Linux Audio

Check our new training course

Loading...
v5.4
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  prepare to run common code
  4 *
  5 *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
  6 */
  7
  8#define DISABLE_BRANCH_PROFILING
  9
 10/* cpu_feature_enabled() cannot be used this early */
 11#define USE_EARLY_PGTABLE_L5
 12
 13#include <linux/init.h>
 14#include <linux/linkage.h>
 15#include <linux/types.h>
 16#include <linux/kernel.h>
 17#include <linux/string.h>
 18#include <linux/percpu.h>
 19#include <linux/start_kernel.h>
 20#include <linux/io.h>
 21#include <linux/memblock.h>
 22#include <linux/mem_encrypt.h>
 
 23
 24#include <asm/processor.h>
 25#include <asm/proto.h>
 26#include <asm/smp.h>
 27#include <asm/setup.h>
 28#include <asm/desc.h>
 29#include <asm/pgtable.h>
 30#include <asm/tlbflush.h>
 31#include <asm/sections.h>
 32#include <asm/kdebug.h>
 33#include <asm/e820/api.h>
 34#include <asm/bios_ebda.h>
 35#include <asm/bootparam_utils.h>
 36#include <asm/microcode.h>
 37#include <asm/kasan.h>
 38#include <asm/fixmap.h>
 
 
 
 
 
 39
 40/*
 41 * Manage page tables very early on.
 42 */
 43extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];
 44static unsigned int __initdata next_early_pgt;
 45pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
 46
 47#ifdef CONFIG_X86_5LEVEL
 48unsigned int __pgtable_l5_enabled __ro_after_init;
 49unsigned int pgdir_shift __ro_after_init = 39;
 50EXPORT_SYMBOL(pgdir_shift);
 51unsigned int ptrs_per_p4d __ro_after_init = 1;
 52EXPORT_SYMBOL(ptrs_per_p4d);
 53#endif
 54
 55#ifdef CONFIG_DYNAMIC_MEMORY_LAYOUT
 56unsigned long page_offset_base __ro_after_init = __PAGE_OFFSET_BASE_L4;
 57EXPORT_SYMBOL(page_offset_base);
 58unsigned long vmalloc_base __ro_after_init = __VMALLOC_BASE_L4;
 59EXPORT_SYMBOL(vmalloc_base);
 60unsigned long vmemmap_base __ro_after_init = __VMEMMAP_BASE_L4;
 61EXPORT_SYMBOL(vmemmap_base);
 62#endif
 63
 64#define __head	__section(.head.text)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 65
 66static void __head *fixup_pointer(void *ptr, unsigned long physaddr)
 67{
 68	return ptr - (void *)_text + (void *)physaddr;
 69}
 70
 71static unsigned long __head *fixup_long(void *ptr, unsigned long physaddr)
 72{
 73	return fixup_pointer(ptr, physaddr);
 74}
 75
 76#ifdef CONFIG_X86_5LEVEL
 77static unsigned int __head *fixup_int(void *ptr, unsigned long physaddr)
 78{
 79	return fixup_pointer(ptr, physaddr);
 80}
 81
 82static bool __head check_la57_support(unsigned long physaddr)
 83{
 84	/*
 85	 * 5-level paging is detected and enabled at kernel decomression
 86	 * stage. Only check if it has been enabled there.
 87	 */
 88	if (!(native_read_cr4() & X86_CR4_LA57))
 89		return false;
 90
 91	*fixup_int(&__pgtable_l5_enabled, physaddr) = 1;
 92	*fixup_int(&pgdir_shift, physaddr) = 48;
 93	*fixup_int(&ptrs_per_p4d, physaddr) = 512;
 94	*fixup_long(&page_offset_base, physaddr) = __PAGE_OFFSET_BASE_L5;
 95	*fixup_long(&vmalloc_base, physaddr) = __VMALLOC_BASE_L5;
 96	*fixup_long(&vmemmap_base, physaddr) = __VMEMMAP_BASE_L5;
 97
 98	return true;
 99}
100#else
101static bool __head check_la57_support(unsigned long physaddr)
102{
103	return false;
104}
105#endif
106
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
107/* Code in __startup_64() can be relocated during execution, but the compiler
108 * doesn't have to generate PC-relative relocations when accessing globals from
109 * that function. Clang actually does not generate them, which leads to
110 * boot-time crashes. To work around this problem, every global pointer must
111 * be adjusted using fixup_pointer().
112 */
113unsigned long __head __startup_64(unsigned long physaddr,
114				  struct boot_params *bp)
115{
116	unsigned long vaddr, vaddr_end;
117	unsigned long load_delta, *p;
118	unsigned long pgtable_flags;
119	pgdval_t *pgd;
120	p4dval_t *p4d;
121	pudval_t *pud;
122	pmdval_t *pmd, pmd_entry;
123	pteval_t *mask_ptr;
124	bool la57;
125	int i;
126	unsigned int *next_pgt_ptr;
127
128	la57 = check_la57_support(physaddr);
129
130	/* Is the address too large? */
131	if (physaddr >> MAX_PHYSMEM_BITS)
132		for (;;);
133
134	/*
135	 * Compute the delta between the address I am compiled to run at
136	 * and the address I am actually running at.
137	 */
138	load_delta = physaddr - (unsigned long)(_text - __START_KERNEL_map);
139
140	/* Is the address not 2M aligned? */
141	if (load_delta & ~PMD_PAGE_MASK)
142		for (;;);
143
144	/* Activate Secure Memory Encryption (SME) if supported and enabled */
145	sme_enable(bp);
146
147	/* Include the SME encryption mask in the fixup value */
148	load_delta += sme_get_me_mask();
149
150	/* Fixup the physical addresses in the page table */
151
152	pgd = fixup_pointer(&early_top_pgt, physaddr);
153	p = pgd + pgd_index(__START_KERNEL_map);
154	if (la57)
155		*p = (unsigned long)level4_kernel_pgt;
156	else
157		*p = (unsigned long)level3_kernel_pgt;
158	*p += _PAGE_TABLE_NOENC - __START_KERNEL_map + load_delta;
159
160	if (la57) {
161		p4d = fixup_pointer(&level4_kernel_pgt, physaddr);
162		p4d[511] += load_delta;
163	}
164
165	pud = fixup_pointer(&level3_kernel_pgt, physaddr);
166	pud[510] += load_delta;
167	pud[511] += load_delta;
168
169	pmd = fixup_pointer(level2_fixmap_pgt, physaddr);
170	for (i = FIXMAP_PMD_TOP; i > FIXMAP_PMD_TOP - FIXMAP_PMD_NUM; i--)
171		pmd[i] += load_delta;
172
173	/*
174	 * Set up the identity mapping for the switchover.  These
175	 * entries should *NOT* have the global bit set!  This also
176	 * creates a bunch of nonsense entries but that is fine --
177	 * it avoids problems around wraparound.
178	 */
179
180	next_pgt_ptr = fixup_pointer(&next_early_pgt, physaddr);
181	pud = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr);
182	pmd = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr);
183
184	pgtable_flags = _KERNPG_TABLE_NOENC + sme_get_me_mask();
185
186	if (la57) {
187		p4d = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++],
188				    physaddr);
189
190		i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
191		pgd[i + 0] = (pgdval_t)p4d + pgtable_flags;
192		pgd[i + 1] = (pgdval_t)p4d + pgtable_flags;
193
194		i = physaddr >> P4D_SHIFT;
195		p4d[(i + 0) % PTRS_PER_P4D] = (pgdval_t)pud + pgtable_flags;
196		p4d[(i + 1) % PTRS_PER_P4D] = (pgdval_t)pud + pgtable_flags;
197	} else {
198		i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
199		pgd[i + 0] = (pgdval_t)pud + pgtable_flags;
200		pgd[i + 1] = (pgdval_t)pud + pgtable_flags;
201	}
202
203	i = physaddr >> PUD_SHIFT;
204	pud[(i + 0) % PTRS_PER_PUD] = (pudval_t)pmd + pgtable_flags;
205	pud[(i + 1) % PTRS_PER_PUD] = (pudval_t)pmd + pgtable_flags;
206
207	pmd_entry = __PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL;
208	/* Filter out unsupported __PAGE_KERNEL_* bits: */
209	mask_ptr = fixup_pointer(&__supported_pte_mask, physaddr);
210	pmd_entry &= *mask_ptr;
211	pmd_entry += sme_get_me_mask();
212	pmd_entry +=  physaddr;
213
214	for (i = 0; i < DIV_ROUND_UP(_end - _text, PMD_SIZE); i++) {
215		int idx = i + (physaddr >> PMD_SHIFT);
216
217		pmd[idx % PTRS_PER_PMD] = pmd_entry + i * PMD_SIZE;
218	}
219
220	/*
221	 * Fixup the kernel text+data virtual addresses. Note that
222	 * we might write invalid pmds, when the kernel is relocated
223	 * cleanup_highmap() fixes this up along with the mappings
224	 * beyond _end.
225	 *
226	 * Only the region occupied by the kernel image has so far
227	 * been checked against the table of usable memory regions
228	 * provided by the firmware, so invalidate pages outside that
229	 * region. A page table entry that maps to a reserved area of
230	 * memory would allow processor speculation into that area,
231	 * and on some hardware (particularly the UV platform) even
232	 * speculative access to some reserved areas is caught as an
233	 * error, causing the BIOS to halt the system.
234	 */
235
236	pmd = fixup_pointer(level2_kernel_pgt, physaddr);
237
238	/* invalidate pages before the kernel image */
239	for (i = 0; i < pmd_index((unsigned long)_text); i++)
240		pmd[i] &= ~_PAGE_PRESENT;
241
242	/* fixup pages that are part of the kernel image */
243	for (; i <= pmd_index((unsigned long)_end); i++)
244		if (pmd[i] & _PAGE_PRESENT)
245			pmd[i] += load_delta;
246
247	/* invalidate pages after the kernel image */
248	for (; i < PTRS_PER_PMD; i++)
249		pmd[i] &= ~_PAGE_PRESENT;
250
251	/*
252	 * Fixup phys_base - remove the memory encryption mask to obtain
253	 * the true physical address.
254	 */
255	*fixup_long(&phys_base, physaddr) += load_delta - sme_get_me_mask();
256
257	/* Encrypt the kernel and related (if SME is active) */
258	sme_encrypt_kernel(bp);
259
260	/*
261	 * Clear the memory encryption mask from the .bss..decrypted section.
262	 * The bss section will be memset to zero later in the initialization so
263	 * there is no need to zero it after changing the memory encryption
264	 * attribute.
265	 */
266	if (mem_encrypt_active()) {
267		vaddr = (unsigned long)__start_bss_decrypted;
268		vaddr_end = (unsigned long)__end_bss_decrypted;
269		for (; vaddr < vaddr_end; vaddr += PMD_SIZE) {
270			i = pmd_index(vaddr);
271			pmd[i] -= sme_get_me_mask();
272		}
273	}
274
275	/*
276	 * Return the SME encryption mask (if SME is active) to be used as a
277	 * modifier for the initial pgdir entry programmed into CR3.
278	 */
279	return sme_get_me_mask();
280}
281
282unsigned long __startup_secondary_64(void)
283{
284	/*
285	 * Return the SME encryption mask (if SME is active) to be used as a
286	 * modifier for the initial pgdir entry programmed into CR3.
287	 */
288	return sme_get_me_mask();
289}
290
291/* Wipe all early page tables except for the kernel symbol map */
292static void __init reset_early_page_tables(void)
293{
294	memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
295	next_early_pgt = 0;
296	write_cr3(__sme_pa_nodebug(early_top_pgt));
297}
298
299/* Create a new PMD entry */
300int __init __early_make_pgtable(unsigned long address, pmdval_t pmd)
301{
302	unsigned long physaddr = address - __PAGE_OFFSET;
303	pgdval_t pgd, *pgd_p;
304	p4dval_t p4d, *p4d_p;
305	pudval_t pud, *pud_p;
306	pmdval_t *pmd_p;
307
308	/* Invalid address or early pgt is done ?  */
309	if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt))
310		return -1;
311
312again:
313	pgd_p = &early_top_pgt[pgd_index(address)].pgd;
314	pgd = *pgd_p;
315
316	/*
317	 * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
318	 * critical -- __PAGE_OFFSET would point us back into the dynamic
319	 * range and we might end up looping forever...
320	 */
321	if (!pgtable_l5_enabled())
322		p4d_p = pgd_p;
323	else if (pgd)
324		p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
325	else {
326		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
327			reset_early_page_tables();
328			goto again;
329		}
330
331		p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++];
332		memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D);
333		*pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
334	}
335	p4d_p += p4d_index(address);
336	p4d = *p4d_p;
337
338	if (p4d)
339		pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
340	else {
341		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
342			reset_early_page_tables();
343			goto again;
344		}
345
346		pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
347		memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
348		*p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
349	}
350	pud_p += pud_index(address);
351	pud = *pud_p;
352
353	if (pud)
354		pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
355	else {
356		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
357			reset_early_page_tables();
358			goto again;
359		}
360
361		pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
362		memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
363		*pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
364	}
365	pmd_p[pmd_index(address)] = pmd;
366
367	return 0;
368}
369
370int __init early_make_pgtable(unsigned long address)
371{
372	unsigned long physaddr = address - __PAGE_OFFSET;
373	pmdval_t pmd;
374
375	pmd = (physaddr & PMD_MASK) + early_pmd_flags;
376
377	return __early_make_pgtable(address, pmd);
378}
379
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
380/* Don't add a printk in there. printk relies on the PDA which is not initialized 
381   yet. */
382static void __init clear_bss(void)
383{
384	memset(__bss_start, 0,
385	       (unsigned long) __bss_stop - (unsigned long) __bss_start);
 
 
386}
387
388static unsigned long get_cmd_line_ptr(void)
389{
390	unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
391
392	cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
393
394	return cmd_line_ptr;
395}
396
397static void __init copy_bootdata(char *real_mode_data)
398{
399	char * command_line;
400	unsigned long cmd_line_ptr;
401
402	/*
403	 * If SME is active, this will create decrypted mappings of the
404	 * boot data in advance of the copy operations.
405	 */
406	sme_map_bootdata(real_mode_data);
407
408	memcpy(&boot_params, real_mode_data, sizeof(boot_params));
409	sanitize_boot_params(&boot_params);
410	cmd_line_ptr = get_cmd_line_ptr();
411	if (cmd_line_ptr) {
412		command_line = __va(cmd_line_ptr);
413		memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
414	}
415
416	/*
417	 * The old boot data is no longer needed and won't be reserved,
418	 * freeing up that memory for use by the system. If SME is active,
419	 * we need to remove the mappings that were created so that the
420	 * memory doesn't remain mapped as decrypted.
421	 */
422	sme_unmap_bootdata(real_mode_data);
423}
424
425asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
426{
427	/*
428	 * Build-time sanity checks on the kernel image and module
429	 * area mappings. (these are purely build-time and produce no code)
430	 */
431	BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
432	BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
433	BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
434	BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
435	BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
436	BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
437	MAYBE_BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
438				(__START_KERNEL & PGDIR_MASK)));
439	BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
440
441	cr4_init_shadow();
442
443	/* Kill off the identity-map trampoline */
444	reset_early_page_tables();
445
446	clear_bss();
447
 
 
 
 
448	clear_page(init_top_pgt);
449
450	/*
451	 * SME support may update early_pmd_flags to include the memory
452	 * encryption mask, so it needs to be called before anything
453	 * that may generate a page fault.
454	 */
455	sme_early_init();
456
457	kasan_early_init();
458
 
 
 
 
 
 
 
 
 
 
459	idt_setup_early_handler();
460
 
 
 
461	copy_bootdata(__va(real_mode_data));
462
463	/*
464	 * Load microcode early on BSP.
465	 */
466	load_ucode_bsp();
467
468	/* set init_top_pgt kernel high mapping*/
469	init_top_pgt[511] = early_top_pgt[511];
470
471	x86_64_start_reservations(real_mode_data);
472}
473
474void __init x86_64_start_reservations(char *real_mode_data)
475{
476	/* version is always not zero if it is copied */
477	if (!boot_params.hdr.version)
478		copy_bootdata(__va(real_mode_data));
479
480	x86_early_init_platform_quirks();
481
482	switch (boot_params.hdr.hardware_subarch) {
483	case X86_SUBARCH_INTEL_MID:
484		x86_intel_mid_early_setup();
485		break;
486	default:
487		break;
488	}
489
490	start_kernel();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
491}
v6.2
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  prepare to run common code
  4 *
  5 *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
  6 */
  7
  8#define DISABLE_BRANCH_PROFILING
  9
 10/* cpu_feature_enabled() cannot be used this early */
 11#define USE_EARLY_PGTABLE_L5
 12
 13#include <linux/init.h>
 14#include <linux/linkage.h>
 15#include <linux/types.h>
 16#include <linux/kernel.h>
 17#include <linux/string.h>
 18#include <linux/percpu.h>
 19#include <linux/start_kernel.h>
 20#include <linux/io.h>
 21#include <linux/memblock.h>
 22#include <linux/cc_platform.h>
 23#include <linux/pgtable.h>
 24
 25#include <asm/processor.h>
 26#include <asm/proto.h>
 27#include <asm/smp.h>
 28#include <asm/setup.h>
 29#include <asm/desc.h>
 
 30#include <asm/tlbflush.h>
 31#include <asm/sections.h>
 32#include <asm/kdebug.h>
 33#include <asm/e820/api.h>
 34#include <asm/bios_ebda.h>
 35#include <asm/bootparam_utils.h>
 36#include <asm/microcode.h>
 37#include <asm/kasan.h>
 38#include <asm/fixmap.h>
 39#include <asm/realmode.h>
 40#include <asm/extable.h>
 41#include <asm/trapnr.h>
 42#include <asm/sev.h>
 43#include <asm/tdx.h>
 44
 45/*
 46 * Manage page tables very early on.
 47 */
 48extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];
 49static unsigned int __initdata next_early_pgt;
 50pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
 51
 52#ifdef CONFIG_X86_5LEVEL
 53unsigned int __pgtable_l5_enabled __ro_after_init;
 54unsigned int pgdir_shift __ro_after_init = 39;
 55EXPORT_SYMBOL(pgdir_shift);
 56unsigned int ptrs_per_p4d __ro_after_init = 1;
 57EXPORT_SYMBOL(ptrs_per_p4d);
 58#endif
 59
 60#ifdef CONFIG_DYNAMIC_MEMORY_LAYOUT
 61unsigned long page_offset_base __ro_after_init = __PAGE_OFFSET_BASE_L4;
 62EXPORT_SYMBOL(page_offset_base);
 63unsigned long vmalloc_base __ro_after_init = __VMALLOC_BASE_L4;
 64EXPORT_SYMBOL(vmalloc_base);
 65unsigned long vmemmap_base __ro_after_init = __VMEMMAP_BASE_L4;
 66EXPORT_SYMBOL(vmemmap_base);
 67#endif
 68
 69/*
 70 * GDT used on the boot CPU before switching to virtual addresses.
 71 */
 72static struct desc_struct startup_gdt[GDT_ENTRIES] = {
 73	[GDT_ENTRY_KERNEL32_CS]         = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
 74	[GDT_ENTRY_KERNEL_CS]           = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
 75	[GDT_ENTRY_KERNEL_DS]           = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
 76};
 77
 78/*
 79 * Address needs to be set at runtime because it references the startup_gdt
 80 * while the kernel still uses a direct mapping.
 81 */
 82static struct desc_ptr startup_gdt_descr = {
 83	.size = sizeof(startup_gdt),
 84	.address = 0,
 85};
 86
 87#define __head	__section(".head.text")
 88
 89static void __head *fixup_pointer(void *ptr, unsigned long physaddr)
 90{
 91	return ptr - (void *)_text + (void *)physaddr;
 92}
 93
 94static unsigned long __head *fixup_long(void *ptr, unsigned long physaddr)
 95{
 96	return fixup_pointer(ptr, physaddr);
 97}
 98
 99#ifdef CONFIG_X86_5LEVEL
100static unsigned int __head *fixup_int(void *ptr, unsigned long physaddr)
101{
102	return fixup_pointer(ptr, physaddr);
103}
104
105static bool __head check_la57_support(unsigned long physaddr)
106{
107	/*
108	 * 5-level paging is detected and enabled at kernel decompression
109	 * stage. Only check if it has been enabled there.
110	 */
111	if (!(native_read_cr4() & X86_CR4_LA57))
112		return false;
113
114	*fixup_int(&__pgtable_l5_enabled, physaddr) = 1;
115	*fixup_int(&pgdir_shift, physaddr) = 48;
116	*fixup_int(&ptrs_per_p4d, physaddr) = 512;
117	*fixup_long(&page_offset_base, physaddr) = __PAGE_OFFSET_BASE_L5;
118	*fixup_long(&vmalloc_base, physaddr) = __VMALLOC_BASE_L5;
119	*fixup_long(&vmemmap_base, physaddr) = __VMEMMAP_BASE_L5;
120
121	return true;
122}
123#else
124static bool __head check_la57_support(unsigned long physaddr)
125{
126	return false;
127}
128#endif
129
130static unsigned long __head sme_postprocess_startup(struct boot_params *bp, pmdval_t *pmd)
131{
132	unsigned long vaddr, vaddr_end;
133	int i;
134
135	/* Encrypt the kernel and related (if SME is active) */
136	sme_encrypt_kernel(bp);
137
138	/*
139	 * Clear the memory encryption mask from the .bss..decrypted section.
140	 * The bss section will be memset to zero later in the initialization so
141	 * there is no need to zero it after changing the memory encryption
142	 * attribute.
143	 */
144	if (sme_get_me_mask()) {
145		vaddr = (unsigned long)__start_bss_decrypted;
146		vaddr_end = (unsigned long)__end_bss_decrypted;
147
148		for (; vaddr < vaddr_end; vaddr += PMD_SIZE) {
149			/*
150			 * On SNP, transition the page to shared in the RMP table so that
151			 * it is consistent with the page table attribute change.
152			 *
153			 * __start_bss_decrypted has a virtual address in the high range
154			 * mapping (kernel .text). PVALIDATE, by way of
155			 * early_snp_set_memory_shared(), requires a valid virtual
156			 * address but the kernel is currently running off of the identity
157			 * mapping so use __pa() to get a *currently* valid virtual address.
158			 */
159			early_snp_set_memory_shared(__pa(vaddr), __pa(vaddr), PTRS_PER_PMD);
160
161			i = pmd_index(vaddr);
162			pmd[i] -= sme_get_me_mask();
163		}
164	}
165
166	/*
167	 * Return the SME encryption mask (if SME is active) to be used as a
168	 * modifier for the initial pgdir entry programmed into CR3.
169	 */
170	return sme_get_me_mask();
171}
172
173/* Code in __startup_64() can be relocated during execution, but the compiler
174 * doesn't have to generate PC-relative relocations when accessing globals from
175 * that function. Clang actually does not generate them, which leads to
176 * boot-time crashes. To work around this problem, every global pointer must
177 * be adjusted using fixup_pointer().
178 */
179unsigned long __head __startup_64(unsigned long physaddr,
180				  struct boot_params *bp)
181{
 
182	unsigned long load_delta, *p;
183	unsigned long pgtable_flags;
184	pgdval_t *pgd;
185	p4dval_t *p4d;
186	pudval_t *pud;
187	pmdval_t *pmd, pmd_entry;
188	pteval_t *mask_ptr;
189	bool la57;
190	int i;
191	unsigned int *next_pgt_ptr;
192
193	la57 = check_la57_support(physaddr);
194
195	/* Is the address too large? */
196	if (physaddr >> MAX_PHYSMEM_BITS)
197		for (;;);
198
199	/*
200	 * Compute the delta between the address I am compiled to run at
201	 * and the address I am actually running at.
202	 */
203	load_delta = physaddr - (unsigned long)(_text - __START_KERNEL_map);
204
205	/* Is the address not 2M aligned? */
206	if (load_delta & ~PMD_MASK)
207		for (;;);
208
 
 
 
209	/* Include the SME encryption mask in the fixup value */
210	load_delta += sme_get_me_mask();
211
212	/* Fixup the physical addresses in the page table */
213
214	pgd = fixup_pointer(&early_top_pgt, physaddr);
215	p = pgd + pgd_index(__START_KERNEL_map);
216	if (la57)
217		*p = (unsigned long)level4_kernel_pgt;
218	else
219		*p = (unsigned long)level3_kernel_pgt;
220	*p += _PAGE_TABLE_NOENC - __START_KERNEL_map + load_delta;
221
222	if (la57) {
223		p4d = fixup_pointer(&level4_kernel_pgt, physaddr);
224		p4d[511] += load_delta;
225	}
226
227	pud = fixup_pointer(&level3_kernel_pgt, physaddr);
228	pud[510] += load_delta;
229	pud[511] += load_delta;
230
231	pmd = fixup_pointer(level2_fixmap_pgt, physaddr);
232	for (i = FIXMAP_PMD_TOP; i > FIXMAP_PMD_TOP - FIXMAP_PMD_NUM; i--)
233		pmd[i] += load_delta;
234
235	/*
236	 * Set up the identity mapping for the switchover.  These
237	 * entries should *NOT* have the global bit set!  This also
238	 * creates a bunch of nonsense entries but that is fine --
239	 * it avoids problems around wraparound.
240	 */
241
242	next_pgt_ptr = fixup_pointer(&next_early_pgt, physaddr);
243	pud = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr);
244	pmd = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr);
245
246	pgtable_flags = _KERNPG_TABLE_NOENC + sme_get_me_mask();
247
248	if (la57) {
249		p4d = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++],
250				    physaddr);
251
252		i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
253		pgd[i + 0] = (pgdval_t)p4d + pgtable_flags;
254		pgd[i + 1] = (pgdval_t)p4d + pgtable_flags;
255
256		i = physaddr >> P4D_SHIFT;
257		p4d[(i + 0) % PTRS_PER_P4D] = (pgdval_t)pud + pgtable_flags;
258		p4d[(i + 1) % PTRS_PER_P4D] = (pgdval_t)pud + pgtable_flags;
259	} else {
260		i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
261		pgd[i + 0] = (pgdval_t)pud + pgtable_flags;
262		pgd[i + 1] = (pgdval_t)pud + pgtable_flags;
263	}
264
265	i = physaddr >> PUD_SHIFT;
266	pud[(i + 0) % PTRS_PER_PUD] = (pudval_t)pmd + pgtable_flags;
267	pud[(i + 1) % PTRS_PER_PUD] = (pudval_t)pmd + pgtable_flags;
268
269	pmd_entry = __PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL;
270	/* Filter out unsupported __PAGE_KERNEL_* bits: */
271	mask_ptr = fixup_pointer(&__supported_pte_mask, physaddr);
272	pmd_entry &= *mask_ptr;
273	pmd_entry += sme_get_me_mask();
274	pmd_entry +=  physaddr;
275
276	for (i = 0; i < DIV_ROUND_UP(_end - _text, PMD_SIZE); i++) {
277		int idx = i + (physaddr >> PMD_SHIFT);
278
279		pmd[idx % PTRS_PER_PMD] = pmd_entry + i * PMD_SIZE;
280	}
281
282	/*
283	 * Fixup the kernel text+data virtual addresses. Note that
284	 * we might write invalid pmds, when the kernel is relocated
285	 * cleanup_highmap() fixes this up along with the mappings
286	 * beyond _end.
287	 *
288	 * Only the region occupied by the kernel image has so far
289	 * been checked against the table of usable memory regions
290	 * provided by the firmware, so invalidate pages outside that
291	 * region. A page table entry that maps to a reserved area of
292	 * memory would allow processor speculation into that area,
293	 * and on some hardware (particularly the UV platform) even
294	 * speculative access to some reserved areas is caught as an
295	 * error, causing the BIOS to halt the system.
296	 */
297
298	pmd = fixup_pointer(level2_kernel_pgt, physaddr);
299
300	/* invalidate pages before the kernel image */
301	for (i = 0; i < pmd_index((unsigned long)_text); i++)
302		pmd[i] &= ~_PAGE_PRESENT;
303
304	/* fixup pages that are part of the kernel image */
305	for (; i <= pmd_index((unsigned long)_end); i++)
306		if (pmd[i] & _PAGE_PRESENT)
307			pmd[i] += load_delta;
308
309	/* invalidate pages after the kernel image */
310	for (; i < PTRS_PER_PMD; i++)
311		pmd[i] &= ~_PAGE_PRESENT;
312
313	/*
314	 * Fixup phys_base - remove the memory encryption mask to obtain
315	 * the true physical address.
316	 */
317	*fixup_long(&phys_base, physaddr) += load_delta - sme_get_me_mask();
318
319	return sme_postprocess_startup(bp, pmd);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
320}
321
322/* Wipe all early page tables except for the kernel symbol map */
323static void __init reset_early_page_tables(void)
324{
325	memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
326	next_early_pgt = 0;
327	write_cr3(__sme_pa_nodebug(early_top_pgt));
328}
329
330/* Create a new PMD entry */
331bool __init __early_make_pgtable(unsigned long address, pmdval_t pmd)
332{
333	unsigned long physaddr = address - __PAGE_OFFSET;
334	pgdval_t pgd, *pgd_p;
335	p4dval_t p4d, *p4d_p;
336	pudval_t pud, *pud_p;
337	pmdval_t *pmd_p;
338
339	/* Invalid address or early pgt is done ?  */
340	if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt))
341		return false;
342
343again:
344	pgd_p = &early_top_pgt[pgd_index(address)].pgd;
345	pgd = *pgd_p;
346
347	/*
348	 * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
349	 * critical -- __PAGE_OFFSET would point us back into the dynamic
350	 * range and we might end up looping forever...
351	 */
352	if (!pgtable_l5_enabled())
353		p4d_p = pgd_p;
354	else if (pgd)
355		p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
356	else {
357		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
358			reset_early_page_tables();
359			goto again;
360		}
361
362		p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++];
363		memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D);
364		*pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
365	}
366	p4d_p += p4d_index(address);
367	p4d = *p4d_p;
368
369	if (p4d)
370		pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
371	else {
372		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
373			reset_early_page_tables();
374			goto again;
375		}
376
377		pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
378		memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
379		*p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
380	}
381	pud_p += pud_index(address);
382	pud = *pud_p;
383
384	if (pud)
385		pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
386	else {
387		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
388			reset_early_page_tables();
389			goto again;
390		}
391
392		pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
393		memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
394		*pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
395	}
396	pmd_p[pmd_index(address)] = pmd;
397
398	return true;
399}
400
401static bool __init early_make_pgtable(unsigned long address)
402{
403	unsigned long physaddr = address - __PAGE_OFFSET;
404	pmdval_t pmd;
405
406	pmd = (physaddr & PMD_MASK) + early_pmd_flags;
407
408	return __early_make_pgtable(address, pmd);
409}
410
411void __init do_early_exception(struct pt_regs *regs, int trapnr)
412{
413	if (trapnr == X86_TRAP_PF &&
414	    early_make_pgtable(native_read_cr2()))
415		return;
416
417	if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT) &&
418	    trapnr == X86_TRAP_VC && handle_vc_boot_ghcb(regs))
419		return;
420
421	if (trapnr == X86_TRAP_VE && tdx_early_handle_ve(regs))
422		return;
423
424	early_fixup_exception(regs, trapnr);
425}
426
427/* Don't add a printk in there. printk relies on the PDA which is not initialized 
428   yet. */
429void __init clear_bss(void)
430{
431	memset(__bss_start, 0,
432	       (unsigned long) __bss_stop - (unsigned long) __bss_start);
433	memset(__brk_base, 0,
434	       (unsigned long) __brk_limit - (unsigned long) __brk_base);
435}
436
437static unsigned long get_cmd_line_ptr(void)
438{
439	unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
440
441	cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
442
443	return cmd_line_ptr;
444}
445
446static void __init copy_bootdata(char *real_mode_data)
447{
448	char * command_line;
449	unsigned long cmd_line_ptr;
450
451	/*
452	 * If SME is active, this will create decrypted mappings of the
453	 * boot data in advance of the copy operations.
454	 */
455	sme_map_bootdata(real_mode_data);
456
457	memcpy(&boot_params, real_mode_data, sizeof(boot_params));
458	sanitize_boot_params(&boot_params);
459	cmd_line_ptr = get_cmd_line_ptr();
460	if (cmd_line_ptr) {
461		command_line = __va(cmd_line_ptr);
462		memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
463	}
464
465	/*
466	 * The old boot data is no longer needed and won't be reserved,
467	 * freeing up that memory for use by the system. If SME is active,
468	 * we need to remove the mappings that were created so that the
469	 * memory doesn't remain mapped as decrypted.
470	 */
471	sme_unmap_bootdata(real_mode_data);
472}
473
474asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
475{
476	/*
477	 * Build-time sanity checks on the kernel image and module
478	 * area mappings. (these are purely build-time and produce no code)
479	 */
480	BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
481	BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
482	BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
483	BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
484	BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
485	BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
486	MAYBE_BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
487				(__START_KERNEL & PGDIR_MASK)));
488	BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
489
490	cr4_init_shadow();
491
492	/* Kill off the identity-map trampoline */
493	reset_early_page_tables();
494
495	clear_bss();
496
497	/*
498	 * This needs to happen *before* kasan_early_init() because latter maps stuff
499	 * into that page.
500	 */
501	clear_page(init_top_pgt);
502
503	/*
504	 * SME support may update early_pmd_flags to include the memory
505	 * encryption mask, so it needs to be called before anything
506	 * that may generate a page fault.
507	 */
508	sme_early_init();
509
510	kasan_early_init();
511
512	/*
513	 * Flush global TLB entries which could be left over from the trampoline page
514	 * table.
515	 *
516	 * This needs to happen *after* kasan_early_init() as KASAN-enabled .configs
517	 * instrument native_write_cr4() so KASAN must be initialized for that
518	 * instrumentation to work.
519	 */
520	__native_tlb_flush_global(this_cpu_read(cpu_tlbstate.cr4));
521
522	idt_setup_early_handler();
523
524	/* Needed before cc_platform_has() can be used for TDX */
525	tdx_early_init();
526
527	copy_bootdata(__va(real_mode_data));
528
529	/*
530	 * Load microcode early on BSP.
531	 */
532	load_ucode_bsp();
533
534	/* set init_top_pgt kernel high mapping*/
535	init_top_pgt[511] = early_top_pgt[511];
536
537	x86_64_start_reservations(real_mode_data);
538}
539
540void __init x86_64_start_reservations(char *real_mode_data)
541{
542	/* version is always not zero if it is copied */
543	if (!boot_params.hdr.version)
544		copy_bootdata(__va(real_mode_data));
545
546	x86_early_init_platform_quirks();
547
548	switch (boot_params.hdr.hardware_subarch) {
549	case X86_SUBARCH_INTEL_MID:
550		x86_intel_mid_early_setup();
551		break;
552	default:
553		break;
554	}
555
556	start_kernel();
557}
558
559/*
560 * Data structures and code used for IDT setup in head_64.S. The bringup-IDT is
561 * used until the idt_table takes over. On the boot CPU this happens in
562 * x86_64_start_kernel(), on secondary CPUs in start_secondary(). In both cases
563 * this happens in the functions called from head_64.S.
564 *
565 * The idt_table can't be used that early because all the code modifying it is
566 * in idt.c and can be instrumented by tracing or KASAN, which both don't work
567 * during early CPU bringup. Also the idt_table has the runtime vectors
568 * configured which require certain CPU state to be setup already (like TSS),
569 * which also hasn't happened yet in early CPU bringup.
570 */
571static gate_desc bringup_idt_table[NUM_EXCEPTION_VECTORS] __page_aligned_data;
572
573static struct desc_ptr bringup_idt_descr = {
574	.size		= (NUM_EXCEPTION_VECTORS * sizeof(gate_desc)) - 1,
575	.address	= 0, /* Set at runtime */
576};
577
578static void set_bringup_idt_handler(gate_desc *idt, int n, void *handler)
579{
580#ifdef CONFIG_AMD_MEM_ENCRYPT
581	struct idt_data data;
582	gate_desc desc;
583
584	init_idt_data(&data, n, handler);
585	idt_init_desc(&desc, &data);
586	native_write_idt_entry(idt, n, &desc);
587#endif
588}
589
590/* This runs while still in the direct mapping */
591static void startup_64_load_idt(unsigned long physbase)
592{
593	struct desc_ptr *desc = fixup_pointer(&bringup_idt_descr, physbase);
594	gate_desc *idt = fixup_pointer(bringup_idt_table, physbase);
595
596
597	if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) {
598		void *handler;
599
600		/* VMM Communication Exception */
601		handler = fixup_pointer(vc_no_ghcb, physbase);
602		set_bringup_idt_handler(idt, X86_TRAP_VC, handler);
603	}
604
605	desc->address = (unsigned long)idt;
606	native_load_idt(desc);
607}
608
609/* This is used when running on kernel addresses */
610void early_setup_idt(void)
611{
612	/* VMM Communication Exception */
613	if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) {
614		setup_ghcb();
615		set_bringup_idt_handler(bringup_idt_table, X86_TRAP_VC, vc_boot_ghcb);
616	}
617
618	bringup_idt_descr.address = (unsigned long)bringup_idt_table;
619	native_load_idt(&bringup_idt_descr);
620}
621
622/*
623 * Setup boot CPU state needed before kernel switches to virtual addresses.
624 */
625void __head startup_64_setup_env(unsigned long physbase)
626{
627	/* Load GDT */
628	startup_gdt_descr.address = (unsigned long)fixup_pointer(startup_gdt, physbase);
629	native_load_gdt(&startup_gdt_descr);
630
631	/* New GDT is live - reload data segment registers */
632	asm volatile("movl %%eax, %%ds\n"
633		     "movl %%eax, %%ss\n"
634		     "movl %%eax, %%es\n" : : "a"(__KERNEL_DS) : "memory");
635
636	startup_64_load_idt(physbase);
637}