xref: /linux/arch/x86/mm/mem_encrypt_identity.c (revision d198b34f3855eee2571dda03eea75a09c7c31480)
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 /*
13  * Since we're dealing with identity mappings, physical and virtual
14  * addresses are the same, so override these defines which are ultimately
15  * used by the headers in misc.h.
16  */
17 #define __pa(x)  ((unsigned long)(x))
18 #define __va(x)  ((void *)((unsigned long)(x)))
19 
20 /*
21  * Special hack: we have to be careful, because no indirections are
22  * allowed here, and paravirt_ops is a kind of one. As it will only run in
23  * baremetal anyway, we just keep it from happening. (This list needs to
24  * be extended when new paravirt and debugging variants are added.)
25  */
26 #undef CONFIG_PARAVIRT
27 #undef CONFIG_PARAVIRT_XXL
28 #undef CONFIG_PARAVIRT_SPINLOCKS
29 
30 #include <linux/kernel.h>
31 #include <linux/mm.h>
32 #include <linux/mem_encrypt.h>
33 
34 #include <asm/setup.h>
35 #include <asm/sections.h>
36 #include <asm/cmdline.h>
37 
38 #include "mm_internal.h"
39 
40 #define PGD_FLAGS		_KERNPG_TABLE_NOENC
41 #define P4D_FLAGS		_KERNPG_TABLE_NOENC
42 #define PUD_FLAGS		_KERNPG_TABLE_NOENC
43 #define PMD_FLAGS		_KERNPG_TABLE_NOENC
44 
45 #define PMD_FLAGS_LARGE		(__PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL)
46 
47 #define PMD_FLAGS_DEC		PMD_FLAGS_LARGE
48 #define PMD_FLAGS_DEC_WP	((PMD_FLAGS_DEC & ~_PAGE_CACHE_MASK) | \
49 				 (_PAGE_PAT | _PAGE_PWT))
50 
51 #define PMD_FLAGS_ENC		(PMD_FLAGS_LARGE | _PAGE_ENC)
52 
53 #define PTE_FLAGS		(__PAGE_KERNEL_EXEC & ~_PAGE_GLOBAL)
54 
55 #define PTE_FLAGS_DEC		PTE_FLAGS
56 #define PTE_FLAGS_DEC_WP	((PTE_FLAGS_DEC & ~_PAGE_CACHE_MASK) | \
57 				 (_PAGE_PAT | _PAGE_PWT))
58 
59 #define PTE_FLAGS_ENC		(PTE_FLAGS | _PAGE_ENC)
60 
61 struct sme_populate_pgd_data {
62 	void    *pgtable_area;
63 	pgd_t   *pgd;
64 
65 	pmdval_t pmd_flags;
66 	pteval_t pte_flags;
67 	unsigned long paddr;
68 
69 	unsigned long vaddr;
70 	unsigned long vaddr_end;
71 };
72 
73 /*
74  * This work area lives in the .init.scratch section, which lives outside of
75  * the kernel proper. It is sized to hold the intermediate copy buffer and
76  * more than enough pagetable pages.
77  *
78  * By using this section, the kernel can be encrypted in place and it
79  * avoids any possibility of boot parameters or initramfs images being
80  * placed such that the in-place encryption logic overwrites them.  This
81  * section is 2MB aligned to allow for simple pagetable setup using only
82  * PMD entries (see vmlinux.lds.S).
83  */
84 static char sme_workarea[2 * PMD_PAGE_SIZE] __section(.init.scratch);
85 
86 static char sme_cmdline_arg[] __initdata = "mem_encrypt";
87 static char sme_cmdline_on[]  __initdata = "on";
88 static char sme_cmdline_off[] __initdata = "off";
89 
90 static void __init sme_clear_pgd(struct sme_populate_pgd_data *ppd)
91 {
92 	unsigned long pgd_start, pgd_end, pgd_size;
93 	pgd_t *pgd_p;
94 
95 	pgd_start = ppd->vaddr & PGDIR_MASK;
96 	pgd_end = ppd->vaddr_end & PGDIR_MASK;
97 
98 	pgd_size = (((pgd_end - pgd_start) / PGDIR_SIZE) + 1) * sizeof(pgd_t);
99 
100 	pgd_p = ppd->pgd + pgd_index(ppd->vaddr);
101 
102 	memset(pgd_p, 0, pgd_size);
103 }
104 
105 static pud_t __init *sme_prepare_pgd(struct sme_populate_pgd_data *ppd)
106 {
107 	pgd_t *pgd;
108 	p4d_t *p4d;
109 	pud_t *pud;
110 	pmd_t *pmd;
111 
112 	pgd = ppd->pgd + pgd_index(ppd->vaddr);
113 	if (pgd_none(*pgd)) {
114 		p4d = ppd->pgtable_area;
115 		memset(p4d, 0, sizeof(*p4d) * PTRS_PER_P4D);
116 		ppd->pgtable_area += sizeof(*p4d) * PTRS_PER_P4D;
117 		set_pgd(pgd, __pgd(PGD_FLAGS | __pa(p4d)));
118 	}
119 
120 	p4d = p4d_offset(pgd, ppd->vaddr);
121 	if (p4d_none(*p4d)) {
122 		pud = ppd->pgtable_area;
123 		memset(pud, 0, sizeof(*pud) * PTRS_PER_PUD);
124 		ppd->pgtable_area += sizeof(*pud) * PTRS_PER_PUD;
125 		set_p4d(p4d, __p4d(P4D_FLAGS | __pa(pud)));
126 	}
127 
128 	pud = pud_offset(p4d, ppd->vaddr);
129 	if (pud_none(*pud)) {
130 		pmd = ppd->pgtable_area;
131 		memset(pmd, 0, sizeof(*pmd) * PTRS_PER_PMD);
132 		ppd->pgtable_area += sizeof(*pmd) * PTRS_PER_PMD;
133 		set_pud(pud, __pud(PUD_FLAGS | __pa(pmd)));
134 	}
135 
136 	if (pud_large(*pud))
137 		return NULL;
138 
139 	return pud;
140 }
141 
142 static void __init sme_populate_pgd_large(struct sme_populate_pgd_data *ppd)
143 {
144 	pud_t *pud;
145 	pmd_t *pmd;
146 
147 	pud = sme_prepare_pgd(ppd);
148 	if (!pud)
149 		return;
150 
151 	pmd = pmd_offset(pud, ppd->vaddr);
152 	if (pmd_large(*pmd))
153 		return;
154 
155 	set_pmd(pmd, __pmd(ppd->paddr | ppd->pmd_flags));
156 }
157 
158 static void __init sme_populate_pgd(struct sme_populate_pgd_data *ppd)
159 {
160 	pud_t *pud;
161 	pmd_t *pmd;
162 	pte_t *pte;
163 
164 	pud = sme_prepare_pgd(ppd);
165 	if (!pud)
166 		return;
167 
168 	pmd = pmd_offset(pud, ppd->vaddr);
169 	if (pmd_none(*pmd)) {
170 		pte = ppd->pgtable_area;
171 		memset(pte, 0, sizeof(*pte) * PTRS_PER_PTE);
172 		ppd->pgtable_area += sizeof(*pte) * PTRS_PER_PTE;
173 		set_pmd(pmd, __pmd(PMD_FLAGS | __pa(pte)));
174 	}
175 
176 	if (pmd_large(*pmd))
177 		return;
178 
179 	pte = pte_offset_map(pmd, ppd->vaddr);
180 	if (pte_none(*pte))
181 		set_pte(pte, __pte(ppd->paddr | ppd->pte_flags));
182 }
183 
184 static void __init __sme_map_range_pmd(struct sme_populate_pgd_data *ppd)
185 {
186 	while (ppd->vaddr < ppd->vaddr_end) {
187 		sme_populate_pgd_large(ppd);
188 
189 		ppd->vaddr += PMD_PAGE_SIZE;
190 		ppd->paddr += PMD_PAGE_SIZE;
191 	}
192 }
193 
194 static void __init __sme_map_range_pte(struct sme_populate_pgd_data *ppd)
195 {
196 	while (ppd->vaddr < ppd->vaddr_end) {
197 		sme_populate_pgd(ppd);
198 
199 		ppd->vaddr += PAGE_SIZE;
200 		ppd->paddr += PAGE_SIZE;
201 	}
202 }
203 
204 static void __init __sme_map_range(struct sme_populate_pgd_data *ppd,
205 				   pmdval_t pmd_flags, pteval_t pte_flags)
206 {
207 	unsigned long vaddr_end;
208 
209 	ppd->pmd_flags = pmd_flags;
210 	ppd->pte_flags = pte_flags;
211 
212 	/* Save original end value since we modify the struct value */
213 	vaddr_end = ppd->vaddr_end;
214 
215 	/* If start is not 2MB aligned, create PTE entries */
216 	ppd->vaddr_end = ALIGN(ppd->vaddr, PMD_PAGE_SIZE);
217 	__sme_map_range_pte(ppd);
218 
219 	/* Create PMD entries */
220 	ppd->vaddr_end = vaddr_end & PMD_PAGE_MASK;
221 	__sme_map_range_pmd(ppd);
222 
223 	/* If end is not 2MB aligned, create PTE entries */
224 	ppd->vaddr_end = vaddr_end;
225 	__sme_map_range_pte(ppd);
226 }
227 
228 static void __init sme_map_range_encrypted(struct sme_populate_pgd_data *ppd)
229 {
230 	__sme_map_range(ppd, PMD_FLAGS_ENC, PTE_FLAGS_ENC);
231 }
232 
233 static void __init sme_map_range_decrypted(struct sme_populate_pgd_data *ppd)
234 {
235 	__sme_map_range(ppd, PMD_FLAGS_DEC, PTE_FLAGS_DEC);
236 }
237 
238 static void __init sme_map_range_decrypted_wp(struct sme_populate_pgd_data *ppd)
239 {
240 	__sme_map_range(ppd, PMD_FLAGS_DEC_WP, PTE_FLAGS_DEC_WP);
241 }
242 
243 static unsigned long __init sme_pgtable_calc(unsigned long len)
244 {
245 	unsigned long entries = 0, tables = 0;
246 
247 	/*
248 	 * Perform a relatively simplistic calculation of the pagetable
249 	 * entries that are needed. Those mappings will be covered mostly
250 	 * by 2MB PMD entries so we can conservatively calculate the required
251 	 * number of P4D, PUD and PMD structures needed to perform the
252 	 * mappings.  For mappings that are not 2MB aligned, PTE mappings
253 	 * would be needed for the start and end portion of the address range
254 	 * that fall outside of the 2MB alignment.  This results in, at most,
255 	 * two extra pages to hold PTE entries for each range that is mapped.
256 	 * Incrementing the count for each covers the case where the addresses
257 	 * cross entries.
258 	 */
259 
260 	/* PGDIR_SIZE is equal to P4D_SIZE on 4-level machine. */
261 	if (PTRS_PER_P4D > 1)
262 		entries += (DIV_ROUND_UP(len, PGDIR_SIZE) + 1) * sizeof(p4d_t) * PTRS_PER_P4D;
263 	entries += (DIV_ROUND_UP(len, P4D_SIZE) + 1) * sizeof(pud_t) * PTRS_PER_PUD;
264 	entries += (DIV_ROUND_UP(len, PUD_SIZE) + 1) * sizeof(pmd_t) * PTRS_PER_PMD;
265 	entries += 2 * sizeof(pte_t) * PTRS_PER_PTE;
266 
267 	/*
268 	 * Now calculate the added pagetable structures needed to populate
269 	 * the new pagetables.
270 	 */
271 
272 	if (PTRS_PER_P4D > 1)
273 		tables += DIV_ROUND_UP(entries, PGDIR_SIZE) * sizeof(p4d_t) * PTRS_PER_P4D;
274 	tables += DIV_ROUND_UP(entries, P4D_SIZE) * sizeof(pud_t) * PTRS_PER_PUD;
275 	tables += DIV_ROUND_UP(entries, PUD_SIZE) * sizeof(pmd_t) * PTRS_PER_PMD;
276 
277 	return entries + tables;
278 }
279 
280 void __init sme_encrypt_kernel(struct boot_params *bp)
281 {
282 	unsigned long workarea_start, workarea_end, workarea_len;
283 	unsigned long execute_start, execute_end, execute_len;
284 	unsigned long kernel_start, kernel_end, kernel_len;
285 	unsigned long initrd_start, initrd_end, initrd_len;
286 	struct sme_populate_pgd_data ppd;
287 	unsigned long pgtable_area_len;
288 	unsigned long decrypted_base;
289 
290 	if (!sme_active())
291 		return;
292 
293 	/*
294 	 * Prepare for encrypting the kernel and initrd by building new
295 	 * pagetables with the necessary attributes needed to encrypt the
296 	 * kernel in place.
297 	 *
298 	 *   One range of virtual addresses will map the memory occupied
299 	 *   by the kernel and initrd as encrypted.
300 	 *
301 	 *   Another range of virtual addresses will map the memory occupied
302 	 *   by the kernel and initrd as decrypted and write-protected.
303 	 *
304 	 *     The use of write-protect attribute will prevent any of the
305 	 *     memory from being cached.
306 	 */
307 
308 	/* Physical addresses gives us the identity mapped virtual addresses */
309 	kernel_start = __pa_symbol(_text);
310 	kernel_end = ALIGN(__pa_symbol(_end), PMD_PAGE_SIZE);
311 	kernel_len = kernel_end - kernel_start;
312 
313 	initrd_start = 0;
314 	initrd_end = 0;
315 	initrd_len = 0;
316 #ifdef CONFIG_BLK_DEV_INITRD
317 	initrd_len = (unsigned long)bp->hdr.ramdisk_size |
318 		     ((unsigned long)bp->ext_ramdisk_size << 32);
319 	if (initrd_len) {
320 		initrd_start = (unsigned long)bp->hdr.ramdisk_image |
321 			       ((unsigned long)bp->ext_ramdisk_image << 32);
322 		initrd_end = PAGE_ALIGN(initrd_start + initrd_len);
323 		initrd_len = initrd_end - initrd_start;
324 	}
325 #endif
326 
327 	/*
328 	 * We're running identity mapped, so we must obtain the address to the
329 	 * SME encryption workarea using rip-relative addressing.
330 	 */
331 	asm ("lea sme_workarea(%%rip), %0"
332 	     : "=r" (workarea_start)
333 	     : "p" (sme_workarea));
334 
335 	/*
336 	 * Calculate required number of workarea bytes needed:
337 	 *   executable encryption area size:
338 	 *     stack page (PAGE_SIZE)
339 	 *     encryption routine page (PAGE_SIZE)
340 	 *     intermediate copy buffer (PMD_PAGE_SIZE)
341 	 *   pagetable structures for the encryption of the kernel
342 	 *   pagetable structures for workarea (in case not currently mapped)
343 	 */
344 	execute_start = workarea_start;
345 	execute_end = execute_start + (PAGE_SIZE * 2) + PMD_PAGE_SIZE;
346 	execute_len = execute_end - execute_start;
347 
348 	/*
349 	 * One PGD for both encrypted and decrypted mappings and a set of
350 	 * PUDs and PMDs for each of the encrypted and decrypted mappings.
351 	 */
352 	pgtable_area_len = sizeof(pgd_t) * PTRS_PER_PGD;
353 	pgtable_area_len += sme_pgtable_calc(execute_end - kernel_start) * 2;
354 	if (initrd_len)
355 		pgtable_area_len += sme_pgtable_calc(initrd_len) * 2;
356 
357 	/* PUDs and PMDs needed in the current pagetables for the workarea */
358 	pgtable_area_len += sme_pgtable_calc(execute_len + pgtable_area_len);
359 
360 	/*
361 	 * The total workarea includes the executable encryption area and
362 	 * the pagetable area. The start of the workarea is already 2MB
363 	 * aligned, align the end of the workarea on a 2MB boundary so that
364 	 * we don't try to create/allocate PTE entries from the workarea
365 	 * before it is mapped.
366 	 */
367 	workarea_len = execute_len + pgtable_area_len;
368 	workarea_end = ALIGN(workarea_start + workarea_len, PMD_PAGE_SIZE);
369 
370 	/*
371 	 * Set the address to the start of where newly created pagetable
372 	 * structures (PGDs, PUDs and PMDs) will be allocated. New pagetable
373 	 * structures are created when the workarea is added to the current
374 	 * pagetables and when the new encrypted and decrypted kernel
375 	 * mappings are populated.
376 	 */
377 	ppd.pgtable_area = (void *)execute_end;
378 
379 	/*
380 	 * Make sure the current pagetable structure has entries for
381 	 * addressing the workarea.
382 	 */
383 	ppd.pgd = (pgd_t *)native_read_cr3_pa();
384 	ppd.paddr = workarea_start;
385 	ppd.vaddr = workarea_start;
386 	ppd.vaddr_end = workarea_end;
387 	sme_map_range_decrypted(&ppd);
388 
389 	/* Flush the TLB - no globals so cr3 is enough */
390 	native_write_cr3(__native_read_cr3());
391 
392 	/*
393 	 * A new pagetable structure is being built to allow for the kernel
394 	 * and initrd to be encrypted. It starts with an empty PGD that will
395 	 * then be populated with new PUDs and PMDs as the encrypted and
396 	 * decrypted kernel mappings are created.
397 	 */
398 	ppd.pgd = ppd.pgtable_area;
399 	memset(ppd.pgd, 0, sizeof(pgd_t) * PTRS_PER_PGD);
400 	ppd.pgtable_area += sizeof(pgd_t) * PTRS_PER_PGD;
401 
402 	/*
403 	 * A different PGD index/entry must be used to get different
404 	 * pagetable entries for the decrypted mapping. Choose the next
405 	 * PGD index and convert it to a virtual address to be used as
406 	 * the base of the mapping.
407 	 */
408 	decrypted_base = (pgd_index(workarea_end) + 1) & (PTRS_PER_PGD - 1);
409 	if (initrd_len) {
410 		unsigned long check_base;
411 
412 		check_base = (pgd_index(initrd_end) + 1) & (PTRS_PER_PGD - 1);
413 		decrypted_base = max(decrypted_base, check_base);
414 	}
415 	decrypted_base <<= PGDIR_SHIFT;
416 
417 	/* Add encrypted kernel (identity) mappings */
418 	ppd.paddr = kernel_start;
419 	ppd.vaddr = kernel_start;
420 	ppd.vaddr_end = kernel_end;
421 	sme_map_range_encrypted(&ppd);
422 
423 	/* Add decrypted, write-protected kernel (non-identity) mappings */
424 	ppd.paddr = kernel_start;
425 	ppd.vaddr = kernel_start + decrypted_base;
426 	ppd.vaddr_end = kernel_end + decrypted_base;
427 	sme_map_range_decrypted_wp(&ppd);
428 
429 	if (initrd_len) {
430 		/* Add encrypted initrd (identity) mappings */
431 		ppd.paddr = initrd_start;
432 		ppd.vaddr = initrd_start;
433 		ppd.vaddr_end = initrd_end;
434 		sme_map_range_encrypted(&ppd);
435 		/*
436 		 * Add decrypted, write-protected initrd (non-identity) mappings
437 		 */
438 		ppd.paddr = initrd_start;
439 		ppd.vaddr = initrd_start + decrypted_base;
440 		ppd.vaddr_end = initrd_end + decrypted_base;
441 		sme_map_range_decrypted_wp(&ppd);
442 	}
443 
444 	/* Add decrypted workarea mappings to both kernel mappings */
445 	ppd.paddr = workarea_start;
446 	ppd.vaddr = workarea_start;
447 	ppd.vaddr_end = workarea_end;
448 	sme_map_range_decrypted(&ppd);
449 
450 	ppd.paddr = workarea_start;
451 	ppd.vaddr = workarea_start + decrypted_base;
452 	ppd.vaddr_end = workarea_end + decrypted_base;
453 	sme_map_range_decrypted(&ppd);
454 
455 	/* Perform the encryption */
456 	sme_encrypt_execute(kernel_start, kernel_start + decrypted_base,
457 			    kernel_len, workarea_start, (unsigned long)ppd.pgd);
458 
459 	if (initrd_len)
460 		sme_encrypt_execute(initrd_start, initrd_start + decrypted_base,
461 				    initrd_len, workarea_start,
462 				    (unsigned long)ppd.pgd);
463 
464 	/*
465 	 * At this point we are running encrypted.  Remove the mappings for
466 	 * the decrypted areas - all that is needed for this is to remove
467 	 * the PGD entry/entries.
468 	 */
469 	ppd.vaddr = kernel_start + decrypted_base;
470 	ppd.vaddr_end = kernel_end + decrypted_base;
471 	sme_clear_pgd(&ppd);
472 
473 	if (initrd_len) {
474 		ppd.vaddr = initrd_start + decrypted_base;
475 		ppd.vaddr_end = initrd_end + decrypted_base;
476 		sme_clear_pgd(&ppd);
477 	}
478 
479 	ppd.vaddr = workarea_start + decrypted_base;
480 	ppd.vaddr_end = workarea_end + decrypted_base;
481 	sme_clear_pgd(&ppd);
482 
483 	/* Flush the TLB - no globals so cr3 is enough */
484 	native_write_cr3(__native_read_cr3());
485 }
486 
487 void __init sme_enable(struct boot_params *bp)
488 {
489 	const char *cmdline_ptr, *cmdline_arg, *cmdline_on, *cmdline_off;
490 	unsigned int eax, ebx, ecx, edx;
491 	unsigned long feature_mask;
492 	bool active_by_default;
493 	unsigned long me_mask;
494 	char buffer[16];
495 	u64 msr;
496 
497 	/* Check for the SME/SEV support leaf */
498 	eax = 0x80000000;
499 	ecx = 0;
500 	native_cpuid(&eax, &ebx, &ecx, &edx);
501 	if (eax < 0x8000001f)
502 		return;
503 
504 #define AMD_SME_BIT	BIT(0)
505 #define AMD_SEV_BIT	BIT(1)
506 	/*
507 	 * Set the feature mask (SME or SEV) based on whether we are
508 	 * running under a hypervisor.
509 	 */
510 	eax = 1;
511 	ecx = 0;
512 	native_cpuid(&eax, &ebx, &ecx, &edx);
513 	feature_mask = (ecx & BIT(31)) ? AMD_SEV_BIT : AMD_SME_BIT;
514 
515 	/*
516 	 * Check for the SME/SEV feature:
517 	 *   CPUID Fn8000_001F[EAX]
518 	 *   - Bit 0 - Secure Memory Encryption support
519 	 *   - Bit 1 - Secure Encrypted Virtualization support
520 	 *   CPUID Fn8000_001F[EBX]
521 	 *   - Bits 5:0 - Pagetable bit position used to indicate encryption
522 	 */
523 	eax = 0x8000001f;
524 	ecx = 0;
525 	native_cpuid(&eax, &ebx, &ecx, &edx);
526 	if (!(eax & feature_mask))
527 		return;
528 
529 	me_mask = 1UL << (ebx & 0x3f);
530 
531 	/* Check if memory encryption is enabled */
532 	if (feature_mask == AMD_SME_BIT) {
533 		/* For SME, check the SYSCFG MSR */
534 		msr = __rdmsr(MSR_K8_SYSCFG);
535 		if (!(msr & MSR_K8_SYSCFG_MEM_ENCRYPT))
536 			return;
537 	} else {
538 		/* For SEV, check the SEV MSR */
539 		msr = __rdmsr(MSR_AMD64_SEV);
540 		if (!(msr & MSR_AMD64_SEV_ENABLED))
541 			return;
542 
543 		/* SEV state cannot be controlled by a command line option */
544 		sme_me_mask = me_mask;
545 		sev_enabled = true;
546 		physical_mask &= ~sme_me_mask;
547 		return;
548 	}
549 
550 	/*
551 	 * Fixups have not been applied to phys_base yet and we're running
552 	 * identity mapped, so we must obtain the address to the SME command
553 	 * line argument data using rip-relative addressing.
554 	 */
555 	asm ("lea sme_cmdline_arg(%%rip), %0"
556 	     : "=r" (cmdline_arg)
557 	     : "p" (sme_cmdline_arg));
558 	asm ("lea sme_cmdline_on(%%rip), %0"
559 	     : "=r" (cmdline_on)
560 	     : "p" (sme_cmdline_on));
561 	asm ("lea sme_cmdline_off(%%rip), %0"
562 	     : "=r" (cmdline_off)
563 	     : "p" (sme_cmdline_off));
564 
565 	if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT))
566 		active_by_default = true;
567 	else
568 		active_by_default = false;
569 
570 	cmdline_ptr = (const char *)((u64)bp->hdr.cmd_line_ptr |
571 				     ((u64)bp->ext_cmd_line_ptr << 32));
572 
573 	cmdline_find_option(cmdline_ptr, cmdline_arg, buffer, sizeof(buffer));
574 
575 	if (!strncmp(buffer, cmdline_on, sizeof(buffer)))
576 		sme_me_mask = me_mask;
577 	else if (!strncmp(buffer, cmdline_off, sizeof(buffer)))
578 		sme_me_mask = 0;
579 	else
580 		sme_me_mask = active_by_default ? me_mask : 0;
581 
582 	physical_mask &= ~sme_me_mask;
583 }
584