Loading...
1/* Paravirtualization interfaces
2 Copyright (C) 2006 Rusty Russell IBM Corporation
3
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.
8
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.
13
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
17
18 2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
19*/
20
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
27#include <asm/bug.h>
28#include <asm/paravirt.h>
29#include <asm/desc.h>
30#include <asm/setup.h>
31#include <asm/pgtable.h>
32#include <asm/time.h>
33#include <asm/pgalloc.h>
34#include <asm/irq.h>
35#include <asm/delay.h>
36#include <asm/fixmap.h>
37#include <asm/apic.h>
38#include <asm/tlbflush.h>
39#include <asm/timer.h>
40
41/* nop stub */
42void _paravirt_nop(void)
43{
44}
45
46/* identity function, which can be inlined */
47u32 _paravirt_ident_32(u32 x)
48{
49 return x;
50}
51
52u64 _paravirt_ident_64(u64 x)
53{
54 return x;
55}
56
57void __init default_banner(void)
58{
59 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
60 pv_info.name);
61}
62
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 ":")
67
68/* Undefined instruction for dealing with missing ops pointers. */
69static const unsigned char ud2a[] = { 0x0f, 0x0b };
70
71unsigned paravirt_patch_nop(void)
72{
73 return 0;
74}
75
76unsigned paravirt_patch_ignore(unsigned len)
77{
78 return len;
79}
80
81struct branch {
82 unsigned char opcode;
83 u32 delta;
84} __attribute__((packed));
85
86unsigned paravirt_patch_call(void *insnbuf,
87 const void *target, u16 tgt_clobbers,
88 unsigned long addr, u16 site_clobbers,
89 unsigned len)
90{
91 struct branch *b = insnbuf;
92 unsigned long delta = (unsigned long)target - (addr+5);
93
94 if (tgt_clobbers & ~site_clobbers)
95 return len; /* target would clobber too much for this site */
96 if (len < 5)
97 return len; /* call too long for patch site */
98
99 b->opcode = 0xe8; /* call */
100 b->delta = delta;
101 BUILD_BUG_ON(sizeof(*b) != 5);
102
103 return 5;
104}
105
106unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
107 unsigned long addr, unsigned len)
108{
109 struct branch *b = insnbuf;
110 unsigned long delta = (unsigned long)target - (addr+5);
111
112 if (len < 5)
113 return len; /* call too long for patch site */
114
115 b->opcode = 0xe9; /* jmp */
116 b->delta = delta;
117
118 return 5;
119}
120
121/* Neat trick to map patch type back to the call within the
122 * corresponding structure. */
123static void *get_call_destination(u8 type)
124{
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,
134#endif
135 };
136 return *((void **)&tmpl + type);
137}
138
139unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
140 unsigned long addr, unsigned len)
141{
142 void *opfunc = get_call_destination(type);
143 unsigned ret;
144
145 if (opfunc == NULL)
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();
151
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);
157
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);
164 else
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);
169
170 return ret;
171}
172
173unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
174 const char *start, const char *end)
175{
176 unsigned insn_len = end - start;
177
178 if (insn_len > len || start == NULL)
179 insn_len = len;
180 else
181 memcpy(insnbuf, start, insn_len);
182
183 return insn_len;
184}
185
186static void native_flush_tlb(void)
187{
188 __native_flush_tlb();
189}
190
191/*
192 * Global pages have to be flushed a bit differently. Not a real
193 * performance problem because this does not happen often.
194 */
195static void native_flush_tlb_global(void)
196{
197 __native_flush_tlb_global();
198}
199
200static void native_flush_tlb_single(unsigned long addr)
201{
202 __native_flush_tlb_single(addr);
203}
204
205struct jump_label_key paravirt_steal_enabled;
206struct jump_label_key paravirt_steal_rq_enabled;
207
208static u64 native_steal_clock(int cpu)
209{
210 return 0;
211}
212
213/* These are in entry.S */
214extern void native_iret(void);
215extern void native_irq_enable_sysexit(void);
216extern void native_usergs_sysret32(void);
217extern void native_usergs_sysret64(void);
218
219static struct resource reserve_ioports = {
220 .start = 0,
221 .end = IO_SPACE_LIMIT,
222 .name = "paravirt-ioport",
223 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
224};
225
226/*
227 * Reserve the whole legacy IO space to prevent any legacy drivers
228 * from wasting time probing for their hardware. This is a fairly
229 * brute-force approach to disabling all non-virtual drivers.
230 *
231 * Note that this must be called very early to have any effect.
232 */
233int paravirt_disable_iospace(void)
234{
235 return request_resource(&ioport_resource, &reserve_ioports);
236}
237
238static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
239
240static inline void enter_lazy(enum paravirt_lazy_mode mode)
241{
242 BUG_ON(percpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
243
244 percpu_write(paravirt_lazy_mode, mode);
245}
246
247static void leave_lazy(enum paravirt_lazy_mode mode)
248{
249 BUG_ON(percpu_read(paravirt_lazy_mode) != mode);
250
251 percpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
252}
253
254void paravirt_enter_lazy_mmu(void)
255{
256 enter_lazy(PARAVIRT_LAZY_MMU);
257}
258
259void paravirt_leave_lazy_mmu(void)
260{
261 leave_lazy(PARAVIRT_LAZY_MMU);
262}
263
264void paravirt_start_context_switch(struct task_struct *prev)
265{
266 BUG_ON(preemptible());
267
268 if (percpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
269 arch_leave_lazy_mmu_mode();
270 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
271 }
272 enter_lazy(PARAVIRT_LAZY_CPU);
273}
274
275void paravirt_end_context_switch(struct task_struct *next)
276{
277 BUG_ON(preemptible());
278
279 leave_lazy(PARAVIRT_LAZY_CPU);
280
281 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
282 arch_enter_lazy_mmu_mode();
283}
284
285enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
286{
287 if (in_interrupt())
288 return PARAVIRT_LAZY_NONE;
289
290 return percpu_read(paravirt_lazy_mode);
291}
292
293void arch_flush_lazy_mmu_mode(void)
294{
295 preempt_disable();
296
297 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
298 arch_leave_lazy_mmu_mode();
299 arch_enter_lazy_mmu_mode();
300 }
301
302 preempt_enable();
303}
304
305struct pv_info pv_info = {
306 .name = "bare hardware",
307 .paravirt_enabled = 0,
308 .kernel_rpl = 0,
309 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
310
311#ifdef CONFIG_X86_64
312 .extra_user_64bit_cs = __USER_CS,
313#endif
314};
315
316struct pv_init_ops pv_init_ops = {
317 .patch = native_patch,
318};
319
320struct pv_time_ops pv_time_ops = {
321 .sched_clock = native_sched_clock,
322 .steal_clock = native_steal_clock,
323};
324
325struct pv_irq_ops pv_irq_ops = {
326 .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
327 .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
328 .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
329 .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
330 .safe_halt = native_safe_halt,
331 .halt = native_halt,
332#ifdef CONFIG_X86_64
333 .adjust_exception_frame = paravirt_nop,
334#endif
335};
336
337struct pv_cpu_ops pv_cpu_ops = {
338 .cpuid = native_cpuid,
339 .get_debugreg = native_get_debugreg,
340 .set_debugreg = native_set_debugreg,
341 .clts = native_clts,
342 .read_cr0 = native_read_cr0,
343 .write_cr0 = native_write_cr0,
344 .read_cr4 = native_read_cr4,
345 .read_cr4_safe = native_read_cr4_safe,
346 .write_cr4 = native_write_cr4,
347#ifdef CONFIG_X86_64
348 .read_cr8 = native_read_cr8,
349 .write_cr8 = native_write_cr8,
350#endif
351 .wbinvd = native_wbinvd,
352 .read_msr = native_read_msr_safe,
353 .rdmsr_regs = native_rdmsr_safe_regs,
354 .write_msr = native_write_msr_safe,
355 .wrmsr_regs = native_wrmsr_safe_regs,
356 .read_tsc = native_read_tsc,
357 .read_pmc = native_read_pmc,
358 .read_tscp = native_read_tscp,
359 .load_tr_desc = native_load_tr_desc,
360 .set_ldt = native_set_ldt,
361 .load_gdt = native_load_gdt,
362 .load_idt = native_load_idt,
363 .store_gdt = native_store_gdt,
364 .store_idt = native_store_idt,
365 .store_tr = native_store_tr,
366 .load_tls = native_load_tls,
367#ifdef CONFIG_X86_64
368 .load_gs_index = native_load_gs_index,
369#endif
370 .write_ldt_entry = native_write_ldt_entry,
371 .write_gdt_entry = native_write_gdt_entry,
372 .write_idt_entry = native_write_idt_entry,
373
374 .alloc_ldt = paravirt_nop,
375 .free_ldt = paravirt_nop,
376
377 .load_sp0 = native_load_sp0,
378
379#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
380 .irq_enable_sysexit = native_irq_enable_sysexit,
381#endif
382#ifdef CONFIG_X86_64
383#ifdef CONFIG_IA32_EMULATION
384 .usergs_sysret32 = native_usergs_sysret32,
385#endif
386 .usergs_sysret64 = native_usergs_sysret64,
387#endif
388 .iret = native_iret,
389 .swapgs = native_swapgs,
390
391 .set_iopl_mask = native_set_iopl_mask,
392 .io_delay = native_io_delay,
393
394 .start_context_switch = paravirt_nop,
395 .end_context_switch = paravirt_nop,
396};
397
398struct pv_apic_ops pv_apic_ops = {
399#ifdef CONFIG_X86_LOCAL_APIC
400 .startup_ipi_hook = paravirt_nop,
401#endif
402};
403
404#if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
405/* 32-bit pagetable entries */
406#define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
407#else
408/* 64-bit pagetable entries */
409#define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
410#endif
411
412struct pv_mmu_ops pv_mmu_ops = {
413
414 .read_cr2 = native_read_cr2,
415 .write_cr2 = native_write_cr2,
416 .read_cr3 = native_read_cr3,
417 .write_cr3 = native_write_cr3,
418
419 .flush_tlb_user = native_flush_tlb,
420 .flush_tlb_kernel = native_flush_tlb_global,
421 .flush_tlb_single = native_flush_tlb_single,
422 .flush_tlb_others = native_flush_tlb_others,
423
424 .pgd_alloc = __paravirt_pgd_alloc,
425 .pgd_free = paravirt_nop,
426
427 .alloc_pte = paravirt_nop,
428 .alloc_pmd = paravirt_nop,
429 .alloc_pud = paravirt_nop,
430 .release_pte = paravirt_nop,
431 .release_pmd = paravirt_nop,
432 .release_pud = paravirt_nop,
433
434 .set_pte = native_set_pte,
435 .set_pte_at = native_set_pte_at,
436 .set_pmd = native_set_pmd,
437 .set_pmd_at = native_set_pmd_at,
438 .pte_update = paravirt_nop,
439 .pte_update_defer = paravirt_nop,
440 .pmd_update = paravirt_nop,
441 .pmd_update_defer = paravirt_nop,
442
443 .ptep_modify_prot_start = __ptep_modify_prot_start,
444 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
445
446#if PAGETABLE_LEVELS >= 3
447#ifdef CONFIG_X86_PAE
448 .set_pte_atomic = native_set_pte_atomic,
449 .pte_clear = native_pte_clear,
450 .pmd_clear = native_pmd_clear,
451#endif
452 .set_pud = native_set_pud,
453
454 .pmd_val = PTE_IDENT,
455 .make_pmd = PTE_IDENT,
456
457#if PAGETABLE_LEVELS == 4
458 .pud_val = PTE_IDENT,
459 .make_pud = PTE_IDENT,
460
461 .set_pgd = native_set_pgd,
462#endif
463#endif /* PAGETABLE_LEVELS >= 3 */
464
465 .pte_val = PTE_IDENT,
466 .pgd_val = PTE_IDENT,
467
468 .make_pte = PTE_IDENT,
469 .make_pgd = PTE_IDENT,
470
471 .dup_mmap = paravirt_nop,
472 .exit_mmap = paravirt_nop,
473 .activate_mm = paravirt_nop,
474
475 .lazy_mode = {
476 .enter = paravirt_nop,
477 .leave = paravirt_nop,
478 },
479
480 .set_fixmap = native_set_fixmap,
481};
482
483EXPORT_SYMBOL_GPL(pv_time_ops);
484EXPORT_SYMBOL (pv_cpu_ops);
485EXPORT_SYMBOL (pv_mmu_ops);
486EXPORT_SYMBOL_GPL(pv_apic_ops);
487EXPORT_SYMBOL_GPL(pv_info);
488EXPORT_SYMBOL (pv_irq_ops);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Paravirtualization interfaces
3 Copyright (C) 2006 Rusty Russell IBM Corporation
4
5
6 2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
7*/
8
9#include <linux/errno.h>
10#include <linux/init.h>
11#include <linux/export.h>
12#include <linux/efi.h>
13#include <linux/bcd.h>
14#include <linux/highmem.h>
15#include <linux/kprobes.h>
16#include <linux/pgtable.h>
17
18#include <asm/bug.h>
19#include <asm/paravirt.h>
20#include <asm/debugreg.h>
21#include <asm/desc.h>
22#include <asm/setup.h>
23#include <asm/time.h>
24#include <asm/pgalloc.h>
25#include <asm/irq.h>
26#include <asm/delay.h>
27#include <asm/fixmap.h>
28#include <asm/apic.h>
29#include <asm/tlbflush.h>
30#include <asm/timer.h>
31#include <asm/special_insns.h>
32#include <asm/tlb.h>
33#include <asm/io_bitmap.h>
34
35/*
36 * nop stub, which must not clobber anything *including the stack* to
37 * avoid confusing the entry prologues.
38 */
39extern void _paravirt_nop(void);
40asm (".pushsection .entry.text, \"ax\"\n"
41 ".global _paravirt_nop\n"
42 "_paravirt_nop:\n\t"
43 "ret\n\t"
44 ".size _paravirt_nop, . - _paravirt_nop\n\t"
45 ".type _paravirt_nop, @function\n\t"
46 ".popsection");
47
48void __init default_banner(void)
49{
50 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
51 pv_info.name);
52}
53
54/* Undefined instruction for dealing with missing ops pointers. */
55static const unsigned char ud2a[] = { 0x0f, 0x0b };
56
57struct branch {
58 unsigned char opcode;
59 u32 delta;
60} __attribute__((packed));
61
62static unsigned paravirt_patch_call(void *insn_buff, const void *target,
63 unsigned long addr, unsigned len)
64{
65 const int call_len = 5;
66 struct branch *b = insn_buff;
67 unsigned long delta = (unsigned long)target - (addr+call_len);
68
69 if (len < call_len) {
70 pr_warn("paravirt: Failed to patch indirect CALL at %ps\n", (void *)addr);
71 /* Kernel might not be viable if patching fails, bail out: */
72 BUG_ON(1);
73 }
74
75 b->opcode = 0xe8; /* call */
76 b->delta = delta;
77 BUILD_BUG_ON(sizeof(*b) != call_len);
78
79 return call_len;
80}
81
82#ifdef CONFIG_PARAVIRT_XXL
83/* identity function, which can be inlined */
84u64 notrace _paravirt_ident_64(u64 x)
85{
86 return x;
87}
88
89static unsigned paravirt_patch_jmp(void *insn_buff, const void *target,
90 unsigned long addr, unsigned len)
91{
92 struct branch *b = insn_buff;
93 unsigned long delta = (unsigned long)target - (addr+5);
94
95 if (len < 5) {
96#ifdef CONFIG_RETPOLINE
97 WARN_ONCE(1, "Failing to patch indirect JMP in %ps\n", (void *)addr);
98#endif
99 return len; /* call too long for patch site */
100 }
101
102 b->opcode = 0xe9; /* jmp */
103 b->delta = delta;
104
105 return 5;
106}
107#endif
108
109DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
110
111void __init native_pv_lock_init(void)
112{
113 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
114 static_branch_disable(&virt_spin_lock_key);
115}
116
117unsigned paravirt_patch_default(u8 type, void *insn_buff,
118 unsigned long addr, unsigned len)
119{
120 /*
121 * Neat trick to map patch type back to the call within the
122 * corresponding structure.
123 */
124 void *opfunc = *((void **)&pv_ops + type);
125 unsigned ret;
126
127 if (opfunc == NULL)
128 /* If there's no function, patch it with a ud2a (BUG) */
129 ret = paravirt_patch_insns(insn_buff, len, ud2a, ud2a+sizeof(ud2a));
130 else if (opfunc == _paravirt_nop)
131 ret = 0;
132
133#ifdef CONFIG_PARAVIRT_XXL
134 /* identity functions just return their single argument */
135 else if (opfunc == _paravirt_ident_64)
136 ret = paravirt_patch_ident_64(insn_buff, len);
137
138 else if (type == PARAVIRT_PATCH(cpu.iret) ||
139 type == PARAVIRT_PATCH(cpu.usergs_sysret64))
140 /* If operation requires a jmp, then jmp */
141 ret = paravirt_patch_jmp(insn_buff, opfunc, addr, len);
142#endif
143 else
144 /* Otherwise call the function. */
145 ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
146
147 return ret;
148}
149
150unsigned paravirt_patch_insns(void *insn_buff, unsigned len,
151 const char *start, const char *end)
152{
153 unsigned insn_len = end - start;
154
155 /* Alternative instruction is too large for the patch site and we cannot continue: */
156 BUG_ON(insn_len > len || start == NULL);
157
158 memcpy(insn_buff, start, insn_len);
159
160 return insn_len;
161}
162
163struct static_key paravirt_steal_enabled;
164struct static_key paravirt_steal_rq_enabled;
165
166static u64 native_steal_clock(int cpu)
167{
168 return 0;
169}
170
171/* These are in entry.S */
172extern void native_iret(void);
173extern void native_usergs_sysret64(void);
174
175static struct resource reserve_ioports = {
176 .start = 0,
177 .end = IO_SPACE_LIMIT,
178 .name = "paravirt-ioport",
179 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
180};
181
182/*
183 * Reserve the whole legacy IO space to prevent any legacy drivers
184 * from wasting time probing for their hardware. This is a fairly
185 * brute-force approach to disabling all non-virtual drivers.
186 *
187 * Note that this must be called very early to have any effect.
188 */
189int paravirt_disable_iospace(void)
190{
191 return request_resource(&ioport_resource, &reserve_ioports);
192}
193
194static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
195
196static inline void enter_lazy(enum paravirt_lazy_mode mode)
197{
198 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
199
200 this_cpu_write(paravirt_lazy_mode, mode);
201}
202
203static void leave_lazy(enum paravirt_lazy_mode mode)
204{
205 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
206
207 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
208}
209
210void paravirt_enter_lazy_mmu(void)
211{
212 enter_lazy(PARAVIRT_LAZY_MMU);
213}
214
215void paravirt_leave_lazy_mmu(void)
216{
217 leave_lazy(PARAVIRT_LAZY_MMU);
218}
219
220void paravirt_flush_lazy_mmu(void)
221{
222 preempt_disable();
223
224 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
225 arch_leave_lazy_mmu_mode();
226 arch_enter_lazy_mmu_mode();
227 }
228
229 preempt_enable();
230}
231
232#ifdef CONFIG_PARAVIRT_XXL
233void paravirt_start_context_switch(struct task_struct *prev)
234{
235 BUG_ON(preemptible());
236
237 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
238 arch_leave_lazy_mmu_mode();
239 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
240 }
241 enter_lazy(PARAVIRT_LAZY_CPU);
242}
243
244void paravirt_end_context_switch(struct task_struct *next)
245{
246 BUG_ON(preemptible());
247
248 leave_lazy(PARAVIRT_LAZY_CPU);
249
250 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
251 arch_enter_lazy_mmu_mode();
252}
253#endif
254
255enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
256{
257 if (in_interrupt())
258 return PARAVIRT_LAZY_NONE;
259
260 return this_cpu_read(paravirt_lazy_mode);
261}
262
263struct pv_info pv_info = {
264 .name = "bare hardware",
265#ifdef CONFIG_PARAVIRT_XXL
266 .kernel_rpl = 0,
267 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
268
269#ifdef CONFIG_X86_64
270 .extra_user_64bit_cs = __USER_CS,
271#endif
272#endif
273};
274
275/* 64-bit pagetable entries */
276#define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
277
278struct paravirt_patch_template pv_ops = {
279 /* Init ops. */
280 .init.patch = native_patch,
281
282 /* Time ops. */
283 .time.sched_clock = native_sched_clock,
284 .time.steal_clock = native_steal_clock,
285
286 /* Cpu ops. */
287 .cpu.io_delay = native_io_delay,
288
289#ifdef CONFIG_PARAVIRT_XXL
290 .cpu.cpuid = native_cpuid,
291 .cpu.get_debugreg = native_get_debugreg,
292 .cpu.set_debugreg = native_set_debugreg,
293 .cpu.read_cr0 = native_read_cr0,
294 .cpu.write_cr0 = native_write_cr0,
295 .cpu.write_cr4 = native_write_cr4,
296 .cpu.wbinvd = native_wbinvd,
297 .cpu.read_msr = native_read_msr,
298 .cpu.write_msr = native_write_msr,
299 .cpu.read_msr_safe = native_read_msr_safe,
300 .cpu.write_msr_safe = native_write_msr_safe,
301 .cpu.read_pmc = native_read_pmc,
302 .cpu.load_tr_desc = native_load_tr_desc,
303 .cpu.set_ldt = native_set_ldt,
304 .cpu.load_gdt = native_load_gdt,
305 .cpu.load_idt = native_load_idt,
306 .cpu.store_tr = native_store_tr,
307 .cpu.load_tls = native_load_tls,
308#ifdef CONFIG_X86_64
309 .cpu.load_gs_index = native_load_gs_index,
310#endif
311 .cpu.write_ldt_entry = native_write_ldt_entry,
312 .cpu.write_gdt_entry = native_write_gdt_entry,
313 .cpu.write_idt_entry = native_write_idt_entry,
314
315 .cpu.alloc_ldt = paravirt_nop,
316 .cpu.free_ldt = paravirt_nop,
317
318 .cpu.load_sp0 = native_load_sp0,
319
320#ifdef CONFIG_X86_64
321 .cpu.usergs_sysret64 = native_usergs_sysret64,
322#endif
323 .cpu.iret = native_iret,
324 .cpu.swapgs = native_swapgs,
325
326#ifdef CONFIG_X86_IOPL_IOPERM
327 .cpu.invalidate_io_bitmap = native_tss_invalidate_io_bitmap,
328 .cpu.update_io_bitmap = native_tss_update_io_bitmap,
329#endif
330
331 .cpu.start_context_switch = paravirt_nop,
332 .cpu.end_context_switch = paravirt_nop,
333
334 /* Irq ops. */
335 .irq.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
336 .irq.restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
337 .irq.irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
338 .irq.irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
339 .irq.safe_halt = native_safe_halt,
340 .irq.halt = native_halt,
341#endif /* CONFIG_PARAVIRT_XXL */
342
343 /* Mmu ops. */
344 .mmu.flush_tlb_user = native_flush_tlb_local,
345 .mmu.flush_tlb_kernel = native_flush_tlb_global,
346 .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
347 .mmu.flush_tlb_others = native_flush_tlb_others,
348 .mmu.tlb_remove_table =
349 (void (*)(struct mmu_gather *, void *))tlb_remove_page,
350
351 .mmu.exit_mmap = paravirt_nop,
352
353#ifdef CONFIG_PARAVIRT_XXL
354 .mmu.read_cr2 = __PV_IS_CALLEE_SAVE(native_read_cr2),
355 .mmu.write_cr2 = native_write_cr2,
356 .mmu.read_cr3 = __native_read_cr3,
357 .mmu.write_cr3 = native_write_cr3,
358
359 .mmu.pgd_alloc = __paravirt_pgd_alloc,
360 .mmu.pgd_free = paravirt_nop,
361
362 .mmu.alloc_pte = paravirt_nop,
363 .mmu.alloc_pmd = paravirt_nop,
364 .mmu.alloc_pud = paravirt_nop,
365 .mmu.alloc_p4d = paravirt_nop,
366 .mmu.release_pte = paravirt_nop,
367 .mmu.release_pmd = paravirt_nop,
368 .mmu.release_pud = paravirt_nop,
369 .mmu.release_p4d = paravirt_nop,
370
371 .mmu.set_pte = native_set_pte,
372 .mmu.set_pte_at = native_set_pte_at,
373 .mmu.set_pmd = native_set_pmd,
374
375 .mmu.ptep_modify_prot_start = __ptep_modify_prot_start,
376 .mmu.ptep_modify_prot_commit = __ptep_modify_prot_commit,
377
378#if CONFIG_PGTABLE_LEVELS >= 3
379#ifdef CONFIG_X86_PAE
380 .mmu.set_pte_atomic = native_set_pte_atomic,
381 .mmu.pte_clear = native_pte_clear,
382 .mmu.pmd_clear = native_pmd_clear,
383#endif
384 .mmu.set_pud = native_set_pud,
385
386 .mmu.pmd_val = PTE_IDENT,
387 .mmu.make_pmd = PTE_IDENT,
388
389#if CONFIG_PGTABLE_LEVELS >= 4
390 .mmu.pud_val = PTE_IDENT,
391 .mmu.make_pud = PTE_IDENT,
392
393 .mmu.set_p4d = native_set_p4d,
394
395#if CONFIG_PGTABLE_LEVELS >= 5
396 .mmu.p4d_val = PTE_IDENT,
397 .mmu.make_p4d = PTE_IDENT,
398
399 .mmu.set_pgd = native_set_pgd,
400#endif /* CONFIG_PGTABLE_LEVELS >= 5 */
401#endif /* CONFIG_PGTABLE_LEVELS >= 4 */
402#endif /* CONFIG_PGTABLE_LEVELS >= 3 */
403
404 .mmu.pte_val = PTE_IDENT,
405 .mmu.pgd_val = PTE_IDENT,
406
407 .mmu.make_pte = PTE_IDENT,
408 .mmu.make_pgd = PTE_IDENT,
409
410 .mmu.dup_mmap = paravirt_nop,
411 .mmu.activate_mm = paravirt_nop,
412
413 .mmu.lazy_mode = {
414 .enter = paravirt_nop,
415 .leave = paravirt_nop,
416 .flush = paravirt_nop,
417 },
418
419 .mmu.set_fixmap = native_set_fixmap,
420#endif /* CONFIG_PARAVIRT_XXL */
421
422#if defined(CONFIG_PARAVIRT_SPINLOCKS)
423 /* Lock ops. */
424#ifdef CONFIG_SMP
425 .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
426 .lock.queued_spin_unlock =
427 PV_CALLEE_SAVE(__native_queued_spin_unlock),
428 .lock.wait = paravirt_nop,
429 .lock.kick = paravirt_nop,
430 .lock.vcpu_is_preempted =
431 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
432#endif /* SMP */
433#endif
434};
435
436#ifdef CONFIG_PARAVIRT_XXL
437/* At this point, native_get/set_debugreg has real function entries */
438NOKPROBE_SYMBOL(native_get_debugreg);
439NOKPROBE_SYMBOL(native_set_debugreg);
440NOKPROBE_SYMBOL(native_load_idt);
441#endif
442
443EXPORT_SYMBOL(pv_ops);
444EXPORT_SYMBOL_GPL(pv_info);