arch/x86/mm/mem_encrypt.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * AMD Memory Encryption Support
   4  *
   5  * Copyright (C) 2016 Advanced Micro Devices, Inc.
   6  *
   7  * Author: Tom Lendacky <thomas.lendacky@amd.com>
   8  */
   9
  10 #define DISABLE_BRANCH_PROFILING
  11
  12 #include <linux/linkage.h>
  13 #include <linux/init.h>
  14 #include <linux/mm.h>
  15 #include <linux/dma-direct.h>
  16 #include <linux/swiotlb.h>
  17 #include <linux/mem_encrypt.h>
  18
  19 #include <asm/tlbflush.h>
  20 #include <asm/fixmap.h>
  21 #include <asm/setup.h>
  22 #include <asm/bootparam.h>
  23 #include <asm/set_memory.h>
  24 #include <asm/cacheflush.h>
  25 #include <asm/processor-flags.h>
  26 #include <asm/msr.h>
  27 #include <asm/cmdline.h>
  28
  29 #include "mm_internal.h"
  30
  31 /*
  32  * Since SME related variables are set early in the boot process they must
  33  * reside in the .data section so as not to be zeroed out when the .bss
  34  * section is later cleared.
  35  */
  36 u64 sme_me_mask __section(.data) = 0;
  37 EXPORT_SYMBOL(sme_me_mask);
  38 DEFINE_STATIC_KEY_FALSE(sev_enable_key);
  39 EXPORT_SYMBOL_GPL(sev_enable_key);
  40
  41 bool sev_enabled __section(.data);
  42
  43 /* Buffer used for early in-place encryption by BSP, no locking needed */
  44 static char sme_early_buffer[PAGE_SIZE] __aligned(PAGE_SIZE);
  45
  46 /*
  47  * This routine does not change the underlying encryption setting of the
  48  * page(s) that map this memory. It assumes that eventually the memory is
  49  * meant to be accessed as either encrypted or decrypted but the contents
  50  * are currently not in the desired state.
  51  *
  52  * This routine follows the steps outlined in the AMD64 Architecture
  53  * Programmer's Manual Volume 2, Section 7.10.8 Encrypt-in-Place.
  54  */
  55 static void __init __sme_early_enc_dec(resource_size_t paddr,
  56                                        unsigned long size, bool enc)
  57 {
  58         void *src, *dst;
  59         size_t len;
  60
  61         if (!sme_me_mask)
  62                 return;
  63
  64         wbinvd();
  65
  66         /*
  67          * There are limited number of early mapping slots, so map (at most)
  68          * one page at time.
  69          */
  70         while (size) {
  71                 len = min_t(size_t, sizeof(sme_early_buffer), size);
  72
  73                 /*
  74                  * Create mappings for the current and desired format of
  75                  * the memory. Use a write-protected mapping for the source.
  76                  */
  77                 src = enc ? early_memremap_decrypted_wp(paddr, len) :
  78                             early_memremap_encrypted_wp(paddr, len);
  79
  80                 dst = enc ? early_memremap_encrypted(paddr, len) :
  81                             early_memremap_decrypted(paddr, len);
  82
  83                 /*
  84                  * If a mapping can't be obtained to perform the operation,
  85                  * then eventual access of that area in the desired mode
  86                  * will cause a crash.
  87                  */
  88                 BUG_ON(!src || !dst);
  89
  90                 /*
  91                  * Use a temporary buffer, of cache-line multiple size, to
  92                  * avoid data corruption as documented in the APM.
  93                  */
  94                 memcpy(sme_early_buffer, src, len);
  95                 memcpy(dst, sme_early_buffer, len);
  96
  97                 early_memunmap(dst, len);
  98                 early_memunmap(src, len);
  99
 100                 paddr += len;
 101                 size -= len;
 102         }
 103 }
 104
 105 void __init sme_early_encrypt(resource_size_t paddr, unsigned long size)
 106 {
 107         __sme_early_enc_dec(paddr, size, true);
 108 }
 109
 110 void __init sme_early_decrypt(resource_size_t paddr, unsigned long size)
 111 {
 112         __sme_early_enc_dec(paddr, size, false);
 113 }
 114
 115 static void __init __sme_early_map_unmap_mem(void *vaddr, unsigned long size,
 116                                              bool map)
 117 {
 118         unsigned long paddr = (unsigned long)vaddr - __PAGE_OFFSET;
 119         pmdval_t pmd_flags, pmd;
 120
 121         /* Use early_pmd_flags but remove the encryption mask */
 122         pmd_flags = __sme_clr(early_pmd_flags);
 123
 124         do {
 125                 pmd = map ? (paddr & PMD_MASK) + pmd_flags : 0;
 126                 __early_make_pgtable((unsigned long)vaddr, pmd);
 127
 128                 vaddr += PMD_SIZE;
 129                 paddr += PMD_SIZE;
 130                 size = (size <= PMD_SIZE) ? 0 : size - PMD_SIZE;
 131         } while (size);
 132
 133         __native_flush_tlb();
 134 }
 135
 136 void __init sme_unmap_bootdata(char *real_mode_data)
 137 {
 138         struct boot_params *boot_data;
 139         unsigned long cmdline_paddr;
 140
 141         if (!sme_active())
 142                 return;
 143
 144         /* Get the command line address before unmapping the real_mode_data */
 145         boot_data = (struct boot_params *)real_mode_data;
 146         cmdline_paddr = boot_data->hdr.cmd_line_ptr | ((u64)boot_data->ext_cmd_line_ptr << 32);
 147
 148         __sme_early_map_unmap_mem(real_mode_data, sizeof(boot_params), false);
 149
 150         if (!cmdline_paddr)
 151                 return;
 152
 153         __sme_early_map_unmap_mem(__va(cmdline_paddr), COMMAND_LINE_SIZE, false);
 154 }
 155
 156 void __init sme_map_bootdata(char *real_mode_data)
 157 {
 158         struct boot_params *boot_data;
 159         unsigned long cmdline_paddr;
 160
 161         if (!sme_active())
 162                 return;
 163
 164         __sme_early_map_unmap_mem(real_mode_data, sizeof(boot_params), true);
 165
 166         /* Get the command line address after mapping the real_mode_data */
 167         boot_data = (struct boot_params *)real_mode_data;
 168         cmdline_paddr = boot_data->hdr.cmd_line_ptr | ((u64)boot_data->ext_cmd_line_ptr << 32);
 169
 170         if (!cmdline_paddr)
 171                 return;
 172
 173         __sme_early_map_unmap_mem(__va(cmdline_paddr), COMMAND_LINE_SIZE, true);
 174 }
 175
 176 void __init sme_early_init(void)
 177 {
 178         unsigned int i;
 179
 180         if (!sme_me_mask)
 181                 return;
 182
 183         early_pmd_flags = __sme_set(early_pmd_flags);
 184
 185         __supported_pte_mask = __sme_set(__supported_pte_mask);
 186
 187         /* Update the protection map with memory encryption mask */
 188         for (i = 0; i < ARRAY_SIZE(protection_map); i++)
 189                 protection_map[i] = pgprot_encrypted(protection_map[i]);
 190
 191         if (sev_active())
 192                 swiotlb_force = SWIOTLB_FORCE;
 193 }
 194
 195 static void __init __set_clr_pte_enc(pte_t *kpte, int level, bool enc)
 196 {
 197         pgprot_t old_prot, new_prot;
 198         unsigned long pfn, pa, size;
 199         pte_t new_pte;
 200
 201         switch (level) {
 202         case PG_LEVEL_4K:
 203                 pfn = pte_pfn(*kpte);
 204                 old_prot = pte_pgprot(*kpte);
 205                 break;
 206         case PG_LEVEL_2M:
 207                 pfn = pmd_pfn(*(pmd_t *)kpte);
 208                 old_prot = pmd_pgprot(*(pmd_t *)kpte);
 209                 break;
 210         case PG_LEVEL_1G:
 211                 pfn = pud_pfn(*(pud_t *)kpte);
 212                 old_prot = pud_pgprot(*(pud_t *)kpte);
 213                 break;
 214         default:
 215                 return;
 216         }
 217
 218         new_prot = old_prot;
 219         if (enc)
 220                 pgprot_val(new_prot) |= _PAGE_ENC;
 221         else
 222                 pgprot_val(new_prot) &= ~_PAGE_ENC;
 223
 224         /* If prot is same then do nothing. */
 225         if (pgprot_val(old_prot) == pgprot_val(new_prot))
 226                 return;
 227
 228         pa = pfn << page_level_shift(level);
 229         size = page_level_size(level);
 230
 231         /*
 232          * We are going to perform in-place en-/decryption and change the
 233          * physical page attribute from C=1 to C=0 or vice versa. Flush the
 234          * caches to ensure that data gets accessed with the correct C-bit.
 235          */
 236         clflush_cache_range(__va(pa), size);
 237
 238         /* Encrypt/decrypt the contents in-place */
 239         if (enc)
 240                 sme_early_encrypt(pa, size);
 241         else
 242                 sme_early_decrypt(pa, size);
 243
 244         /* Change the page encryption mask. */
 245         new_pte = pfn_pte(pfn, new_prot);
 246         set_pte_atomic(kpte, new_pte);
 247 }
 248
 249 static int __init early_set_memory_enc_dec(unsigned long vaddr,
 250                                            unsigned long size, bool enc)
 251 {
 252         unsigned long vaddr_end, vaddr_next;
 253         unsigned long psize, pmask;
 254         int split_page_size_mask;
 255         int level, ret;
 256         pte_t *kpte;
 257
 258         vaddr_next = vaddr;
 259         vaddr_end = vaddr + size;
 260
 261         for (; vaddr < vaddr_end; vaddr = vaddr_next) {
 262                 kpte = lookup_address(vaddr, &level);
 263                 if (!kpte || pte_none(*kpte)) {
 264                         ret = 1;
 265                         goto out;
 266                 }
 267
 268                 if (level == PG_LEVEL_4K) {
 269                         __set_clr_pte_enc(kpte, level, enc);
 270                         vaddr_next = (vaddr & PAGE_MASK) + PAGE_SIZE;
 271                         continue;
 272                 }
 273
 274                 psize = page_level_size(level);
 275                 pmask = page_level_mask(level);
 276
 277                 /*
 278                  * Check whether we can change the large page in one go.
 279                  * We request a split when the address is not aligned and
 280                  * the number of pages to set/clear encryption bit is smaller
 281                  * than the number of pages in the large page.
 282                  */
 283                 if (vaddr == (vaddr & pmask) &&
 284                     ((vaddr_end - vaddr) >= psize)) {
 285                         __set_clr_pte_enc(kpte, level, enc);
 286                         vaddr_next = (vaddr & pmask) + psize;
 287                         continue;
 288                 }
 289
 290                 /*
 291                  * The virtual address is part of a larger page, create the next
 292                  * level page table mapping (4K or 2M). If it is part of a 2M
 293                  * page then we request a split of the large page into 4K
 294                  * chunks. A 1GB large page is split into 2M pages, resp.
 295                  */
 296                 if (level == PG_LEVEL_2M)
 297                         split_page_size_mask = 0;
 298                 else
 299                         split_page_size_mask = 1 << PG_LEVEL_2M;
 300
 301                 /*
 302                  * kernel_physical_mapping_change() does not flush the TLBs, so
 303                  * a TLB flush is required after we exit from the for loop.
 304                  */
 305                 kernel_physical_mapping_change(__pa(vaddr & pmask),
 306                                                __pa((vaddr_end & pmask) + psize),
 307                                                split_page_size_mask);
 308         }
 309
 310         ret = 0;
 311
 312 out:
 313         __flush_tlb_all();
 314         return ret;
 315 }
 316
 317 int __init early_set_memory_decrypted(unsigned long vaddr, unsigned long size)
 318 {
 319         return early_set_memory_enc_dec(vaddr, size, false);
 320 }
 321
 322 int __init early_set_memory_encrypted(unsigned long vaddr, unsigned long size)
 323 {
 324         return early_set_memory_enc_dec(vaddr, size, true);
 325 }
 326
 327 /*
 328  * SME and SEV are very similar but they are not the same, so there are
 329  * times that the kernel will need to distinguish between SME and SEV. The
 330  * sme_active() and sev_active() functions are used for this.  When a
 331  * distinction isn't needed, the mem_encrypt_active() function can be used.
 332  *
 333  * The trampoline code is a good example for this requirement.  Before
 334  * paging is activated, SME will access all memory as decrypted, but SEV
 335  * will access all memory as encrypted.  So, when APs are being brought
 336  * up under SME the trampoline area cannot be encrypted, whereas under SEV
 337  * the trampoline area must be encrypted.
 338  */
 339 bool sme_active(void)
 340 {
 341         return sme_me_mask && !sev_enabled;
 342 }
 343 EXPORT_SYMBOL(sme_active);
 344
 345 bool sev_active(void)
 346 {
 347         return sme_me_mask && sev_enabled;
 348 }
 349 EXPORT_SYMBOL(sev_active);
 350
 351 /* Architecture __weak replacement functions */
 352 void __init mem_encrypt_free_decrypted_mem(void)
 353 {
 354         unsigned long vaddr, vaddr_end, npages;
 355         int r;
 356
 357         vaddr = (unsigned long)__start_bss_decrypted_unused;
 358         vaddr_end = (unsigned long)__end_bss_decrypted;
 359         npages = (vaddr_end - vaddr) >> PAGE_SHIFT;
 360
 361         /*
 362          * The unused memory range was mapped decrypted, change the encryption
 363          * attribute from decrypted to encrypted before freeing it.
 364          */
 365         if (mem_encrypt_active()) {
 366                 r = set_memory_encrypted(vaddr, npages);
 367                 if (r) {
 368                         pr_warn("failed to free unused decrypted pages\n");
 369                         return;
 370                 }
 371         }
 372
 373         free_init_pages("unused decrypted", vaddr, vaddr_end);
 374 }
 375
 376 void __init mem_encrypt_init(void)
 377 {
 378         if (!sme_me_mask)
 379                 return;
 380
 381         /* Call into SWIOTLB to update the SWIOTLB DMA buffers */
 382         swiotlb_update_mem_attributes();
 383
 384         /*
 385          * With SEV, we need to unroll the rep string I/O instructions.
 386          */
 387         if (sev_active())
 388                 static_branch_enable(&sev_enable_key);
 389
 390         pr_info("AMD %s active\n",
 391                 sev_active() ? "Secure Encrypted Virtualization (SEV)"
 392                              : "Secure Memory Encryption (SME)");
 393 }
 394