root/arch/x86/mm/mem_encrypt.c

DEFINITIONS

1. __sme_early_enc_dec
2. sme_early_encrypt
3. sme_early_decrypt
4. __sme_early_map_unmap_mem
5. sme_unmap_bootdata
6. sme_map_bootdata
7. sme_early_init
8. __set_clr_pte_enc
9. early_set_memory_enc_dec
10. early_set_memory_decrypted
11. early_set_memory_encrypted
12. sme_active
13. sev_active
14. force_dma_unencrypted
15. mem_encrypt_free_decrypted_mem
16. mem_encrypt_init

   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 #include <linux/device.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bitops.h>
  21 #include <linux/dma-mapping.h>
  22 
  23 #include <asm/tlbflush.h>
  24 #include <asm/fixmap.h>
  25 #include <asm/setup.h>
  26 #include <asm/bootparam.h>
  27 #include <asm/set_memory.h>
  28 #include <asm/cacheflush.h>
  29 #include <asm/processor-flags.h>
  30 #include <asm/msr.h>
  31 #include <asm/cmdline.h>
  32 
  33 #include "mm_internal.h"
  34 
  35 /*
  36  * Since SME related variables are set early in the boot process they must
  37  * reside in the .data section so as not to be zeroed out when the .bss
  38  * section is later cleared.
  39  */
  40 u64 sme_me_mask __section(.data) = 0;
  41 EXPORT_SYMBOL(sme_me_mask);
  42 DEFINE_STATIC_KEY_FALSE(sev_enable_key);
  43 EXPORT_SYMBOL_GPL(sev_enable_key);
  44 
  45 bool sev_enabled __section(.data);
  46 
  47 /* Buffer used for early in-place encryption by BSP, no locking needed */
  48 static char sme_early_buffer[PAGE_SIZE] __initdata __aligned(PAGE_SIZE);
  49 
  50 /*
  51  * This routine does not change the underlying encryption setting of the
  52  * page(s) that map this memory. It assumes that eventually the memory is
  53  * meant to be accessed as either encrypted or decrypted but the contents
  54  * are currently not in the desired state.
  55  *
  56  * This routine follows the steps outlined in the AMD64 Architecture
  57  * Programmer's Manual Volume 2, Section 7.10.8 Encrypt-in-Place.
  58  */
  59 static void __init __sme_early_enc_dec(resource_size_t paddr,
  60                                        unsigned long size, bool enc)
  61 {
  62         void *src, *dst;
  63         size_t len;
  64 
  65         if (!sme_me_mask)
  66                 return;
  67 
  68         wbinvd();
  69 
  70         /*
  71          * There are limited number of early mapping slots, so map (at most)
  72          * one page at time.
  73          */
  74         while (size) {
  75                 len = min_t(size_t, sizeof(sme_early_buffer), size);
  76 
  77                 /*
  78                  * Create mappings for the current and desired format of
  79                  * the memory. Use a write-protected mapping for the source.
  80                  */
  81                 src = enc ? early_memremap_decrypted_wp(paddr, len) :
  82                             early_memremap_encrypted_wp(paddr, len);
  83 
  84                 dst = enc ? early_memremap_encrypted(paddr, len) :
  85                             early_memremap_decrypted(paddr, len);
  86 
  87                 /*
  88                  * If a mapping can't be obtained to perform the operation,
  89                  * then eventual access of that area in the desired mode
  90                  * will cause a crash.
  91                  */
  92                 BUG_ON(!src || !dst);
  93 
  94                 /*
  95                  * Use a temporary buffer, of cache-line multiple size, to
  96                  * avoid data corruption as documented in the APM.
  97                  */
  98                 memcpy(sme_early_buffer, src, len);
  99                 memcpy(dst, sme_early_buffer, len);
 100 
 101                 early_memunmap(dst, len);
 102                 early_memunmap(src, len);
 103 
 104                 paddr += len;
 105                 size -= len;
 106         }
 107 }
 108 
 109 void __init sme_early_encrypt(resource_size_t paddr, unsigned long size)
 110 {
 111         __sme_early_enc_dec(paddr, size, true);
 112 }
 113 
 114 void __init sme_early_decrypt(resource_size_t paddr, unsigned long size)
 115 {
 116         __sme_early_enc_dec(paddr, size, false);
 117 }
 118 
 119 static void __init __sme_early_map_unmap_mem(void *vaddr, unsigned long size,
 120                                              bool map)
 121 {
 122         unsigned long paddr = (unsigned long)vaddr - __PAGE_OFFSET;
 123         pmdval_t pmd_flags, pmd;
 124 
 125         /* Use early_pmd_flags but remove the encryption mask */
 126         pmd_flags = __sme_clr(early_pmd_flags);
 127 
 128         do {
 129                 pmd = map ? (paddr & PMD_MASK) + pmd_flags : 0;
 130                 __early_make_pgtable((unsigned long)vaddr, pmd);
 131 
 132                 vaddr += PMD_SIZE;
 133                 paddr += PMD_SIZE;
 134                 size = (size <= PMD_SIZE) ? 0 : size - PMD_SIZE;
 135         } while (size);
 136 
 137         __native_flush_tlb();
 138 }
 139 
 140 void __init sme_unmap_bootdata(char *real_mode_data)
 141 {
 142         struct boot_params *boot_data;
 143         unsigned long cmdline_paddr;
 144 
 145         if (!sme_active())
 146                 return;
 147 
 148         /* Get the command line address before unmapping the real_mode_data */
 149         boot_data = (struct boot_params *)real_mode_data;
 150         cmdline_paddr = boot_data->hdr.cmd_line_ptr | ((u64)boot_data->ext_cmd_line_ptr << 32);
 151 
 152         __sme_early_map_unmap_mem(real_mode_data, sizeof(boot_params), false);
 153 
 154         if (!cmdline_paddr)
 155                 return;
 156 
 157         __sme_early_map_unmap_mem(__va(cmdline_paddr), COMMAND_LINE_SIZE, false);
 158 }
 159 
 160 void __init sme_map_bootdata(char *real_mode_data)
 161 {
 162         struct boot_params *boot_data;
 163         unsigned long cmdline_paddr;
 164 
 165         if (!sme_active())
 166                 return;
 167 
 168         __sme_early_map_unmap_mem(real_mode_data, sizeof(boot_params), true);
 169 
 170         /* Get the command line address after mapping the real_mode_data */
 171         boot_data = (struct boot_params *)real_mode_data;
 172         cmdline_paddr = boot_data->hdr.cmd_line_ptr | ((u64)boot_data->ext_cmd_line_ptr << 32);
 173 
 174         if (!cmdline_paddr)
 175                 return;
 176 
 177         __sme_early_map_unmap_mem(__va(cmdline_paddr), COMMAND_LINE_SIZE, true);
 178 }
 179 
 180 void __init sme_early_init(void)
 181 {
 182         unsigned int i;
 183 
 184         if (!sme_me_mask)
 185                 return;
 186 
 187         early_pmd_flags = __sme_set(early_pmd_flags);
 188 
 189         __supported_pte_mask = __sme_set(__supported_pte_mask);
 190 
 191         /* Update the protection map with memory encryption mask */
 192         for (i = 0; i < ARRAY_SIZE(protection_map); i++)
 193                 protection_map[i] = pgprot_encrypted(protection_map[i]);
 194 
 195         if (sev_active())
 196                 swiotlb_force = SWIOTLB_FORCE;
 197 }
 198 
 199 static void __init __set_clr_pte_enc(pte_t *kpte, int level, bool enc)
 200 {
 201         pgprot_t old_prot, new_prot;
 202         unsigned long pfn, pa, size;
 203         pte_t new_pte;
 204 
 205         switch (level) {
 206         case PG_LEVEL_4K:
 207                 pfn = pte_pfn(*kpte);
 208                 old_prot = pte_pgprot(*kpte);
 209                 break;
 210         case PG_LEVEL_2M:
 211                 pfn = pmd_pfn(*(pmd_t *)kpte);
 212                 old_prot = pmd_pgprot(*(pmd_t *)kpte);
 213                 break;
 214         case PG_LEVEL_1G:
 215                 pfn = pud_pfn(*(pud_t *)kpte);
 216                 old_prot = pud_pgprot(*(pud_t *)kpte);
 217                 break;
 218         default:
 219                 return;
 220         }
 221 
 222         new_prot = old_prot;
 223         if (enc)
 224                 pgprot_val(new_prot) |= _PAGE_ENC;
 225         else
 226                 pgprot_val(new_prot) &= ~_PAGE_ENC;
 227 
 228         /* If prot is same then do nothing. */
 229         if (pgprot_val(old_prot) == pgprot_val(new_prot))
 230                 return;
 231 
 232         pa = pfn << page_level_shift(level);
 233         size = page_level_size(level);
 234 
 235         /*
 236          * We are going to perform in-place en-/decryption and change the
 237          * physical page attribute from C=1 to C=0 or vice versa. Flush the
 238          * caches to ensure that data gets accessed with the correct C-bit.
 239          */
 240         clflush_cache_range(__va(pa), size);
 241 
 242         /* Encrypt/decrypt the contents in-place */
 243         if (enc)
 244                 sme_early_encrypt(pa, size);
 245         else
 246                 sme_early_decrypt(pa, size);
 247 
 248         /* Change the page encryption mask. */
 249         new_pte = pfn_pte(pfn, new_prot);
 250         set_pte_atomic(kpte, new_pte);
 251 }
 252 
 253 static int __init early_set_memory_enc_dec(unsigned long vaddr,
 254                                            unsigned long size, bool enc)
 255 {
 256         unsigned long vaddr_end, vaddr_next;
 257         unsigned long psize, pmask;
 258         int split_page_size_mask;
 259         int level, ret;
 260         pte_t *kpte;
 261 
 262         vaddr_next = vaddr;
 263         vaddr_end = vaddr + size;
 264 
 265         for (; vaddr < vaddr_end; vaddr = vaddr_next) {
 266                 kpte = lookup_address(vaddr, &level);
 267                 if (!kpte || pte_none(*kpte)) {
 268                         ret = 1;
 269                         goto out;
 270                 }
 271 
 272                 if (level == PG_LEVEL_4K) {
 273                         __set_clr_pte_enc(kpte, level, enc);
 274                         vaddr_next = (vaddr & PAGE_MASK) + PAGE_SIZE;
 275                         continue;
 276                 }
 277 
 278                 psize = page_level_size(level);
 279                 pmask = page_level_mask(level);
 280 
 281                 /*
 282                  * Check whether we can change the large page in one go.
 283                  * We request a split when the address is not aligned and
 284                  * the number of pages to set/clear encryption bit is smaller
 285                  * than the number of pages in the large page.
 286                  */
 287                 if (vaddr == (vaddr & pmask) &&
 288                     ((vaddr_end - vaddr) >= psize)) {
 289                         __set_clr_pte_enc(kpte, level, enc);
 290                         vaddr_next = (vaddr & pmask) + psize;
 291                         continue;
 292                 }
 293 
 294                 /*
 295                  * The virtual address is part of a larger page, create the next
 296                  * level page table mapping (4K or 2M). If it is part of a 2M
 297                  * page then we request a split of the large page into 4K
 298                  * chunks. A 1GB large page is split into 2M pages, resp.
 299                  */
 300                 if (level == PG_LEVEL_2M)
 301                         split_page_size_mask = 0;
 302                 else
 303                         split_page_size_mask = 1 << PG_LEVEL_2M;
 304 
 305                 /*
 306                  * kernel_physical_mapping_change() does not flush the TLBs, so
 307                  * a TLB flush is required after we exit from the for loop.
 308                  */
 309                 kernel_physical_mapping_change(__pa(vaddr & pmask),
 310                                                __pa((vaddr_end & pmask) + psize),
 311                                                split_page_size_mask);
 312         }
 313 
 314         ret = 0;
 315 
 316 out:
 317         __flush_tlb_all();
 318         return ret;
 319 }
 320 
 321 int __init early_set_memory_decrypted(unsigned long vaddr, unsigned long size)
 322 {
 323         return early_set_memory_enc_dec(vaddr, size, false);
 324 }
 325 
 326 int __init early_set_memory_encrypted(unsigned long vaddr, unsigned long size)
 327 {
 328         return early_set_memory_enc_dec(vaddr, size, true);
 329 }
 330 
 331 /*
 332  * SME and SEV are very similar but they are not the same, so there are
 333  * times that the kernel will need to distinguish between SME and SEV. The
 334  * sme_active() and sev_active() functions are used for this.  When a
 335  * distinction isn't needed, the mem_encrypt_active() function can be used.
 336  *
 337  * The trampoline code is a good example for this requirement.  Before
 338  * paging is activated, SME will access all memory as decrypted, but SEV
 339  * will access all memory as encrypted.  So, when APs are being brought
 340  * up under SME the trampoline area cannot be encrypted, whereas under SEV
 341  * the trampoline area must be encrypted.
 342  */
 343 bool sme_active(void)
 344 {
 345         return sme_me_mask && !sev_enabled;
 346 }
 347 
 348 bool sev_active(void)
 349 {
 350         return sme_me_mask && sev_enabled;
 351 }
 352 
 353 /* Override for DMA direct allocation check - ARCH_HAS_FORCE_DMA_UNENCRYPTED */
 354 bool force_dma_unencrypted(struct device *dev)
 355 {
 356         /*
 357          * For SEV, all DMA must be to unencrypted addresses.
 358          */
 359         if (sev_active())
 360                 return true;
 361 
 362         /*
 363          * For SME, all DMA must be to unencrypted addresses if the
 364          * device does not support DMA to addresses that include the
 365          * encryption mask.
 366          */
 367         if (sme_active()) {
 368                 u64 dma_enc_mask = DMA_BIT_MASK(__ffs64(sme_me_mask));
 369                 u64 dma_dev_mask = min_not_zero(dev->coherent_dma_mask,
 370                                                 dev->bus_dma_mask);
 371 
 372                 if (dma_dev_mask <= dma_enc_mask)
 373                         return true;
 374         }
 375 
 376         return false;
 377 }
 378 
 379 /* Architecture __weak replacement functions */
 380 void __init mem_encrypt_free_decrypted_mem(void)
 381 {
 382         unsigned long vaddr, vaddr_end, npages;
 383         int r;
 384 
 385         vaddr = (unsigned long)__start_bss_decrypted_unused;
 386         vaddr_end = (unsigned long)__end_bss_decrypted;
 387         npages = (vaddr_end - vaddr) >> PAGE_SHIFT;
 388 
 389         /*
 390          * The unused memory range was mapped decrypted, change the encryption
 391          * attribute from decrypted to encrypted before freeing it.
 392          */
 393         if (mem_encrypt_active()) {
 394                 r = set_memory_encrypted(vaddr, npages);
 395                 if (r) {
 396                         pr_warn("failed to free unused decrypted pages\n");
 397                         return;
 398                 }
 399         }
 400 
 401         free_init_pages("unused decrypted", vaddr, vaddr_end);
 402 }
 403 
 404 void __init mem_encrypt_init(void)
 405 {
 406         if (!sme_me_mask)
 407                 return;
 408 
 409         /* Call into SWIOTLB to update the SWIOTLB DMA buffers */
 410         swiotlb_update_mem_attributes();
 411 
 412         /*
 413          * With SEV, we need to unroll the rep string I/O instructions.
 414          */
 415         if (sev_active())
 416                 static_branch_enable(&sev_enable_key);
 417 
 418         pr_info("AMD %s active\n",
 419                 sev_active() ? "Secure Encrypted Virtualization (SEV)"
 420                              : "Secure Memory Encryption (SME)");
 421 }
 422