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>
26 #include <linux/kprobes.h>
29 #include <asm/paravirt.h>
30 #include <asm/debugreg.h>
32 #include <asm/setup.h>
33 #include <asm/pgtable.h>
35 #include <asm/pgalloc.h>
37 #include <asm/delay.h>
38 #include <asm/fixmap.h>
40 #include <asm/tlbflush.h>
41 #include <asm/timer.h>
42 #include <asm/special_insns.h>
45 * nop stub, which must not clobber anything *including the stack* to
46 * avoid confusing the entry prologues.
48 extern void _paravirt_nop(void);
49 asm (".pushsection .entry.text, \"ax\"\n"
50 ".global _paravirt_nop\n"
53 ".size _paravirt_nop, . - _paravirt_nop\n\t"
54 ".type _paravirt_nop, @function\n\t"
57 /* identity function, which can be inlined */
58 u32 _paravirt_ident_32(u32 x)
63 u64 _paravirt_ident_64(u64 x)
68 void __init default_banner(void)
70 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
74 /* Undefined instruction for dealing with missing ops pointers. */
75 static const unsigned char ud2a[] = { 0x0f, 0x0b };
77 unsigned paravirt_patch_nop(void)
82 unsigned paravirt_patch_ignore(unsigned len)
90 } __attribute__((packed));
92 unsigned paravirt_patch_call(void *insnbuf,
93 const void *target, u16 tgt_clobbers,
94 unsigned long addr, u16 site_clobbers,
97 struct branch *b = insnbuf;
98 unsigned long delta = (unsigned long)target - (addr+5);
100 if (tgt_clobbers & ~site_clobbers)
101 return len; /* target would clobber too much for this site */
103 return len; /* call too long for patch site */
105 b->opcode = 0xe8; /* call */
107 BUILD_BUG_ON(sizeof(*b) != 5);
112 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
113 unsigned long addr, unsigned len)
115 struct branch *b = insnbuf;
116 unsigned long delta = (unsigned long)target - (addr+5);
119 return len; /* call too long for patch site */
121 b->opcode = 0xe9; /* jmp */
127 /* Neat trick to map patch type back to the call within the
128 * corresponding structure. */
129 static void *get_call_destination(u8 type)
131 struct paravirt_patch_template tmpl = {
132 .pv_init_ops = pv_init_ops,
133 .pv_time_ops = pv_time_ops,
134 .pv_cpu_ops = pv_cpu_ops,
135 .pv_irq_ops = pv_irq_ops,
136 .pv_apic_ops = pv_apic_ops,
137 .pv_mmu_ops = pv_mmu_ops,
138 #ifdef CONFIG_PARAVIRT_SPINLOCKS
139 .pv_lock_ops = pv_lock_ops,
142 return *((void **)&tmpl + type);
145 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
146 unsigned long addr, unsigned len)
148 void *opfunc = get_call_destination(type);
152 /* If there's no function, patch it with a ud2a (BUG) */
153 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
154 else if (opfunc == _paravirt_nop)
155 /* If the operation is a nop, then nop the callsite */
156 ret = paravirt_patch_nop();
158 /* identity functions just return their single argument */
159 else if (opfunc == _paravirt_ident_32)
160 ret = paravirt_patch_ident_32(insnbuf, len);
161 else if (opfunc == _paravirt_ident_64)
162 ret = paravirt_patch_ident_64(insnbuf, len);
164 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
166 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
168 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
169 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
170 /* If operation requires a jmp, then jmp */
171 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
173 /* Otherwise call the function; assume target could
174 clobber any caller-save reg */
175 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
176 addr, clobbers, len);
181 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
182 const char *start, const char *end)
184 unsigned insn_len = end - start;
186 if (insn_len > len || start == NULL)
189 memcpy(insnbuf, start, insn_len);
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 struct static_key paravirt_steal_enabled;
214 struct static_key paravirt_steal_rq_enabled;
216 static u64 native_steal_clock(int cpu)
221 /* These are in entry.S */
222 extern void native_iret(void);
223 extern void native_irq_enable_sysexit(void);
224 extern void native_usergs_sysret32(void);
225 extern void native_usergs_sysret64(void);
227 static struct resource reserve_ioports = {
229 .end = IO_SPACE_LIMIT,
230 .name = "paravirt-ioport",
231 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
235 * Reserve the whole legacy IO space to prevent any legacy drivers
236 * from wasting time probing for their hardware. This is a fairly
237 * brute-force approach to disabling all non-virtual drivers.
239 * Note that this must be called very early to have any effect.
241 int paravirt_disable_iospace(void)
243 return request_resource(&ioport_resource, &reserve_ioports);
246 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
248 static inline void enter_lazy(enum paravirt_lazy_mode mode)
250 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
252 this_cpu_write(paravirt_lazy_mode, mode);
255 static void leave_lazy(enum paravirt_lazy_mode mode)
257 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
259 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
262 void paravirt_enter_lazy_mmu(void)
264 enter_lazy(PARAVIRT_LAZY_MMU);
267 void paravirt_leave_lazy_mmu(void)
269 leave_lazy(PARAVIRT_LAZY_MMU);
272 void paravirt_flush_lazy_mmu(void)
276 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
277 arch_leave_lazy_mmu_mode();
278 arch_enter_lazy_mmu_mode();
284 void paravirt_start_context_switch(struct task_struct *prev)
286 BUG_ON(preemptible());
288 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
289 arch_leave_lazy_mmu_mode();
290 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
292 enter_lazy(PARAVIRT_LAZY_CPU);
295 void paravirt_end_context_switch(struct task_struct *next)
297 BUG_ON(preemptible());
299 leave_lazy(PARAVIRT_LAZY_CPU);
301 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
302 arch_enter_lazy_mmu_mode();
305 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
308 return PARAVIRT_LAZY_NONE;
310 return this_cpu_read(paravirt_lazy_mode);
313 struct pv_info pv_info = {
314 .name = "bare hardware",
315 .paravirt_enabled = 0,
317 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
320 .extra_user_64bit_cs = __USER_CS,
324 struct pv_init_ops pv_init_ops = {
325 .patch = native_patch,
328 struct pv_time_ops pv_time_ops = {
329 .sched_clock = native_sched_clock,
330 .steal_clock = native_steal_clock,
333 __visible struct pv_irq_ops pv_irq_ops = {
334 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
335 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
336 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
337 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
338 .safe_halt = native_safe_halt,
341 .adjust_exception_frame = paravirt_nop,
345 __visible struct pv_cpu_ops pv_cpu_ops = {
346 .cpuid = native_cpuid,
347 .get_debugreg = native_get_debugreg,
348 .set_debugreg = native_set_debugreg,
350 .read_cr0 = native_read_cr0,
351 .write_cr0 = native_write_cr0,
352 .read_cr4 = native_read_cr4,
353 .read_cr4_safe = native_read_cr4_safe,
354 .write_cr4 = native_write_cr4,
356 .read_cr8 = native_read_cr8,
357 .write_cr8 = native_write_cr8,
359 .wbinvd = native_wbinvd,
360 .read_msr = native_read_msr_safe,
361 .write_msr = native_write_msr_safe,
362 .read_pmc = native_read_pmc,
363 .load_tr_desc = native_load_tr_desc,
364 .set_ldt = native_set_ldt,
365 .load_gdt = native_load_gdt,
366 .load_idt = native_load_idt,
367 .store_idt = native_store_idt,
368 .store_tr = native_store_tr,
369 .load_tls = native_load_tls,
371 .load_gs_index = native_load_gs_index,
373 .write_ldt_entry = native_write_ldt_entry,
374 .write_gdt_entry = native_write_gdt_entry,
375 .write_idt_entry = native_write_idt_entry,
377 .alloc_ldt = paravirt_nop,
378 .free_ldt = paravirt_nop,
380 .load_sp0 = native_load_sp0,
382 #if defined(CONFIG_X86_32)
383 .irq_enable_sysexit = native_irq_enable_sysexit,
386 #ifdef CONFIG_IA32_EMULATION
387 .usergs_sysret32 = native_usergs_sysret32,
389 .usergs_sysret64 = native_usergs_sysret64,
392 .swapgs = native_swapgs,
394 .set_iopl_mask = native_set_iopl_mask,
395 .io_delay = native_io_delay,
397 .start_context_switch = paravirt_nop,
398 .end_context_switch = paravirt_nop,
401 /* At this point, native_get/set_debugreg has real function entries */
402 NOKPROBE_SYMBOL(native_get_debugreg);
403 NOKPROBE_SYMBOL(native_set_debugreg);
404 NOKPROBE_SYMBOL(native_load_idt);
406 struct pv_apic_ops pv_apic_ops = {
407 #ifdef CONFIG_X86_LOCAL_APIC
408 .startup_ipi_hook = paravirt_nop,
412 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
413 /* 32-bit pagetable entries */
414 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
416 /* 64-bit pagetable entries */
417 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
420 struct pv_mmu_ops pv_mmu_ops = {
422 .read_cr2 = native_read_cr2,
423 .write_cr2 = native_write_cr2,
424 .read_cr3 = native_read_cr3,
425 .write_cr3 = native_write_cr3,
427 .flush_tlb_user = native_flush_tlb,
428 .flush_tlb_kernel = native_flush_tlb_global,
429 .flush_tlb_single = native_flush_tlb_single,
430 .flush_tlb_others = native_flush_tlb_others,
432 .pgd_alloc = __paravirt_pgd_alloc,
433 .pgd_free = paravirt_nop,
435 .alloc_pte = paravirt_nop,
436 .alloc_pmd = paravirt_nop,
437 .alloc_pud = paravirt_nop,
438 .release_pte = paravirt_nop,
439 .release_pmd = paravirt_nop,
440 .release_pud = paravirt_nop,
442 .set_pte = native_set_pte,
443 .set_pte_at = native_set_pte_at,
444 .set_pmd = native_set_pmd,
445 .set_pmd_at = native_set_pmd_at,
446 .pte_update = paravirt_nop,
447 .pte_update_defer = paravirt_nop,
448 .pmd_update = paravirt_nop,
449 .pmd_update_defer = paravirt_nop,
451 .ptep_modify_prot_start = __ptep_modify_prot_start,
452 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
454 #if CONFIG_PGTABLE_LEVELS >= 3
455 #ifdef CONFIG_X86_PAE
456 .set_pte_atomic = native_set_pte_atomic,
457 .pte_clear = native_pte_clear,
458 .pmd_clear = native_pmd_clear,
460 .set_pud = native_set_pud,
462 .pmd_val = PTE_IDENT,
463 .make_pmd = PTE_IDENT,
465 #if CONFIG_PGTABLE_LEVELS == 4
466 .pud_val = PTE_IDENT,
467 .make_pud = PTE_IDENT,
469 .set_pgd = native_set_pgd,
471 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
473 .pte_val = PTE_IDENT,
474 .pgd_val = PTE_IDENT,
476 .make_pte = PTE_IDENT,
477 .make_pgd = PTE_IDENT,
479 .dup_mmap = paravirt_nop,
480 .exit_mmap = paravirt_nop,
481 .activate_mm = paravirt_nop,
484 .enter = paravirt_nop,
485 .leave = paravirt_nop,
486 .flush = paravirt_nop,
489 .set_fixmap = native_set_fixmap,
492 EXPORT_SYMBOL_GPL(pv_time_ops);
493 EXPORT_SYMBOL (pv_cpu_ops);
494 EXPORT_SYMBOL (pv_mmu_ops);
495 EXPORT_SYMBOL_GPL(pv_apic_ops);
496 EXPORT_SYMBOL_GPL(pv_info);
497 EXPORT_SYMBOL (pv_irq_ops);