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.
6 * $Id: setup64.c,v 1.12 2002/03/21 10:09:17 ak Exp $
8 #include <linux/config.h>
9 #include <linux/init.h>
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <linux/string.h>
13 #include <linux/bootmem.h>
14 #include <linux/bitops.h>
15 #include <linux/module.h>
16 #include <asm/bootsetup.h>
18 #include <asm/pgtable.h>
19 #include <asm/processor.h>
21 #include <asm/atomic.h>
22 #include <asm/mmu_context.h>
25 #include <asm/percpu.h>
26 #include <asm/proto.h>
27 #include <asm/sections.h>
29 char x86_boot_params[BOOT_PARAM_SIZE] __initdata = {0,};
31 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
33 struct x8664_pda cpu_pda[NR_CPUS] __cacheline_aligned;
35 struct desc_ptr idt_descr = { 256 * 16, (unsigned long) idt_table };
37 char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
39 unsigned long __supported_pte_mask __read_mostly = ~0UL;
40 static int do_not_nx __initdata = 0;
43 Control non executable mappings for 64bit processes.
48 int __init nonx_setup(char *str)
50 if (!strncmp(str, "on", 2)) {
51 __supported_pte_mask |= _PAGE_NX;
53 } else if (!strncmp(str, "off", 3)) {
55 __supported_pte_mask &= ~_PAGE_NX;
59 __setup("noexec=", nonx_setup); /* parsed early actually */
61 int force_personality32 = READ_IMPLIES_EXEC;
64 Control non executable heap for 32bit processes.
65 To control the stack too use noexec=off
67 on PROT_READ does not imply PROT_EXEC for 32bit processes
68 off PROT_READ implies PROT_EXEC (default)
70 static int __init nonx32_setup(char *str)
72 if (!strcmp(str, "on"))
73 force_personality32 &= ~READ_IMPLIES_EXEC;
74 else if (!strcmp(str, "off"))
75 force_personality32 |= READ_IMPLIES_EXEC;
78 __setup("noexec32=", nonx32_setup);
82 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
83 * Always point %gs to its beginning
85 void __init setup_per_cpu_areas(void)
90 #ifdef CONFIG_HOTPLUG_CPU
91 prefill_possible_map();
94 /* Copy section for each CPU (we discard the original) */
95 size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
97 if (size < PERCPU_ENOUGH_ROOM)
98 size = PERCPU_ENOUGH_ROOM;
101 for_each_cpu_mask (i, cpu_possible_map) {
104 if (!NODE_DATA(cpu_to_node(i))) {
105 printk("cpu with no node %d, num_online_nodes %d\n",
106 i, num_online_nodes());
107 ptr = alloc_bootmem(size);
109 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
112 panic("Cannot allocate cpu data for CPU %d\n", i);
113 cpu_pda[i].data_offset = ptr - __per_cpu_start;
114 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
118 void pda_init(int cpu)
120 struct x8664_pda *pda = &cpu_pda[cpu];
122 /* Setup up data that may be needed in __get_free_pages early */
123 asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
124 wrmsrl(MSR_GS_BASE, cpu_pda + cpu);
126 pda->cpunumber = cpu;
129 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
130 pda->active_mm = &init_mm;
134 /* others are initialized in smpboot.c */
135 pda->pcurrent = &init_task;
136 pda->irqstackptr = boot_cpu_stack;
138 pda->irqstackptr = (char *)
139 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
140 if (!pda->irqstackptr)
141 panic("cannot allocate irqstack for cpu %d", cpu);
145 pda->irqstackptr += IRQSTACKSIZE-64;
148 char boot_exception_stacks[(N_EXCEPTION_STACKS - 2) * EXCEPTION_STKSZ + DEBUG_STKSZ]
149 __attribute__((section(".bss.page_aligned")));
151 /* May not be marked __init: used by software suspend */
152 void syscall_init(void)
155 * LSTAR and STAR live in a bit strange symbiosis.
156 * They both write to the same internal register. STAR allows to set CS/DS
157 * but only a 32bit target. LSTAR sets the 64bit rip.
159 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
160 wrmsrl(MSR_LSTAR, system_call);
162 #ifdef CONFIG_IA32_EMULATION
163 syscall32_cpu_init ();
166 /* Flags to clear on syscall */
167 wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
170 void __cpuinit check_efer(void)
174 rdmsrl(MSR_EFER, efer);
175 if (!(efer & EFER_NX) || do_not_nx) {
176 __supported_pte_mask &= ~_PAGE_NX;
181 * cpu_init() initializes state that is per-CPU. Some data is already
182 * initialized (naturally) in the bootstrap process, such as the GDT
183 * and IDT. We reload them nevertheless, this function acts as a
184 * 'CPU state barrier', nothing should get across.
185 * A lot of state is already set up in PDA init.
187 void __cpuinit cpu_init (void)
189 int cpu = stack_smp_processor_id();
190 struct tss_struct *t = &per_cpu(init_tss, cpu);
192 char *estacks = NULL;
193 struct task_struct *me;
196 /* CPU 0 is initialised in head64.c */
199 zap_low_mappings(cpu);
201 estacks = boot_exception_stacks;
205 if (cpu_test_and_set(cpu, cpu_initialized))
206 panic("CPU#%d already initialized!\n", cpu);
208 printk("Initializing CPU#%d\n", cpu);
210 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
213 * Initialize the per-CPU GDT with the boot GDT,
214 * and set up the GDT descriptor:
217 memcpy(cpu_gdt_table[cpu], cpu_gdt_table[0], GDT_SIZE);
220 cpu_gdt_descr[cpu].size = GDT_SIZE;
221 cpu_gdt_descr[cpu].address = (unsigned long)cpu_gdt_table[cpu];
222 asm volatile("lgdt %0" :: "m" (cpu_gdt_descr[cpu]));
223 asm volatile("lidt %0" :: "m" (idt_descr));
225 memcpy(me->thread.tls_array, cpu_gdt_table[cpu], GDT_ENTRY_TLS_ENTRIES * 8);
228 wrmsrl(MSR_FS_BASE, 0);
229 wrmsrl(MSR_KERNEL_GS_BASE, 0);
235 * set up and load the per-CPU TSS
237 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
239 static const unsigned int order[N_EXCEPTION_STACKS] = {
240 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
241 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
244 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
246 panic("Cannot allocate exception stack %ld %d\n",
250 #if DEBUG_STKSZ > EXCEPTION_STKSZ
252 cpu_pda[cpu].debugstack = (unsigned long)estacks;
253 estacks += DEBUG_STKSZ;
257 estacks += EXCEPTION_STKSZ;
260 t->ist[v] = (unsigned long)estacks;
263 t->io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
265 * <= is required because the CPU will access up to
266 * 8 bits beyond the end of the IO permission bitmap.
268 for (i = 0; i <= IO_BITMAP_LONGS; i++)
269 t->io_bitmap[i] = ~0UL;
271 atomic_inc(&init_mm.mm_count);
272 me->active_mm = &init_mm;
275 enter_lazy_tlb(&init_mm, me);
277 set_tss_desc(cpu, t);
279 load_LDT(&init_mm.context);
282 * Clear all 6 debug registers: