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 /* Simple instruction patching code. */
64 #define DEF_NATIVE(ops, name, code) \
65 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
66 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
68 /* Undefined instruction for dealing with missing ops pointers. */
69 static const unsigned char ud2a[] = { 0x0f, 0x0b };
71 unsigned paravirt_patch_nop(void)
76 unsigned paravirt_patch_ignore(unsigned len)
84 } __attribute__((packed));
86 unsigned paravirt_patch_call(void *insnbuf,
87 const void *target, u16 tgt_clobbers,
88 unsigned long addr, u16 site_clobbers,
91 struct branch *b = insnbuf;
92 unsigned long delta = (unsigned long)target - (addr+5);
94 if (tgt_clobbers & ~site_clobbers)
95 return len; /* target would clobber too much for this site */
97 return len; /* call too long for patch site */
99 b->opcode = 0xe8; /* call */
101 BUILD_BUG_ON(sizeof(*b) != 5);
106 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
107 unsigned long addr, unsigned len)
109 struct branch *b = insnbuf;
110 unsigned long delta = (unsigned long)target - (addr+5);
113 return len; /* call too long for patch site */
115 b->opcode = 0xe9; /* jmp */
121 /* Neat trick to map patch type back to the call within the
122 * corresponding structure. */
123 static void *get_call_destination(u8 type)
125 struct paravirt_patch_template tmpl = {
126 .pv_init_ops = pv_init_ops,
127 .pv_time_ops = pv_time_ops,
128 .pv_cpu_ops = pv_cpu_ops,
129 .pv_irq_ops = pv_irq_ops,
130 .pv_apic_ops = pv_apic_ops,
131 .pv_mmu_ops = pv_mmu_ops,
132 #ifdef CONFIG_PARAVIRT_SPINLOCKS
133 .pv_lock_ops = pv_lock_ops,
136 return *((void **)&tmpl + type);
139 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
140 unsigned long addr, unsigned len)
142 void *opfunc = get_call_destination(type);
146 /* If there's no function, patch it with a ud2a (BUG) */
147 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
148 else if (opfunc == _paravirt_nop)
149 /* If the operation is a nop, then nop the callsite */
150 ret = paravirt_patch_nop();
152 /* identity functions just return their single argument */
153 else if (opfunc == _paravirt_ident_32)
154 ret = paravirt_patch_ident_32(insnbuf, len);
155 else if (opfunc == _paravirt_ident_64)
156 ret = paravirt_patch_ident_64(insnbuf, len);
158 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
159 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
160 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
161 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
162 /* If operation requires a jmp, then jmp */
163 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
165 /* Otherwise call the function; assume target could
166 clobber any caller-save reg */
167 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
168 addr, clobbers, len);
173 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
174 const char *start, const char *end)
176 unsigned insn_len = end - start;
178 if (insn_len > len || start == NULL)
181 memcpy(insnbuf, start, insn_len);
188 pv_irq_ops.init_IRQ();
191 static void native_flush_tlb(void)
193 __native_flush_tlb();
197 * Global pages have to be flushed a bit differently. Not a real
198 * performance problem because this does not happen often.
200 static void native_flush_tlb_global(void)
202 __native_flush_tlb_global();
205 static void native_flush_tlb_single(unsigned long addr)
207 __native_flush_tlb_single(addr);
210 /* These are in entry.S */
211 extern void native_iret(void);
212 extern void native_irq_enable_sysexit(void);
213 extern void native_usergs_sysret32(void);
214 extern void native_usergs_sysret64(void);
216 static int __init print_banner(void)
218 pv_init_ops.banner();
221 core_initcall(print_banner);
223 static struct resource reserve_ioports = {
225 .end = IO_SPACE_LIMIT,
226 .name = "paravirt-ioport",
227 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
231 * Reserve the whole legacy IO space to prevent any legacy drivers
232 * from wasting time probing for their hardware. This is a fairly
233 * brute-force approach to disabling all non-virtual drivers.
235 * Note that this must be called very early to have any effect.
237 int paravirt_disable_iospace(void)
239 return request_resource(&ioport_resource, &reserve_ioports);
242 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
244 static inline void enter_lazy(enum paravirt_lazy_mode mode)
246 BUG_ON(percpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
248 percpu_write(paravirt_lazy_mode, mode);
251 static void leave_lazy(enum paravirt_lazy_mode mode)
253 BUG_ON(percpu_read(paravirt_lazy_mode) != mode);
255 percpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
258 void paravirt_enter_lazy_mmu(void)
260 enter_lazy(PARAVIRT_LAZY_MMU);
263 void paravirt_leave_lazy_mmu(void)
265 leave_lazy(PARAVIRT_LAZY_MMU);
268 void paravirt_start_context_switch(struct task_struct *prev)
270 BUG_ON(preemptible());
272 if (percpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
273 arch_leave_lazy_mmu_mode();
274 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
276 enter_lazy(PARAVIRT_LAZY_CPU);
279 void paravirt_end_context_switch(struct task_struct *next)
281 BUG_ON(preemptible());
283 leave_lazy(PARAVIRT_LAZY_CPU);
285 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
286 arch_enter_lazy_mmu_mode();
289 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
292 return PARAVIRT_LAZY_NONE;
294 return percpu_read(paravirt_lazy_mode);
297 void arch_flush_lazy_mmu_mode(void)
301 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
302 arch_leave_lazy_mmu_mode();
303 arch_enter_lazy_mmu_mode();
309 struct pv_info pv_info = {
310 .name = "bare hardware",
311 .paravirt_enabled = 0,
313 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
316 struct pv_init_ops pv_init_ops = {
317 .patch = native_patch,
318 .banner = default_banner,
319 .arch_setup = paravirt_nop,
322 struct pv_time_ops pv_time_ops = {
323 .time_init = hpet_time_init,
324 .get_wallclock = native_get_wallclock,
325 .set_wallclock = native_set_wallclock,
326 .sched_clock = native_sched_clock,
327 .get_tsc_khz = native_calibrate_tsc,
330 struct pv_irq_ops pv_irq_ops = {
331 .init_IRQ = native_init_IRQ,
332 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
333 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
334 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
335 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
336 .safe_halt = native_safe_halt,
339 .adjust_exception_frame = paravirt_nop,
343 struct pv_cpu_ops pv_cpu_ops = {
344 .cpuid = native_cpuid,
345 .get_debugreg = native_get_debugreg,
346 .set_debugreg = native_set_debugreg,
348 .read_cr0 = native_read_cr0,
349 .write_cr0 = native_write_cr0,
350 .read_cr4 = native_read_cr4,
351 .read_cr4_safe = native_read_cr4_safe,
352 .write_cr4 = native_write_cr4,
354 .read_cr8 = native_read_cr8,
355 .write_cr8 = native_write_cr8,
357 .wbinvd = native_wbinvd,
358 .read_msr = native_read_msr_safe,
359 .read_msr_amd = native_read_msr_amd_safe,
360 .write_msr = native_write_msr_safe,
361 .read_tsc = native_read_tsc,
362 .read_pmc = native_read_pmc,
363 .read_tscp = native_read_tscp,
364 .load_tr_desc = native_load_tr_desc,
365 .set_ldt = native_set_ldt,
366 .load_gdt = native_load_gdt,
367 .load_idt = native_load_idt,
368 .store_gdt = native_store_gdt,
369 .store_idt = native_store_idt,
370 .store_tr = native_store_tr,
371 .load_tls = native_load_tls,
373 .load_gs_index = native_load_gs_index,
375 .write_ldt_entry = native_write_ldt_entry,
376 .write_gdt_entry = native_write_gdt_entry,
377 .write_idt_entry = native_write_idt_entry,
379 .alloc_ldt = paravirt_nop,
380 .free_ldt = paravirt_nop,
382 .load_sp0 = native_load_sp0,
384 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
385 .irq_enable_sysexit = native_irq_enable_sysexit,
388 #ifdef CONFIG_IA32_EMULATION
389 .usergs_sysret32 = native_usergs_sysret32,
391 .usergs_sysret64 = native_usergs_sysret64,
394 .swapgs = native_swapgs,
396 .set_iopl_mask = native_set_iopl_mask,
397 .io_delay = native_io_delay,
399 .start_context_switch = paravirt_nop,
400 .end_context_switch = paravirt_nop,
403 struct pv_apic_ops pv_apic_ops = {
404 #ifdef CONFIG_X86_LOCAL_APIC
405 .setup_boot_clock = setup_boot_APIC_clock,
406 .setup_secondary_clock = setup_secondary_APIC_clock,
407 .startup_ipi_hook = paravirt_nop,
411 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
412 /* 32-bit pagetable entries */
413 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
415 /* 64-bit pagetable entries */
416 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
419 struct pv_mmu_ops pv_mmu_ops = {
420 #ifndef CONFIG_X86_64
421 .pagetable_setup_start = native_pagetable_setup_start,
422 .pagetable_setup_done = native_pagetable_setup_done,
424 .pagetable_setup_start = paravirt_nop,
425 .pagetable_setup_done = paravirt_nop,
428 .read_cr2 = native_read_cr2,
429 .write_cr2 = native_write_cr2,
430 .read_cr3 = native_read_cr3,
431 .write_cr3 = native_write_cr3,
433 .flush_tlb_user = native_flush_tlb,
434 .flush_tlb_kernel = native_flush_tlb_global,
435 .flush_tlb_single = native_flush_tlb_single,
436 .flush_tlb_others = native_flush_tlb_others,
438 .pgd_alloc = __paravirt_pgd_alloc,
439 .pgd_free = paravirt_nop,
441 .alloc_pte = paravirt_nop,
442 .alloc_pmd = paravirt_nop,
443 .alloc_pmd_clone = paravirt_nop,
444 .alloc_pud = paravirt_nop,
445 .release_pte = paravirt_nop,
446 .release_pmd = paravirt_nop,
447 .release_pud = paravirt_nop,
449 .set_pte = native_set_pte,
450 .set_pte_at = native_set_pte_at,
451 .set_pmd = native_set_pmd,
452 .pte_update = paravirt_nop,
453 .pte_update_defer = paravirt_nop,
455 .ptep_modify_prot_start = __ptep_modify_prot_start,
456 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
458 #ifdef CONFIG_HIGHPTE
459 .kmap_atomic_pte = kmap_atomic,
462 #if PAGETABLE_LEVELS >= 3
463 #ifdef CONFIG_X86_PAE
464 .set_pte_atomic = native_set_pte_atomic,
465 .pte_clear = native_pte_clear,
466 .pmd_clear = native_pmd_clear,
468 .set_pud = native_set_pud,
470 .pmd_val = PTE_IDENT,
471 .make_pmd = PTE_IDENT,
473 #if PAGETABLE_LEVELS == 4
474 .pud_val = PTE_IDENT,
475 .make_pud = PTE_IDENT,
477 .set_pgd = native_set_pgd,
479 #endif /* PAGETABLE_LEVELS >= 3 */
481 .pte_val = PTE_IDENT,
482 .pgd_val = PTE_IDENT,
484 .make_pte = PTE_IDENT,
485 .make_pgd = PTE_IDENT,
487 .dup_mmap = paravirt_nop,
488 .exit_mmap = paravirt_nop,
489 .activate_mm = paravirt_nop,
492 .enter = paravirt_nop,
493 .leave = paravirt_nop,
496 .set_fixmap = native_set_fixmap,
499 EXPORT_SYMBOL_GPL(pv_time_ops);
500 EXPORT_SYMBOL (pv_cpu_ops);
501 EXPORT_SYMBOL (pv_mmu_ops);
502 EXPORT_SYMBOL_GPL(pv_apic_ops);
503 EXPORT_SYMBOL_GPL(pv_info);
504 EXPORT_SYMBOL (pv_irq_ops);