
  58 static void __init __sme_early_enc_dec(resource_size_t paddr,
  59                                        unsigned long size, bool enc)
  60 {
  61         void *src, *dst;
  62         size_t len;
  63
  64         if (!sme_me_mask)
  65                 return;
  66
  67         wbinvd();
  68
  69         /*
  70          * There are limited number of early mapping slots, so map (at most)
  71          * one page at time.
  72          */
  73         while (size) {
  74                 len = min_t(size_t, sizeof(sme_early_buffer), size);
  75
  76                 /*
  77                  * Create mappings for the current and desired format of
  78                  * the memory. Use a write-protected mapping for the source.
  79                  */
  80                 src = enc ? early_memremap_decrypted_wp(paddr, len) :
  81                             early_memremap_encrypted_wp(paddr, len);
  82
  83                 dst = enc ? early_memremap_encrypted(paddr, len) :
  84                             early_memremap_decrypted(paddr, len);
  85
  86                 /*
  87                  * If a mapping can't be obtained to perform the operation,
  88                  * then eventual access of that area in the desired mode
  89                  * will cause a crash.
  90                  */
  91                 BUG_ON(!src || !dst);
  92
  93                 /*
  94                  * Use a temporary buffer, of cache-line multiple size, to
  95                  * avoid data corruption as documented in the APM.
  96                  */
  97                 memcpy(sme_early_buffer, src, len);
  98                 memcpy(dst, sme_early_buffer, len);
  99
 100                 early_memunmap(dst, len);
 101                 early_memunmap(src, len);
 102
 103                 paddr += len;
 104                 size -= len;
 105         }
 106 }
 107
 108 void __init sme_early_encrypt(resource_size_t paddr, unsigned long size)
 109 {
 110         __sme_early_enc_dec(paddr, size, true);
 111 }
 112
 113 void __init sme_early_decrypt(resource_size_t paddr, unsigned long size)
 114 {
 115         __sme_early_enc_dec(paddr, size, false);
 116 }
 117
 118 static void __init __sme_early_map_unmap_mem(void *vaddr, unsigned long size,
 119                                              bool map)
 120 {
 121         unsigned long paddr = (unsigned long)vaddr - __PAGE_OFFSET;
 122         pmdval_t pmd_flags, pmd;
 123
 124         /* Use early_pmd_flags but remove the encryption mask */
 125         pmd_flags = __sme_clr(early_pmd_flags);
 126
 127         do {
 128                 pmd = map ? (paddr & PMD_MASK) + pmd_flags : 0;
 129                 __early_make_pgtable((unsigned long)vaddr, pmd);
 130
 131                 vaddr += PMD_SIZE;
 132                 paddr += PMD_SIZE;
 133                 size = (size <= PMD_SIZE) ? 0 : size - PMD_SIZE;
 134         } while (size);
 135
 136         __native_flush_tlb();
 137 }
 138
 139 void __init sme_unmap_bootdata(char *real_mode_data)
 140 {
 141         struct boot_params *boot_data;
 142         unsigned long cmdline_paddr;
 143
 144         if (!sme_active())
 145                 return;
 146
 147         /* Get the command line address before unmapping the real_mode_data */
 148         boot_data = (struct boot_params *)real_mode_data;
 149         cmdline_paddr = boot_data->hdr.cmd_line_ptr | ((u64)boot_data->ext_cmd_line_ptr << 32);
 150
 151         __sme_early_map_unmap_mem(real_mode_data, sizeof(boot_params), false);
 152
 153         if (!cmdline_paddr)
 154                 return;
 155
 156         __sme_early_map_unmap_mem(__va(cmdline_paddr), COMMAND_LINE_SIZE, false);
 157 }
 158
 159 void __init sme_map_bootdata(char *real_mode_data)
 160 {
 161         struct boot_params *boot_data;
 162         unsigned long cmdline_paddr;
 163
 164         if (!sme_active())
 165                 return;
 166
 167         __sme_early_map_unmap_mem(real_mode_data, sizeof(boot_params), true);
 168
 169         /* Get the command line address after mapping the real_mode_data */
 170         boot_data = (struct boot_params *)real_mode_data;
 171         cmdline_paddr = boot_data->hdr.cmd_line_ptr | ((u64)boot_data->ext_cmd_line_ptr << 32);
 172
 173         if (!cmdline_paddr)
 174                 return;
 175
 176         __sme_early_map_unmap_mem(__va(cmdline_paddr), COMMAND_LINE_SIZE, true);
 177 }
 178
 179 void __init sme_early_init(void)
 180 {
 181         unsigned int i;
 182
 183         if (!sme_me_mask)
 184                 return;
 185
 186         early_pmd_flags = __sme_set(early_pmd_flags);
 187
 188         __supported_pte_mask = __sme_set(__supported_pte_mask);
 189
 190         /* Update the protection map with memory encryption mask */
 191         for (i = 0; i < ARRAY_SIZE(protection_map); i++)
 192                 protection_map[i] = pgprot_encrypted(protection_map[i]);
 193
 194         if (sev_active())
 195                 swiotlb_force = SWIOTLB_FORCE;
 196 }
 197
 198 static void __init __set_clr_pte_enc(pte_t *kpte, int level, bool enc)
 199 {
 200         pgprot_t old_prot, new_prot;
 201         unsigned long pfn, pa, size;
 202         pte_t new_pte;
 203
 204         switch (level) {
 205         case PG_LEVEL_4K:
 206                 pfn = pte_pfn(*kpte);
 207                 old_prot = pte_pgprot(*kpte);
 208                 break;
 209         case PG_LEVEL_2M:
 210                 pfn = pmd_pfn(*(pmd_t *)kpte);
 211                 old_prot = pmd_pgprot(*(pmd_t *)kpte);
 212                 break;
 213         case PG_LEVEL_1G:
 214                 pfn = pud_pfn(*(pud_t *)kpte);
 215                 old_prot = pud_pgprot(*(pud_t *)kpte);
 216                 break;
 217         default:
 218                 return;
 219         }
 220
 221         new_prot = old_prot;
 222         if (enc)
 223                 pgprot_val(new_prot) |= _PAGE_ENC;
 224         else
 225                 pgprot_val(new_prot) &= ~_PAGE_ENC;
 226
 227         /* If prot is same then do nothing. */
 228         if (pgprot_val(old_prot) == pgprot_val(new_prot))
 229                 return;
 230
 231         pa = pfn << page_level_shift(level);
 232         size = page_level_size(level);
 233
 234         /*
 235          * We are going to perform in-place en-/decryption and change the
 236          * physical page attribute from C=1 to C=0 or vice versa. Flush the
 237          * caches to ensure that data gets accessed with the correct C-bit.
 238          */
 239         clflush_cache_range(__va(pa), size);
 240
 241         /* Encrypt/decrypt the contents in-place */
 242         if (enc)
 243                 sme_early_encrypt(pa, size);
 244         else
 245                 sme_early_decrypt(pa, size);
 246
 247         /* Change the page encryption mask. */
 248         new_pte = pfn_pte(pfn, new_prot);
 249         set_pte_atomic(kpte, new_pte);
 250 }
 251
 252 static int __init early_set_memory_enc_dec(unsigned long vaddr,
 253                                            unsigned long size, bool enc)
 254 {
 255         unsigned long vaddr_end, vaddr_next;
 256         unsigned long psize, pmask;
 257         int split_page_size_mask;
 258         int level, ret;
 259         pte_t *kpte;
 260
 261         vaddr_next = vaddr;
 262         vaddr_end = vaddr + size;
 263
 264         for (; vaddr < vaddr_end; vaddr = vaddr_next) {
 265                 kpte = lookup_address(vaddr, &level);
 266                 if (!kpte || pte_none(*kpte)) {
 267                         ret = 1;
 268                         goto out;
 269                 }
 270
 271                 if (level == PG_LEVEL_4K) {
 272                         __set_clr_pte_enc(kpte, level, enc);
 273                         vaddr_next = (vaddr & PAGE_MASK) + PAGE_SIZE;
 274                         continue;
 275                 }
 276
 277                 psize = page_level_size(level);
 278                 pmask = page_level_mask(level);
 279
 280                 /*
 281                  * Check whether we can change the large page in one go.
 282                  * We request a split when the address is not aligned and
 283                  * the number of pages to set/clear encryption bit is smaller
 284                  * than the number of pages in the large page.
 285                  */
 286                 if (vaddr == (vaddr & pmask) &&
 287                     ((vaddr_end - vaddr) >= psize)) {
 288                         __set_clr_pte_enc(kpte, level, enc);
 289                         vaddr_next = (vaddr & pmask) + psize;
 290                         continue;
 291                 }
 292
 293                 /*
 294                  * The virtual address is part of a larger page, create the next
 295                  * level page table mapping (4K or 2M). If it is part of a 2M
 296                  * page then we request a split of the large page into 4K
 297                  * chunks. A 1GB large page is split into 2M pages, resp.
 298                  */
 299                 if (level == PG_LEVEL_2M)
 300                         split_page_size_mask = 0;
 301                 else
 302                         split_page_size_mask = 1 << PG_LEVEL_2M;
 303
 304                 kernel_physical_mapping_init(__pa(vaddr & pmask),
 305                                              __pa((vaddr_end & pmask) + psize),
 306                                              split_page_size_mask);
 307         }
 308
 309         ret = 0;
 310
 311 out:
 312         __flush_tlb_all();
 313         return ret;
 314 }
 315
 316 int __init early_set_memory_decrypted(unsigned long vaddr, unsigned long size)
 317 {
 318         return early_set_memory_enc_dec(vaddr, size, false);
 319 }
 320
 321 int __init early_set_memory_encrypted(unsigned long vaddr, unsigned long size)
 322 {
 323         return early_set_memory_enc_dec(vaddr, size, true);
 324 }
 325
 326 /*
 327  * SME and SEV are very similar but they are not the same, so there are
 328  * times that the kernel will need to distinguish between SME and SEV. The
 329  * sme_active() and sev_active() functions are used for this.  When a
 330  * distinction isn't needed, the mem_encrypt_active() function can be used.
 331  *
 332  * The trampoline code is a good example for this requirement.  Before
 333  * paging is activated, SME will access all memory as decrypted, but SEV
 334  * will access all memory as encrypted.  So, when APs are being brought
 335  * up under SME the trampoline area cannot be encrypted, whereas under SEV
 336  * the trampoline area must be encrypted.
 337  */
 338 bool sme_active(void)
 339 {
 340         return sme_me_mask && !sev_enabled;
 341 }
 342 EXPORT_SYMBOL(sme_active);
 343
 344 bool sev_active(void)
 345 {
 346         return sme_me_mask && sev_enabled;
 347 }
 348 EXPORT_SYMBOL(sev_active);
 349
 350 /* Architecture __weak replacement functions */
 351 void __init mem_encrypt_free_decrypted_mem(void)
 352 {
 353         unsigned long vaddr, vaddr_end, npages;
 354         int r;
 355
 356         vaddr = (unsigned long)__start_bss_decrypted_unused;
 357         vaddr_end = (unsigned long)__end_bss_decrypted;
 358         npages = (vaddr_end - vaddr) >> PAGE_SHIFT;
 359
 360         /*
 361          * The unused memory range was mapped decrypted, change the encryption
 362          * attribute from decrypted to encrypted before freeing it.
 363          */
 364         if (mem_encrypt_active()) {
 365                 r = set_memory_encrypted(vaddr, npages);
 366                 if (r) {
 367                         pr_warn("failed to free unused decrypted pages\n");
 368                         return;
 369                 }
 370         }
 371
 372         free_init_pages("unused decrypted", vaddr, vaddr_end);
 373 }
 374
 375 void __init mem_encrypt_init(void)
 376 {
 377         if (!sme_me_mask)
 378                 return;
 379
 380         /* Call into SWIOTLB to update the SWIOTLB DMA buffers */
 381         swiotlb_update_mem_attributes();
 382
 383         /*
 384          * With SEV, we need to unroll the rep string I/O instructions.
 385          */
 386         if (sev_active())
 387                 static_branch_enable(&sev_enable_key);
 388
 389         pr_info("AMD %s active\n",
 390                 sev_active() ? "Secure Encrypted Virtualization (SEV)"
 391                              : "Secure Memory Encryption (SME)");
 392 }
 393