2 * X86-64 specific CPU setup.
3 * Copyright (C) 1995 Linus Torvalds
4 * Copyright 2001, 2002, 2003 SuSE Labs / Andi Kleen.
5 * See setup.c for older changelog.
7 #include <linux/init.h>
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
10 #include <linux/string.h>
11 #include <linux/bootmem.h>
12 #include <linux/bitops.h>
13 #include <linux/module.h>
15 #include <asm/pgtable.h>
16 #include <asm/processor.h>
18 #include <asm/atomic.h>
19 #include <asm/mmu_context.h>
22 #include <asm/percpu.h>
23 #include <asm/proto.h>
24 #include <asm/sections.h>
25 #include <asm/setup.h>
27 #ifndef CONFIG_DEBUG_BOOT_PARAMS
28 struct boot_params __initdata boot_params;
30 struct boot_params boot_params;
33 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
35 struct x8664_pda *_cpu_pda[NR_CPUS] __read_mostly;
36 EXPORT_SYMBOL(_cpu_pda);
37 struct x8664_pda boot_cpu_pda[NR_CPUS] __cacheline_aligned;
39 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
41 char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
43 unsigned long __supported_pte_mask __read_mostly = ~0UL;
44 EXPORT_SYMBOL_GPL(__supported_pte_mask);
46 static int do_not_nx __cpuinitdata = 0;
49 Control non executable mappings for 64bit processes.
54 static int __init nonx_setup(char *str)
58 if (!strncmp(str, "on", 2)) {
59 __supported_pte_mask |= _PAGE_NX;
61 } else if (!strncmp(str, "off", 3)) {
63 __supported_pte_mask &= ~_PAGE_NX;
67 early_param("noexec", nonx_setup);
69 int force_personality32 = 0;
72 Control non executable heap for 32bit processes.
73 To control the stack too use noexec=off
75 on PROT_READ does not imply PROT_EXEC for 32bit processes
76 off PROT_READ implies PROT_EXEC (default)
78 static int __init nonx32_setup(char *str)
80 if (!strcmp(str, "on"))
81 force_personality32 &= ~READ_IMPLIES_EXEC;
82 else if (!strcmp(str, "off"))
83 force_personality32 |= READ_IMPLIES_EXEC;
86 __setup("noexec32=", nonx32_setup);
88 void pda_init(int cpu)
90 struct x8664_pda *pda = cpu_pda(cpu);
92 /* Setup up data that may be needed in __get_free_pages early */
93 asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
94 /* Memory clobbers used to order PDA accessed */
96 wrmsrl(MSR_GS_BASE, pda);
102 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
103 pda->active_mm = &init_mm;
107 /* others are initialized in smpboot.c */
108 pda->pcurrent = &init_task;
109 pda->irqstackptr = boot_cpu_stack;
111 pda->irqstackptr = (char *)
112 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
113 if (!pda->irqstackptr)
114 panic("cannot allocate irqstack for cpu %d", cpu);
118 pda->irqstackptr += IRQSTACKSIZE-64;
121 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
122 __attribute__((section(".bss.page_aligned")));
124 extern asmlinkage void ignore_sysret(void);
126 /* May not be marked __init: used by software suspend */
127 void syscall_init(void)
130 * LSTAR and STAR live in a bit strange symbiosis.
131 * They both write to the same internal register. STAR allows to set CS/DS
132 * but only a 32bit target. LSTAR sets the 64bit rip.
134 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
135 wrmsrl(MSR_LSTAR, system_call);
136 wrmsrl(MSR_CSTAR, ignore_sysret);
138 #ifdef CONFIG_IA32_EMULATION
139 syscall32_cpu_init ();
142 /* Flags to clear on syscall */
143 wrmsrl(MSR_SYSCALL_MASK,
144 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
147 void __cpuinit check_efer(void)
151 rdmsrl(MSR_EFER, efer);
152 if (!(efer & EFER_NX) || do_not_nx) {
153 __supported_pte_mask &= ~_PAGE_NX;
157 unsigned long kernel_eflags;
160 * Copies of the original ist values from the tss are only accessed during
161 * debugging, no special alignment required.
163 DEFINE_PER_CPU(struct orig_ist, orig_ist);
166 * cpu_init() initializes state that is per-CPU. Some data is already
167 * initialized (naturally) in the bootstrap process, such as the GDT
168 * and IDT. We reload them nevertheless, this function acts as a
169 * 'CPU state barrier', nothing should get across.
170 * A lot of state is already set up in PDA init.
172 void __cpuinit cpu_init (void)
174 int cpu = stack_smp_processor_id();
175 struct tss_struct *t = &per_cpu(init_tss, cpu);
176 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
178 char *estacks = NULL;
179 struct task_struct *me;
182 /* CPU 0 is initialised in head64.c */
186 estacks = boot_exception_stacks;
190 if (cpu_test_and_set(cpu, cpu_initialized))
191 panic("CPU#%d already initialized!\n", cpu);
193 printk("Initializing CPU#%d\n", cpu);
195 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
198 * Initialize the per-CPU GDT with the boot GDT,
199 * and set up the GDT descriptor:
202 memcpy(get_cpu_gdt_table(cpu), cpu_gdt_table, GDT_SIZE);
204 cpu_gdt_descr[cpu].size = GDT_SIZE;
205 load_gdt((const struct desc_ptr *)&cpu_gdt_descr[cpu]);
206 load_idt((const struct desc_ptr *)&idt_descr);
208 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
211 wrmsrl(MSR_FS_BASE, 0);
212 wrmsrl(MSR_KERNEL_GS_BASE, 0);
218 * set up and load the per-CPU TSS
220 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
221 static const unsigned int order[N_EXCEPTION_STACKS] = {
222 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
223 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
226 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
228 panic("Cannot allocate exception stack %ld %d\n",
231 estacks += PAGE_SIZE << order[v];
232 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
235 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
237 * <= is required because the CPU will access up to
238 * 8 bits beyond the end of the IO permission bitmap.
240 for (i = 0; i <= IO_BITMAP_LONGS; i++)
241 t->io_bitmap[i] = ~0UL;
243 atomic_inc(&init_mm.mm_count);
244 me->active_mm = &init_mm;
247 enter_lazy_tlb(&init_mm, me);
249 set_tss_desc(cpu, t);
251 load_LDT(&init_mm.context);
254 * Clear all 6 debug registers:
257 set_debugreg(0UL, 0);
258 set_debugreg(0UL, 1);
259 set_debugreg(0UL, 2);
260 set_debugreg(0UL, 3);
261 set_debugreg(0UL, 6);
262 set_debugreg(0UL, 7);
266 raw_local_save_flags(kernel_eflags);