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 };
80 } __attribute__((packed));
82 unsigned paravirt_patch_call(void *insnbuf,
83 const void *target, u16 tgt_clobbers,
84 unsigned long addr, u16 site_clobbers,
87 struct branch *b = insnbuf;
88 unsigned long delta = (unsigned long)target - (addr+5);
90 if (tgt_clobbers & ~site_clobbers)
91 return len; /* target would clobber too much for this site */
93 return len; /* call too long for patch site */
95 b->opcode = 0xe8; /* call */
97 BUILD_BUG_ON(sizeof(*b) != 5);
102 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
103 unsigned long addr, unsigned len)
105 struct branch *b = insnbuf;
106 unsigned long delta = (unsigned long)target - (addr+5);
109 return len; /* call too long for patch site */
111 b->opcode = 0xe9; /* jmp */
117 /* Neat trick to map patch type back to the call within the
118 * corresponding structure. */
119 static void *get_call_destination(u8 type)
121 struct paravirt_patch_template tmpl = {
122 .pv_init_ops = pv_init_ops,
123 .pv_time_ops = pv_time_ops,
124 .pv_cpu_ops = pv_cpu_ops,
125 .pv_irq_ops = pv_irq_ops,
126 .pv_mmu_ops = pv_mmu_ops,
127 #ifdef CONFIG_PARAVIRT_SPINLOCKS
128 .pv_lock_ops = pv_lock_ops,
131 return *((void **)&tmpl + type);
134 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
135 unsigned long addr, unsigned len)
137 void *opfunc = get_call_destination(type);
141 /* If there's no function, patch it with a ud2a (BUG) */
142 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
143 else if (opfunc == _paravirt_nop)
146 /* identity functions just return their single argument */
147 else if (opfunc == _paravirt_ident_32)
148 ret = paravirt_patch_ident_32(insnbuf, len);
149 else if (opfunc == _paravirt_ident_64)
150 ret = paravirt_patch_ident_64(insnbuf, len);
152 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
154 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
156 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
157 type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
158 /* If operation requires a jmp, then jmp */
159 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
161 /* Otherwise call the function; assume target could
162 clobber any caller-save reg */
163 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
164 addr, clobbers, len);
169 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
170 const char *start, const char *end)
172 unsigned insn_len = end - start;
174 if (insn_len > len || start == NULL)
177 memcpy(insnbuf, start, insn_len);
182 static void native_flush_tlb(void)
184 __native_flush_tlb();
188 * Global pages have to be flushed a bit differently. Not a real
189 * performance problem because this does not happen often.
191 static void native_flush_tlb_global(void)
193 __native_flush_tlb_global();
196 static void native_flush_tlb_single(unsigned long addr)
198 __native_flush_tlb_single(addr);
201 struct static_key paravirt_steal_enabled;
202 struct static_key paravirt_steal_rq_enabled;
204 static u64 native_steal_clock(int cpu)
209 /* These are in entry.S */
210 extern void native_iret(void);
211 extern void native_irq_enable_sysexit(void);
212 extern void native_usergs_sysret32(void);
213 extern void native_usergs_sysret64(void);
215 static struct resource reserve_ioports = {
217 .end = IO_SPACE_LIMIT,
218 .name = "paravirt-ioport",
219 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
223 * Reserve the whole legacy IO space to prevent any legacy drivers
224 * from wasting time probing for their hardware. This is a fairly
225 * brute-force approach to disabling all non-virtual drivers.
227 * Note that this must be called very early to have any effect.
229 int paravirt_disable_iospace(void)
231 return request_resource(&ioport_resource, &reserve_ioports);
234 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
236 static inline void enter_lazy(enum paravirt_lazy_mode mode)
238 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
240 this_cpu_write(paravirt_lazy_mode, mode);
243 static void leave_lazy(enum paravirt_lazy_mode mode)
245 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
247 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
250 void paravirt_enter_lazy_mmu(void)
252 enter_lazy(PARAVIRT_LAZY_MMU);
255 void paravirt_leave_lazy_mmu(void)
257 leave_lazy(PARAVIRT_LAZY_MMU);
260 void paravirt_flush_lazy_mmu(void)
264 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
265 arch_leave_lazy_mmu_mode();
266 arch_enter_lazy_mmu_mode();
272 void paravirt_start_context_switch(struct task_struct *prev)
274 BUG_ON(preemptible());
276 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
277 arch_leave_lazy_mmu_mode();
278 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
280 enter_lazy(PARAVIRT_LAZY_CPU);
283 void paravirt_end_context_switch(struct task_struct *next)
285 BUG_ON(preemptible());
287 leave_lazy(PARAVIRT_LAZY_CPU);
289 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
290 arch_enter_lazy_mmu_mode();
293 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
296 return PARAVIRT_LAZY_NONE;
298 return this_cpu_read(paravirt_lazy_mode);
301 struct pv_info pv_info = {
302 .name = "bare hardware",
303 .paravirt_enabled = 0,
305 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
308 .extra_user_64bit_cs = __USER_CS,
312 struct pv_init_ops pv_init_ops = {
313 .patch = native_patch,
316 struct pv_time_ops pv_time_ops = {
317 .sched_clock = native_sched_clock,
318 .steal_clock = native_steal_clock,
321 __visible struct pv_irq_ops pv_irq_ops = {
322 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
323 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
324 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
325 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
326 .safe_halt = native_safe_halt,
329 .adjust_exception_frame = paravirt_nop,
333 __visible struct pv_cpu_ops pv_cpu_ops = {
334 .cpuid = native_cpuid,
335 .get_debugreg = native_get_debugreg,
336 .set_debugreg = native_set_debugreg,
338 .read_cr0 = native_read_cr0,
339 .write_cr0 = native_write_cr0,
340 .read_cr4 = native_read_cr4,
341 .read_cr4_safe = native_read_cr4_safe,
342 .write_cr4 = native_write_cr4,
344 .read_cr8 = native_read_cr8,
345 .write_cr8 = native_write_cr8,
347 .wbinvd = native_wbinvd,
348 .read_msr = native_read_msr_safe,
349 .write_msr = native_write_msr_safe,
350 .read_pmc = native_read_pmc,
351 .load_tr_desc = native_load_tr_desc,
352 .set_ldt = native_set_ldt,
353 .load_gdt = native_load_gdt,
354 .load_idt = native_load_idt,
355 .store_idt = native_store_idt,
356 .store_tr = native_store_tr,
357 .load_tls = native_load_tls,
359 .load_gs_index = native_load_gs_index,
361 .write_ldt_entry = native_write_ldt_entry,
362 .write_gdt_entry = native_write_gdt_entry,
363 .write_idt_entry = native_write_idt_entry,
365 .alloc_ldt = paravirt_nop,
366 .free_ldt = paravirt_nop,
368 .load_sp0 = native_load_sp0,
370 #if defined(CONFIG_X86_32)
371 .irq_enable_sysexit = native_irq_enable_sysexit,
374 #ifdef CONFIG_IA32_EMULATION
375 .usergs_sysret32 = native_usergs_sysret32,
377 .usergs_sysret64 = native_usergs_sysret64,
380 .swapgs = native_swapgs,
382 .set_iopl_mask = native_set_iopl_mask,
383 .io_delay = native_io_delay,
385 .start_context_switch = paravirt_nop,
386 .end_context_switch = paravirt_nop,
389 /* At this point, native_get/set_debugreg has real function entries */
390 NOKPROBE_SYMBOL(native_get_debugreg);
391 NOKPROBE_SYMBOL(native_set_debugreg);
392 NOKPROBE_SYMBOL(native_load_idt);
394 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
395 /* 32-bit pagetable entries */
396 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
398 /* 64-bit pagetable entries */
399 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
402 struct pv_mmu_ops pv_mmu_ops = {
404 .read_cr2 = native_read_cr2,
405 .write_cr2 = native_write_cr2,
406 .read_cr3 = native_read_cr3,
407 .write_cr3 = native_write_cr3,
409 .flush_tlb_user = native_flush_tlb,
410 .flush_tlb_kernel = native_flush_tlb_global,
411 .flush_tlb_single = native_flush_tlb_single,
412 .flush_tlb_others = native_flush_tlb_others,
414 .pgd_alloc = __paravirt_pgd_alloc,
415 .pgd_free = paravirt_nop,
417 .alloc_pte = paravirt_nop,
418 .alloc_pmd = paravirt_nop,
419 .alloc_pud = paravirt_nop,
420 .release_pte = paravirt_nop,
421 .release_pmd = paravirt_nop,
422 .release_pud = paravirt_nop,
424 .set_pte = native_set_pte,
425 .set_pte_at = native_set_pte_at,
426 .set_pmd = native_set_pmd,
427 .set_pmd_at = native_set_pmd_at,
428 .pte_update = paravirt_nop,
429 .pte_update_defer = paravirt_nop,
430 .pmd_update = paravirt_nop,
431 .pmd_update_defer = paravirt_nop,
433 .ptep_modify_prot_start = __ptep_modify_prot_start,
434 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
436 #if CONFIG_PGTABLE_LEVELS >= 3
437 #ifdef CONFIG_X86_PAE
438 .set_pte_atomic = native_set_pte_atomic,
439 .pte_clear = native_pte_clear,
440 .pmd_clear = native_pmd_clear,
442 .set_pud = native_set_pud,
444 .pmd_val = PTE_IDENT,
445 .make_pmd = PTE_IDENT,
447 #if CONFIG_PGTABLE_LEVELS == 4
448 .pud_val = PTE_IDENT,
449 .make_pud = PTE_IDENT,
451 .set_pgd = native_set_pgd,
453 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
455 .pte_val = PTE_IDENT,
456 .pgd_val = PTE_IDENT,
458 .make_pte = PTE_IDENT,
459 .make_pgd = PTE_IDENT,
461 .dup_mmap = paravirt_nop,
462 .exit_mmap = paravirt_nop,
463 .activate_mm = paravirt_nop,
466 .enter = paravirt_nop,
467 .leave = paravirt_nop,
468 .flush = paravirt_nop,
471 .set_fixmap = native_set_fixmap,
474 EXPORT_SYMBOL_GPL(pv_time_ops);
475 EXPORT_SYMBOL (pv_cpu_ops);
476 EXPORT_SYMBOL (pv_mmu_ops);
477 EXPORT_SYMBOL_GPL(pv_info);
478 EXPORT_SYMBOL (pv_irq_ops);