1 /* Paravirtualization interfaces
2 Copyright (C) 2006 Rusty Russell IBM Corporation
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
21 #include <linux/errno.h>
22 #include <linux/module.h>
23 #include <linux/efi.h>
24 #include <linux/bcd.h>
25 #include <linux/highmem.h>
28 #include <asm/paravirt.h>
30 #include <asm/setup.h>
31 #include <asm/pgtable.h>
33 #include <asm/pgalloc.h>
35 #include <asm/delay.h>
36 #include <asm/fixmap.h>
38 #include <asm/tlbflush.h>
39 #include <asm/timer.h>
42 void _paravirt_nop(void)
46 /* identity function, which can be inlined */
47 u32 _paravirt_ident_32(u32 x)
52 u64 _paravirt_ident_64(u64 x)
57 static void __init default_banner(void)
59 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
63 char *memory_setup(void)
65 return pv_init_ops.memory_setup();
68 /* Simple instruction patching code. */
69 #define DEF_NATIVE(ops, name, code) \
70 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
71 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
73 /* Undefined instruction for dealing with missing ops pointers. */
74 static const unsigned char ud2a[] = { 0x0f, 0x0b };
76 unsigned paravirt_patch_nop(void)
81 unsigned paravirt_patch_ignore(unsigned len)
89 } __attribute__((packed));
91 unsigned paravirt_patch_call(void *insnbuf,
92 const void *target, u16 tgt_clobbers,
93 unsigned long addr, u16 site_clobbers,
96 struct branch *b = insnbuf;
97 unsigned long delta = (unsigned long)target - (addr+5);
99 if (tgt_clobbers & ~site_clobbers)
100 return len; /* target would clobber too much for this site */
102 return len; /* call too long for patch site */
104 b->opcode = 0xe8; /* call */
106 BUILD_BUG_ON(sizeof(*b) != 5);
111 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
112 unsigned long addr, unsigned len)
114 struct branch *b = insnbuf;
115 unsigned long delta = (unsigned long)target - (addr+5);
118 return len; /* call too long for patch site */
120 b->opcode = 0xe9; /* jmp */
126 /* Neat trick to map patch type back to the call within the
127 * corresponding structure. */
128 static void *get_call_destination(u8 type)
130 struct paravirt_patch_template tmpl = {
131 .pv_init_ops = pv_init_ops,
132 .pv_time_ops = pv_time_ops,
133 .pv_cpu_ops = pv_cpu_ops,
134 .pv_irq_ops = pv_irq_ops,
135 .pv_apic_ops = pv_apic_ops,
136 .pv_mmu_ops = pv_mmu_ops,
137 .pv_lock_ops = pv_lock_ops,
139 return *((void **)&tmpl + type);
142 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
143 unsigned long addr, unsigned len)
145 void *opfunc = get_call_destination(type);
149 /* If there's no function, patch it with a ud2a (BUG) */
150 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
151 else if (opfunc == _paravirt_nop)
152 /* If the operation is a nop, then nop the callsite */
153 ret = paravirt_patch_nop();
155 /* identity functions just return their single argument */
156 else if (opfunc == _paravirt_ident_32)
157 ret = paravirt_patch_ident_32(insnbuf, len);
158 else if (opfunc == _paravirt_ident_64)
159 ret = paravirt_patch_ident_64(insnbuf, len);
161 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
162 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
163 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
164 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
165 /* If operation requires a jmp, then jmp */
166 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
168 /* Otherwise call the function; assume target could
169 clobber any caller-save reg */
170 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
171 addr, clobbers, len);
176 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
177 const char *start, const char *end)
179 unsigned insn_len = end - start;
181 if (insn_len > len || start == NULL)
184 memcpy(insnbuf, start, insn_len);
191 pv_irq_ops.init_IRQ();
194 static void native_flush_tlb(void)
196 __native_flush_tlb();
200 * Global pages have to be flushed a bit differently. Not a real
201 * performance problem because this does not happen often.
203 static void native_flush_tlb_global(void)
205 __native_flush_tlb_global();
208 static void native_flush_tlb_single(unsigned long addr)
210 __native_flush_tlb_single(addr);
213 /* These are in entry.S */
214 extern void native_iret(void);
215 extern void native_irq_enable_sysexit(void);
216 extern void native_usergs_sysret32(void);
217 extern void native_usergs_sysret64(void);
219 static int __init print_banner(void)
221 pv_init_ops.banner();
224 core_initcall(print_banner);
226 static struct resource reserve_ioports = {
228 .end = IO_SPACE_LIMIT,
229 .name = "paravirt-ioport",
230 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
234 * Reserve the whole legacy IO space to prevent any legacy drivers
235 * from wasting time probing for their hardware. This is a fairly
236 * brute-force approach to disabling all non-virtual drivers.
238 * Note that this must be called very early to have any effect.
240 int paravirt_disable_iospace(void)
242 return request_resource(&ioport_resource, &reserve_ioports);
245 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
247 static inline void enter_lazy(enum paravirt_lazy_mode mode)
249 BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
250 BUG_ON(preemptible());
252 __get_cpu_var(paravirt_lazy_mode) = mode;
255 static void leave_lazy(enum paravirt_lazy_mode mode)
257 BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode);
258 BUG_ON(preemptible());
260 __get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
263 void paravirt_enter_lazy_mmu(void)
265 enter_lazy(PARAVIRT_LAZY_MMU);
268 void paravirt_leave_lazy_mmu(void)
270 leave_lazy(PARAVIRT_LAZY_MMU);
273 void paravirt_start_context_switch(struct task_struct *prev)
275 if (percpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
276 arch_leave_lazy_mmu_mode();
277 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
279 enter_lazy(PARAVIRT_LAZY_CPU);
282 void paravirt_end_context_switch(struct task_struct *next)
284 leave_lazy(PARAVIRT_LAZY_CPU);
286 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
287 arch_enter_lazy_mmu_mode();
290 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
293 return PARAVIRT_LAZY_NONE;
295 return __get_cpu_var(paravirt_lazy_mode);
298 void arch_flush_lazy_mmu_mode(void)
302 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
303 WARN_ON(preempt_count() == 1);
304 arch_leave_lazy_mmu_mode();
305 arch_enter_lazy_mmu_mode();
311 struct pv_info pv_info = {
312 .name = "bare hardware",
313 .paravirt_enabled = 0,
315 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
318 struct pv_init_ops pv_init_ops = {
319 .patch = native_patch,
320 .banner = default_banner,
321 .arch_setup = paravirt_nop,
322 .memory_setup = machine_specific_memory_setup,
325 struct pv_time_ops pv_time_ops = {
326 .time_init = hpet_time_init,
327 .get_wallclock = native_get_wallclock,
328 .set_wallclock = native_set_wallclock,
329 .sched_clock = native_sched_clock,
330 .get_tsc_khz = native_calibrate_tsc,
333 struct pv_irq_ops pv_irq_ops = {
334 .init_IRQ = native_init_IRQ,
335 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
336 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
337 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
338 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
339 .safe_halt = native_safe_halt,
342 .adjust_exception_frame = paravirt_nop,
346 struct pv_cpu_ops pv_cpu_ops = {
347 .cpuid = native_cpuid,
348 .get_debugreg = native_get_debugreg,
349 .set_debugreg = native_set_debugreg,
351 .read_cr0 = native_read_cr0,
352 .write_cr0 = native_write_cr0,
353 .read_cr4 = native_read_cr4,
354 .read_cr4_safe = native_read_cr4_safe,
355 .write_cr4 = native_write_cr4,
357 .read_cr8 = native_read_cr8,
358 .write_cr8 = native_write_cr8,
360 .wbinvd = native_wbinvd,
361 .read_msr = native_read_msr_safe,
362 .read_msr_amd = native_read_msr_amd_safe,
363 .write_msr = native_write_msr_safe,
364 .read_tsc = native_read_tsc,
365 .read_pmc = native_read_pmc,
366 .read_tscp = native_read_tscp,
367 .load_tr_desc = native_load_tr_desc,
368 .set_ldt = native_set_ldt,
369 .load_gdt = native_load_gdt,
370 .load_idt = native_load_idt,
371 .store_gdt = native_store_gdt,
372 .store_idt = native_store_idt,
373 .store_tr = native_store_tr,
374 .load_tls = native_load_tls,
376 .load_gs_index = native_load_gs_index,
378 .write_ldt_entry = native_write_ldt_entry,
379 .write_gdt_entry = native_write_gdt_entry,
380 .write_idt_entry = native_write_idt_entry,
382 .alloc_ldt = paravirt_nop,
383 .free_ldt = paravirt_nop,
385 .load_sp0 = native_load_sp0,
387 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
388 .irq_enable_sysexit = native_irq_enable_sysexit,
391 #ifdef CONFIG_IA32_EMULATION
392 .usergs_sysret32 = native_usergs_sysret32,
394 .usergs_sysret64 = native_usergs_sysret64,
397 .swapgs = native_swapgs,
399 .set_iopl_mask = native_set_iopl_mask,
400 .io_delay = native_io_delay,
402 .start_context_switch = paravirt_nop,
403 .end_context_switch = paravirt_nop,
406 struct pv_apic_ops pv_apic_ops = {
407 #ifdef CONFIG_X86_LOCAL_APIC
408 .setup_boot_clock = setup_boot_APIC_clock,
409 .setup_secondary_clock = setup_secondary_APIC_clock,
410 .startup_ipi_hook = paravirt_nop,
414 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
415 /* 32-bit pagetable entries */
416 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
418 /* 64-bit pagetable entries */
419 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
422 struct pv_mmu_ops pv_mmu_ops = {
423 #ifndef CONFIG_X86_64
424 .pagetable_setup_start = native_pagetable_setup_start,
425 .pagetable_setup_done = native_pagetable_setup_done,
427 .pagetable_setup_start = paravirt_nop,
428 .pagetable_setup_done = paravirt_nop,
431 .read_cr2 = native_read_cr2,
432 .write_cr2 = native_write_cr2,
433 .read_cr3 = native_read_cr3,
434 .write_cr3 = native_write_cr3,
436 .flush_tlb_user = native_flush_tlb,
437 .flush_tlb_kernel = native_flush_tlb_global,
438 .flush_tlb_single = native_flush_tlb_single,
439 .flush_tlb_others = native_flush_tlb_others,
441 .pgd_alloc = __paravirt_pgd_alloc,
442 .pgd_free = paravirt_nop,
444 .alloc_pte = paravirt_nop,
445 .alloc_pmd = paravirt_nop,
446 .alloc_pmd_clone = paravirt_nop,
447 .alloc_pud = paravirt_nop,
448 .release_pte = paravirt_nop,
449 .release_pmd = paravirt_nop,
450 .release_pud = paravirt_nop,
452 .set_pte = native_set_pte,
453 .set_pte_at = native_set_pte_at,
454 .set_pmd = native_set_pmd,
455 .pte_update = paravirt_nop,
456 .pte_update_defer = paravirt_nop,
458 .ptep_modify_prot_start = __ptep_modify_prot_start,
459 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
461 #ifdef CONFIG_HIGHPTE
462 .kmap_atomic_pte = kmap_atomic,
465 #if PAGETABLE_LEVELS >= 3
466 #ifdef CONFIG_X86_PAE
467 .set_pte_atomic = native_set_pte_atomic,
468 .set_pte_present = native_set_pte_present,
469 .pte_clear = native_pte_clear,
470 .pmd_clear = native_pmd_clear,
472 .set_pud = native_set_pud,
474 .pmd_val = PTE_IDENT,
475 .make_pmd = PTE_IDENT,
477 #if PAGETABLE_LEVELS == 4
478 .pud_val = PTE_IDENT,
479 .make_pud = PTE_IDENT,
481 .set_pgd = native_set_pgd,
483 #endif /* PAGETABLE_LEVELS >= 3 */
485 .pte_val = PTE_IDENT,
486 .pgd_val = PTE_IDENT,
488 .make_pte = PTE_IDENT,
489 .make_pgd = PTE_IDENT,
491 .dup_mmap = paravirt_nop,
492 .exit_mmap = paravirt_nop,
493 .activate_mm = paravirt_nop,
496 .enter = paravirt_nop,
497 .leave = paravirt_nop,
500 .set_fixmap = native_set_fixmap,
503 EXPORT_SYMBOL_GPL(pv_time_ops);
504 EXPORT_SYMBOL (pv_cpu_ops);
505 EXPORT_SYMBOL (pv_mmu_ops);
506 EXPORT_SYMBOL_GPL(pv_apic_ops);
507 EXPORT_SYMBOL_GPL(pv_info);
508 EXPORT_SYMBOL (pv_irq_ops);