1 #include <linux/init.h>
2 #include <linux/kernel.h>
3 #include <linux/sched.h>
4 #include <linux/string.h>
5 #include <linux/bootmem.h>
6 #include <linux/bitops.h>
7 #include <linux/module.h>
8 #include <linux/kgdb.h>
9 #include <linux/topology.h>
10 #include <linux/delay.h>
11 #include <linux/smp.h>
12 #include <linux/percpu.h>
16 #include <asm/linkage.h>
17 #include <asm/mmu_context.h>
23 #ifdef CONFIG_X86_LOCAL_APIC
24 #include <asm/mpspec.h>
26 #include <mach_apic.h>
29 #include <asm/pgtable.h>
30 #include <asm/processor.h>
32 #include <asm/atomic.h>
33 #include <asm/proto.h>
34 #include <asm/sections.h>
35 #include <asm/setup.h>
36 #include <asm/genapic.h>
40 /* We need valid kernel segments for data and code in long mode too
41 * IRET will check the segment types kkeil 2000/10/28
42 * Also sysret mandates a special GDT layout
44 /* The TLS descriptors are currently at a different place compared to i386.
45 Hopefully nobody expects them at a fixed place (Wine?) */
46 DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
47 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
48 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
49 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
50 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
51 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
52 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
54 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
56 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
58 /* Current gdt points %fs at the "master" per-cpu area: after this,
59 * it's on the real one. */
60 void switch_to_new_gdt(void)
62 struct desc_ptr gdt_descr;
64 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
65 gdt_descr.size = GDT_SIZE - 1;
69 struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
71 static void __cpuinit default_init(struct cpuinfo_x86 *c)
76 static struct cpu_dev __cpuinitdata default_cpu = {
77 .c_init = default_init,
78 .c_vendor = "Unknown",
80 static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
82 int __cpuinit get_model_name(struct cpuinfo_x86 *c)
86 if (c->extended_cpuid_level < 0x80000004)
89 v = (unsigned int *) c->x86_model_id;
90 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
91 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
92 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
93 c->x86_model_id[48] = 0;
98 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
100 unsigned int n, dummy, ebx, ecx, edx;
102 n = c->extended_cpuid_level;
104 if (n >= 0x80000005) {
105 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
106 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
107 "D cache %dK (%d bytes/line)\n",
108 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
109 c->x86_cache_size = (ecx>>24) + (edx>>24);
110 /* On K8 L1 TLB is inclusive, so don't count it */
114 if (n >= 0x80000006) {
115 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
116 ecx = cpuid_ecx(0x80000006);
117 c->x86_cache_size = ecx >> 16;
118 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
120 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
121 c->x86_cache_size, ecx & 0xFF);
125 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
128 u32 eax, ebx, ecx, edx;
129 int index_msb, core_bits;
131 cpuid(1, &eax, &ebx, &ecx, &edx);
134 if (!cpu_has(c, X86_FEATURE_HT))
136 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
139 smp_num_siblings = (ebx & 0xff0000) >> 16;
141 if (smp_num_siblings == 1) {
142 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
143 } else if (smp_num_siblings > 1) {
145 if (smp_num_siblings > NR_CPUS) {
146 printk(KERN_WARNING "CPU: Unsupported number of "
147 "siblings %d", smp_num_siblings);
148 smp_num_siblings = 1;
152 index_msb = get_count_order(smp_num_siblings);
153 c->phys_proc_id = phys_pkg_id(index_msb);
155 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
157 index_msb = get_count_order(smp_num_siblings);
159 core_bits = get_count_order(c->x86_max_cores);
161 c->cpu_core_id = phys_pkg_id(index_msb) &
162 ((1 << core_bits) - 1);
165 if ((c->x86_max_cores * smp_num_siblings) > 1) {
166 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
168 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
175 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
177 char *v = c->x86_vendor_id;
181 for (i = 0; i < X86_VENDOR_NUM; i++) {
183 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
184 (cpu_devs[i]->c_ident[1] &&
185 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
187 this_cpu = cpu_devs[i];
194 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
195 printk(KERN_ERR "CPU: Your system may be unstable.\n");
197 c->x86_vendor = X86_VENDOR_UNKNOWN;
200 static void __init early_cpu_support_print(void)
203 struct cpu_dev *cpu_devx;
205 printk("KERNEL supported cpus:\n");
206 for (i = 0; i < X86_VENDOR_NUM; i++) {
207 cpu_devx = cpu_devs[i];
210 for (j = 0; j < 2; j++) {
211 if (!cpu_devx->c_ident[j])
213 printk(" %s %s\n", cpu_devx->c_vendor,
214 cpu_devx->c_ident[j]);
220 * The NOPL instruction is supposed to exist on all CPUs with
221 * family >= 6, unfortunately, that's not true in practice because
222 * of early VIA chips and (more importantly) broken virtualizers that
223 * are not easy to detect. Hence, probe for it based on first
226 * Note: no 64-bit chip is known to lack these, but put the code here
227 * for consistency with 32 bits, and to make it utterly trivial to
228 * diagnose the problem should it ever surface.
230 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
232 const u32 nopl_signature = 0x888c53b1; /* Random number */
233 u32 has_nopl = nopl_signature;
235 clear_cpu_cap(c, X86_FEATURE_NOPL);
238 "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
240 " .section .fixup,\"ax\"\n"
247 if (has_nopl == nopl_signature)
248 set_cpu_cap(c, X86_FEATURE_NOPL);
252 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
254 void __init early_cpu_init(void)
256 struct cpu_vendor_dev *cvdev;
258 for (cvdev = __x86cpuvendor_start ;
259 cvdev < __x86cpuvendor_end ;
261 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
262 early_cpu_support_print();
263 early_identify_cpu(&boot_cpu_data);
266 /* Do some early cpuid on the boot CPU to get some parameter that are
267 needed before check_bugs. Everything advanced is in identify_cpu
269 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
273 c->loops_per_jiffy = loops_per_jiffy;
274 c->x86_cache_size = -1;
275 c->x86_vendor = X86_VENDOR_UNKNOWN;
276 c->x86_model = c->x86_mask = 0; /* So far unknown... */
277 c->x86_vendor_id[0] = '\0'; /* Unset */
278 c->x86_model_id[0] = '\0'; /* Unset */
279 c->x86_clflush_size = 64;
280 c->x86_cache_alignment = c->x86_clflush_size;
281 c->x86_max_cores = 1;
282 c->x86_coreid_bits = 0;
283 c->extended_cpuid_level = 0;
284 memset(&c->x86_capability, 0, sizeof c->x86_capability);
286 /* Get vendor name */
287 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
288 (unsigned int *)&c->x86_vendor_id[0],
289 (unsigned int *)&c->x86_vendor_id[8],
290 (unsigned int *)&c->x86_vendor_id[4]);
294 /* Initialize the standard set of capabilities */
295 /* Note that the vendor-specific code below might override */
297 /* Intel-defined flags: level 0x00000001 */
298 if (c->cpuid_level >= 0x00000001) {
300 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
301 &c->x86_capability[0]);
302 c->x86 = (tfms >> 8) & 0xf;
303 c->x86_model = (tfms >> 4) & 0xf;
304 c->x86_mask = tfms & 0xf;
306 c->x86 += (tfms >> 20) & 0xff;
308 c->x86_model += ((tfms >> 16) & 0xF) << 4;
309 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
310 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
312 /* Have CPUID level 0 only - unheard of */
316 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
318 c->phys_proc_id = c->initial_apicid;
320 /* AMD-defined flags: level 0x80000001 */
321 xlvl = cpuid_eax(0x80000000);
322 c->extended_cpuid_level = xlvl;
323 if ((xlvl & 0xffff0000) == 0x80000000) {
324 if (xlvl >= 0x80000001) {
325 c->x86_capability[1] = cpuid_edx(0x80000001);
326 c->x86_capability[6] = cpuid_ecx(0x80000001);
328 if (xlvl >= 0x80000004)
329 get_model_name(c); /* Default name */
332 /* Transmeta-defined flags: level 0x80860001 */
333 xlvl = cpuid_eax(0x80860000);
334 if ((xlvl & 0xffff0000) == 0x80860000) {
335 /* Don't set x86_cpuid_level here for now to not confuse. */
336 if (xlvl >= 0x80860001)
337 c->x86_capability[2] = cpuid_edx(0x80860001);
340 if (c->extended_cpuid_level >= 0x80000007)
341 c->x86_power = cpuid_edx(0x80000007);
343 if (c->extended_cpuid_level >= 0x80000008) {
344 u32 eax = cpuid_eax(0x80000008);
346 c->x86_virt_bits = (eax >> 8) & 0xff;
347 c->x86_phys_bits = eax & 0xff;
352 if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
353 cpu_devs[c->x86_vendor]->c_early_init)
354 cpu_devs[c->x86_vendor]->c_early_init(c);
356 validate_pat_support(c);
360 * This does the hard work of actually picking apart the CPU stuff...
362 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
366 early_identify_cpu(c);
368 init_scattered_cpuid_features(c);
370 c->apicid = phys_pkg_id(0);
373 * Vendor-specific initialization. In this section we
374 * canonicalize the feature flags, meaning if there are
375 * features a certain CPU supports which CPUID doesn't
376 * tell us, CPUID claiming incorrect flags, or other bugs,
377 * we handle them here.
379 * At the end of this section, c->x86_capability better
380 * indicate the features this CPU genuinely supports!
382 if (this_cpu->c_init)
388 * On SMP, boot_cpu_data holds the common feature set between
389 * all CPUs; so make sure that we indicate which features are
390 * common between the CPUs. The first time this routine gets
391 * executed, c == &boot_cpu_data.
393 if (c != &boot_cpu_data) {
394 /* AND the already accumulated flags with these */
395 for (i = 0; i < NCAPINTS; i++)
396 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
399 /* Clear all flags overriden by options */
400 for (i = 0; i < NCAPINTS; i++)
401 c->x86_capability[i] &= ~cleared_cpu_caps[i];
403 #ifdef CONFIG_X86_MCE
406 select_idle_routine(c);
409 numa_add_cpu(smp_processor_id());
414 void __cpuinit identify_boot_cpu(void)
416 identify_cpu(&boot_cpu_data);
419 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
421 BUG_ON(c == &boot_cpu_data);
426 static __init int setup_noclflush(char *arg)
428 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
431 __setup("noclflush", setup_noclflush);
433 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
435 if (c->x86_model_id[0])
436 printk(KERN_CONT "%s", c->x86_model_id);
438 if (c->x86_mask || c->cpuid_level >= 0)
439 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
441 printk(KERN_CONT "\n");
444 static __init int setup_disablecpuid(char *arg)
447 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
448 setup_clear_cpu_cap(bit);
453 __setup("clearcpuid=", setup_disablecpuid);
455 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
457 struct x8664_pda **_cpu_pda __read_mostly;
458 EXPORT_SYMBOL(_cpu_pda);
460 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
462 char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
464 unsigned long __supported_pte_mask __read_mostly = ~0UL;
465 EXPORT_SYMBOL_GPL(__supported_pte_mask);
467 static int do_not_nx __cpuinitdata;
470 Control non executable mappings for 64bit processes.
475 static int __init nonx_setup(char *str)
479 if (!strncmp(str, "on", 2)) {
480 __supported_pte_mask |= _PAGE_NX;
482 } else if (!strncmp(str, "off", 3)) {
484 __supported_pte_mask &= ~_PAGE_NX;
488 early_param("noexec", nonx_setup);
490 int force_personality32;
493 Control non executable heap for 32bit processes.
494 To control the stack too use noexec=off
496 on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
497 off PROT_READ implies PROT_EXEC
499 static int __init nonx32_setup(char *str)
501 if (!strcmp(str, "on"))
502 force_personality32 &= ~READ_IMPLIES_EXEC;
503 else if (!strcmp(str, "off"))
504 force_personality32 |= READ_IMPLIES_EXEC;
507 __setup("noexec32=", nonx32_setup);
509 void pda_init(int cpu)
511 struct x8664_pda *pda = cpu_pda(cpu);
513 /* Setup up data that may be needed in __get_free_pages early */
516 /* Memory clobbers used to order PDA accessed */
518 wrmsrl(MSR_GS_BASE, pda);
521 pda->cpunumber = cpu;
523 pda->kernelstack = (unsigned long)stack_thread_info() -
524 PDA_STACKOFFSET + THREAD_SIZE;
525 pda->active_mm = &init_mm;
529 /* others are initialized in smpboot.c */
530 pda->pcurrent = &init_task;
531 pda->irqstackptr = boot_cpu_stack;
532 pda->irqstackptr += IRQSTACKSIZE - 64;
534 if (!pda->irqstackptr) {
535 pda->irqstackptr = (char *)
536 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
537 if (!pda->irqstackptr)
538 panic("cannot allocate irqstack for cpu %d",
540 pda->irqstackptr += IRQSTACKSIZE - 64;
543 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
544 pda->nodenumber = cpu_to_node(cpu);
548 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
549 DEBUG_STKSZ] __page_aligned_bss;
551 extern asmlinkage void ignore_sysret(void);
553 /* May not be marked __init: used by software suspend */
554 void syscall_init(void)
557 * LSTAR and STAR live in a bit strange symbiosis.
558 * They both write to the same internal register. STAR allows to
559 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
561 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
562 wrmsrl(MSR_LSTAR, system_call);
563 wrmsrl(MSR_CSTAR, ignore_sysret);
565 #ifdef CONFIG_IA32_EMULATION
566 syscall32_cpu_init();
569 /* Flags to clear on syscall */
570 wrmsrl(MSR_SYSCALL_MASK,
571 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
574 void __cpuinit check_efer(void)
578 rdmsrl(MSR_EFER, efer);
579 if (!(efer & EFER_NX) || do_not_nx)
580 __supported_pte_mask &= ~_PAGE_NX;
583 unsigned long kernel_eflags;
586 * Copies of the original ist values from the tss are only accessed during
587 * debugging, no special alignment required.
589 DEFINE_PER_CPU(struct orig_ist, orig_ist);
592 * cpu_init() initializes state that is per-CPU. Some data is already
593 * initialized (naturally) in the bootstrap process, such as the GDT
594 * and IDT. We reload them nevertheless, this function acts as a
595 * 'CPU state barrier', nothing should get across.
596 * A lot of state is already set up in PDA init.
598 void __cpuinit cpu_init(void)
600 int cpu = stack_smp_processor_id();
601 struct tss_struct *t = &per_cpu(init_tss, cpu);
602 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
604 char *estacks = NULL;
605 struct task_struct *me;
608 /* CPU 0 is initialised in head64.c */
612 estacks = boot_exception_stacks;
616 if (cpu_test_and_set(cpu, cpu_initialized))
617 panic("CPU#%d already initialized!\n", cpu);
619 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
621 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
624 * Initialize the per-CPU GDT with the boot GDT,
625 * and set up the GDT descriptor:
629 load_idt((const struct desc_ptr *)&idt_descr);
631 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
634 wrmsrl(MSR_FS_BASE, 0);
635 wrmsrl(MSR_KERNEL_GS_BASE, 0);
641 * set up and load the per-CPU TSS
643 if (!orig_ist->ist[0]) {
644 static const unsigned int order[N_EXCEPTION_STACKS] = {
645 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
646 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
648 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
650 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
652 panic("Cannot allocate exception "
653 "stack %ld %d\n", v, cpu);
655 estacks += PAGE_SIZE << order[v];
656 orig_ist->ist[v] = t->x86_tss.ist[v] =
657 (unsigned long)estacks;
661 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
663 * <= is required because the CPU will access up to
664 * 8 bits beyond the end of the IO permission bitmap.
666 for (i = 0; i <= IO_BITMAP_LONGS; i++)
667 t->io_bitmap[i] = ~0UL;
669 atomic_inc(&init_mm.mm_count);
670 me->active_mm = &init_mm;
673 enter_lazy_tlb(&init_mm, me);
675 load_sp0(t, ¤t->thread);
676 set_tss_desc(cpu, t);
678 load_LDT(&init_mm.context);
682 * If the kgdb is connected no debug regs should be altered. This
683 * is only applicable when KGDB and a KGDB I/O module are built
684 * into the kernel and you are using early debugging with
685 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
687 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
688 arch_kgdb_ops.correct_hw_break();
692 * Clear all 6 debug registers:
695 set_debugreg(0UL, 0);
696 set_debugreg(0UL, 1);
697 set_debugreg(0UL, 2);
698 set_debugreg(0UL, 3);
699 set_debugreg(0UL, 6);
700 set_debugreg(0UL, 7);
702 /* If the kgdb is connected no debug regs should be altered. */
708 raw_local_save_flags(kernel_eflags);