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);
89 * Copy data used in early init routines from the initial arrays to the
90 * per cpu data areas. These arrays then become expendable and the
91 * *_early_ptr's are zeroed indicating that the static arrays are gone.
93 static void __init setup_per_cpu_maps(void)
97 for_each_possible_cpu(cpu) {
99 if (per_cpu_offset(cpu)) {
101 per_cpu(x86_cpu_to_apicid, cpu) =
102 x86_cpu_to_apicid_init[cpu];
103 per_cpu(x86_bios_cpu_apicid, cpu) =
104 x86_bios_cpu_apicid_init[cpu];
106 per_cpu(x86_cpu_to_node_map, cpu) =
107 x86_cpu_to_node_map_init[cpu];
112 printk(KERN_NOTICE "per_cpu_offset zero for cpu %d\n",
117 /* indicate the early static arrays will soon be gone */
118 x86_cpu_to_apicid_early_ptr = NULL;
119 x86_bios_cpu_apicid_early_ptr = NULL;
121 x86_cpu_to_node_map_early_ptr = NULL;
127 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
128 * Always point %gs to its beginning
130 void __init setup_per_cpu_areas(void)
135 #ifdef CONFIG_HOTPLUG_CPU
136 prefill_possible_map();
139 /* Copy section for each CPU (we discard the original) */
140 size = PERCPU_ENOUGH_ROOM;
142 printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n", size);
143 for_each_cpu_mask (i, cpu_possible_map) {
146 if (!NODE_DATA(early_cpu_to_node(i))) {
147 printk("cpu with no node %d, num_online_nodes %d\n",
148 i, num_online_nodes());
149 ptr = alloc_bootmem_pages(size);
151 ptr = alloc_bootmem_pages_node(NODE_DATA(early_cpu_to_node(i)), size);
154 panic("Cannot allocate cpu data for CPU %d\n", i);
155 cpu_pda(i)->data_offset = ptr - __per_cpu_start;
156 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
159 /* setup percpu data maps early */
160 setup_per_cpu_maps();
163 void pda_init(int cpu)
165 struct x8664_pda *pda = cpu_pda(cpu);
167 /* Setup up data that may be needed in __get_free_pages early */
168 asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
169 /* Memory clobbers used to order PDA accessed */
171 wrmsrl(MSR_GS_BASE, pda);
174 pda->cpunumber = cpu;
177 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
178 pda->active_mm = &init_mm;
182 /* others are initialized in smpboot.c */
183 pda->pcurrent = &init_task;
184 pda->irqstackptr = boot_cpu_stack;
186 pda->irqstackptr = (char *)
187 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
188 if (!pda->irqstackptr)
189 panic("cannot allocate irqstack for cpu %d", cpu);
193 pda->irqstackptr += IRQSTACKSIZE-64;
196 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
197 __attribute__((section(".bss.page_aligned")));
199 extern asmlinkage void ignore_sysret(void);
201 /* May not be marked __init: used by software suspend */
202 void syscall_init(void)
205 * LSTAR and STAR live in a bit strange symbiosis.
206 * They both write to the same internal register. STAR allows to set CS/DS
207 * but only a 32bit target. LSTAR sets the 64bit rip.
209 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
210 wrmsrl(MSR_LSTAR, system_call);
211 wrmsrl(MSR_CSTAR, ignore_sysret);
213 #ifdef CONFIG_IA32_EMULATION
214 syscall32_cpu_init ();
217 /* Flags to clear on syscall */
218 wrmsrl(MSR_SYSCALL_MASK,
219 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
222 void __cpuinit check_efer(void)
226 rdmsrl(MSR_EFER, efer);
227 if (!(efer & EFER_NX) || do_not_nx) {
228 __supported_pte_mask &= ~_PAGE_NX;
232 unsigned long kernel_eflags;
235 * Copies of the original ist values from the tss are only accessed during
236 * debugging, no special alignment required.
238 DEFINE_PER_CPU(struct orig_ist, orig_ist);
241 * cpu_init() initializes state that is per-CPU. Some data is already
242 * initialized (naturally) in the bootstrap process, such as the GDT
243 * and IDT. We reload them nevertheless, this function acts as a
244 * 'CPU state barrier', nothing should get across.
245 * A lot of state is already set up in PDA init.
247 void __cpuinit cpu_init (void)
249 int cpu = stack_smp_processor_id();
250 struct tss_struct *t = &per_cpu(init_tss, cpu);
251 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
253 char *estacks = NULL;
254 struct task_struct *me;
257 /* CPU 0 is initialised in head64.c */
261 estacks = boot_exception_stacks;
265 if (cpu_test_and_set(cpu, cpu_initialized))
266 panic("CPU#%d already initialized!\n", cpu);
268 printk("Initializing CPU#%d\n", cpu);
270 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
273 * Initialize the per-CPU GDT with the boot GDT,
274 * and set up the GDT descriptor:
277 memcpy(get_cpu_gdt_table(cpu), cpu_gdt_table, GDT_SIZE);
279 cpu_gdt_descr[cpu].size = GDT_SIZE;
280 load_gdt((const struct desc_ptr *)&cpu_gdt_descr[cpu]);
281 load_idt((const struct desc_ptr *)&idt_descr);
283 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
286 wrmsrl(MSR_FS_BASE, 0);
287 wrmsrl(MSR_KERNEL_GS_BASE, 0);
293 * set up and load the per-CPU TSS
295 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
296 static const unsigned int order[N_EXCEPTION_STACKS] = {
297 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
298 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
301 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
303 panic("Cannot allocate exception stack %ld %d\n",
306 estacks += PAGE_SIZE << order[v];
307 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
310 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
312 * <= is required because the CPU will access up to
313 * 8 bits beyond the end of the IO permission bitmap.
315 for (i = 0; i <= IO_BITMAP_LONGS; i++)
316 t->io_bitmap[i] = ~0UL;
318 atomic_inc(&init_mm.mm_count);
319 me->active_mm = &init_mm;
322 enter_lazy_tlb(&init_mm, me);
324 set_tss_desc(cpu, t);
326 load_LDT(&init_mm.context);
329 * Clear all 6 debug registers:
332 set_debugreg(0UL, 0);
333 set_debugreg(0UL, 1);
334 set_debugreg(0UL, 2);
335 set_debugreg(0UL, 3);
336 set_debugreg(0UL, 6);
337 set_debugreg(0UL, 7);
341 raw_local_save_flags(kernel_eflags);