xref: /linux/arch/x86/kernel/paravirt.c (revision c0e297dc61f8d4453e07afbea1fa8d0e67cd4a34)
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 #include <linux/kprobes.h>
27 
28 #include <asm/bug.h>
29 #include <asm/paravirt.h>
30 #include <asm/debugreg.h>
31 #include <asm/desc.h>
32 #include <asm/setup.h>
33 #include <asm/pgtable.h>
34 #include <asm/time.h>
35 #include <asm/pgalloc.h>
36 #include <asm/irq.h>
37 #include <asm/delay.h>
38 #include <asm/fixmap.h>
39 #include <asm/apic.h>
40 #include <asm/tlbflush.h>
41 #include <asm/timer.h>
42 #include <asm/special_insns.h>
43 
44 /* nop stub */
45 void _paravirt_nop(void)
46 {
47 }
48 
49 /* identity function, which can be inlined */
50 u32 _paravirt_ident_32(u32 x)
51 {
52 	return x;
53 }
54 
55 u64 _paravirt_ident_64(u64 x)
56 {
57 	return x;
58 }
59 
60 void __init default_banner(void)
61 {
62 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
63 	       pv_info.name);
64 }
65 
66 /* Undefined instruction for dealing with missing ops pointers. */
67 static const unsigned char ud2a[] = { 0x0f, 0x0b };
68 
69 unsigned paravirt_patch_nop(void)
70 {
71 	return 0;
72 }
73 
74 unsigned paravirt_patch_ignore(unsigned len)
75 {
76 	return len;
77 }
78 
79 struct branch {
80 	unsigned char opcode;
81 	u32 delta;
82 } __attribute__((packed));
83 
84 unsigned paravirt_patch_call(void *insnbuf,
85 			     const void *target, u16 tgt_clobbers,
86 			     unsigned long addr, u16 site_clobbers,
87 			     unsigned len)
88 {
89 	struct branch *b = insnbuf;
90 	unsigned long delta = (unsigned long)target - (addr+5);
91 
92 	if (tgt_clobbers & ~site_clobbers)
93 		return len;	/* target would clobber too much for this site */
94 	if (len < 5)
95 		return len;	/* call too long for patch site */
96 
97 	b->opcode = 0xe8; /* call */
98 	b->delta = delta;
99 	BUILD_BUG_ON(sizeof(*b) != 5);
100 
101 	return 5;
102 }
103 
104 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
105 			    unsigned long addr, unsigned len)
106 {
107 	struct branch *b = insnbuf;
108 	unsigned long delta = (unsigned long)target - (addr+5);
109 
110 	if (len < 5)
111 		return len;	/* call too long for patch site */
112 
113 	b->opcode = 0xe9;	/* jmp */
114 	b->delta = delta;
115 
116 	return 5;
117 }
118 
119 /* Neat trick to map patch type back to the call within the
120  * corresponding structure. */
121 static void *get_call_destination(u8 type)
122 {
123 	struct paravirt_patch_template tmpl = {
124 		.pv_init_ops = pv_init_ops,
125 		.pv_time_ops = pv_time_ops,
126 		.pv_cpu_ops = pv_cpu_ops,
127 		.pv_irq_ops = pv_irq_ops,
128 		.pv_apic_ops = pv_apic_ops,
129 		.pv_mmu_ops = pv_mmu_ops,
130 #ifdef CONFIG_PARAVIRT_SPINLOCKS
131 		.pv_lock_ops = pv_lock_ops,
132 #endif
133 	};
134 	return *((void **)&tmpl + type);
135 }
136 
137 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
138 				unsigned long addr, unsigned len)
139 {
140 	void *opfunc = get_call_destination(type);
141 	unsigned ret;
142 
143 	if (opfunc == NULL)
144 		/* If there's no function, patch it with a ud2a (BUG) */
145 		ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
146 	else if (opfunc == _paravirt_nop)
147 		/* If the operation is a nop, then nop the callsite */
148 		ret = paravirt_patch_nop();
149 
150 	/* identity functions just return their single argument */
151 	else if (opfunc == _paravirt_ident_32)
152 		ret = paravirt_patch_ident_32(insnbuf, len);
153 	else if (opfunc == _paravirt_ident_64)
154 		ret = paravirt_patch_ident_64(insnbuf, len);
155 
156 	else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
157 #ifdef CONFIG_X86_32
158 		 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
159 #endif
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 
173 unsigned 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 
186 static 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  */
195 static void native_flush_tlb_global(void)
196 {
197 	__native_flush_tlb_global();
198 }
199 
200 static void native_flush_tlb_single(unsigned long addr)
201 {
202 	__native_flush_tlb_single(addr);
203 }
204 
205 struct static_key paravirt_steal_enabled;
206 struct static_key paravirt_steal_rq_enabled;
207 
208 static u64 native_steal_clock(int cpu)
209 {
210 	return 0;
211 }
212 
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);
218 
219 static 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  */
233 int paravirt_disable_iospace(void)
234 {
235 	return request_resource(&ioport_resource, &reserve_ioports);
236 }
237 
238 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
239 
240 static inline void enter_lazy(enum paravirt_lazy_mode mode)
241 {
242 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
243 
244 	this_cpu_write(paravirt_lazy_mode, mode);
245 }
246 
247 static void leave_lazy(enum paravirt_lazy_mode mode)
248 {
249 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
250 
251 	this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
252 }
253 
254 void paravirt_enter_lazy_mmu(void)
255 {
256 	enter_lazy(PARAVIRT_LAZY_MMU);
257 }
258 
259 void paravirt_leave_lazy_mmu(void)
260 {
261 	leave_lazy(PARAVIRT_LAZY_MMU);
262 }
263 
264 void paravirt_flush_lazy_mmu(void)
265 {
266 	preempt_disable();
267 
268 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
269 		arch_leave_lazy_mmu_mode();
270 		arch_enter_lazy_mmu_mode();
271 	}
272 
273 	preempt_enable();
274 }
275 
276 void paravirt_start_context_switch(struct task_struct *prev)
277 {
278 	BUG_ON(preemptible());
279 
280 	if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
281 		arch_leave_lazy_mmu_mode();
282 		set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
283 	}
284 	enter_lazy(PARAVIRT_LAZY_CPU);
285 }
286 
287 void paravirt_end_context_switch(struct task_struct *next)
288 {
289 	BUG_ON(preemptible());
290 
291 	leave_lazy(PARAVIRT_LAZY_CPU);
292 
293 	if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
294 		arch_enter_lazy_mmu_mode();
295 }
296 
297 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
298 {
299 	if (in_interrupt())
300 		return PARAVIRT_LAZY_NONE;
301 
302 	return this_cpu_read(paravirt_lazy_mode);
303 }
304 
305 struct 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 
316 struct pv_init_ops pv_init_ops = {
317 	.patch = native_patch,
318 };
319 
320 struct pv_time_ops pv_time_ops = {
321 	.sched_clock = native_sched_clock,
322 	.steal_clock = native_steal_clock,
323 };
324 
325 __visible struct 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 
337 __visible struct 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 	.write_msr = native_write_msr_safe,
354 	.read_tsc = native_read_tsc,
355 	.read_pmc = native_read_pmc,
356 	.read_tscp = native_read_tscp,
357 	.load_tr_desc = native_load_tr_desc,
358 	.set_ldt = native_set_ldt,
359 	.load_gdt = native_load_gdt,
360 	.load_idt = native_load_idt,
361 	.store_idt = native_store_idt,
362 	.store_tr = native_store_tr,
363 	.load_tls = native_load_tls,
364 #ifdef CONFIG_X86_64
365 	.load_gs_index = native_load_gs_index,
366 #endif
367 	.write_ldt_entry = native_write_ldt_entry,
368 	.write_gdt_entry = native_write_gdt_entry,
369 	.write_idt_entry = native_write_idt_entry,
370 
371 	.alloc_ldt = paravirt_nop,
372 	.free_ldt = paravirt_nop,
373 
374 	.load_sp0 = native_load_sp0,
375 
376 #if defined(CONFIG_X86_32)
377 	.irq_enable_sysexit = native_irq_enable_sysexit,
378 #endif
379 #ifdef CONFIG_X86_64
380 #ifdef CONFIG_IA32_EMULATION
381 	.usergs_sysret32 = native_usergs_sysret32,
382 #endif
383 	.usergs_sysret64 = native_usergs_sysret64,
384 #endif
385 	.iret = native_iret,
386 	.swapgs = native_swapgs,
387 
388 	.set_iopl_mask = native_set_iopl_mask,
389 	.io_delay = native_io_delay,
390 
391 	.start_context_switch = paravirt_nop,
392 	.end_context_switch = paravirt_nop,
393 };
394 
395 /* At this point, native_get/set_debugreg has real function entries */
396 NOKPROBE_SYMBOL(native_get_debugreg);
397 NOKPROBE_SYMBOL(native_set_debugreg);
398 NOKPROBE_SYMBOL(native_load_idt);
399 
400 struct pv_apic_ops pv_apic_ops = {
401 #ifdef CONFIG_X86_LOCAL_APIC
402 	.startup_ipi_hook = paravirt_nop,
403 #endif
404 };
405 
406 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
407 /* 32-bit pagetable entries */
408 #define PTE_IDENT	__PV_IS_CALLEE_SAVE(_paravirt_ident_32)
409 #else
410 /* 64-bit pagetable entries */
411 #define PTE_IDENT	__PV_IS_CALLEE_SAVE(_paravirt_ident_64)
412 #endif
413 
414 struct pv_mmu_ops pv_mmu_ops = {
415 
416 	.read_cr2 = native_read_cr2,
417 	.write_cr2 = native_write_cr2,
418 	.read_cr3 = native_read_cr3,
419 	.write_cr3 = native_write_cr3,
420 
421 	.flush_tlb_user = native_flush_tlb,
422 	.flush_tlb_kernel = native_flush_tlb_global,
423 	.flush_tlb_single = native_flush_tlb_single,
424 	.flush_tlb_others = native_flush_tlb_others,
425 
426 	.pgd_alloc = __paravirt_pgd_alloc,
427 	.pgd_free = paravirt_nop,
428 
429 	.alloc_pte = paravirt_nop,
430 	.alloc_pmd = paravirt_nop,
431 	.alloc_pud = paravirt_nop,
432 	.release_pte = paravirt_nop,
433 	.release_pmd = paravirt_nop,
434 	.release_pud = paravirt_nop,
435 
436 	.set_pte = native_set_pte,
437 	.set_pte_at = native_set_pte_at,
438 	.set_pmd = native_set_pmd,
439 	.set_pmd_at = native_set_pmd_at,
440 	.pte_update = paravirt_nop,
441 	.pte_update_defer = paravirt_nop,
442 	.pmd_update = paravirt_nop,
443 	.pmd_update_defer = paravirt_nop,
444 
445 	.ptep_modify_prot_start = __ptep_modify_prot_start,
446 	.ptep_modify_prot_commit = __ptep_modify_prot_commit,
447 
448 #if CONFIG_PGTABLE_LEVELS >= 3
449 #ifdef CONFIG_X86_PAE
450 	.set_pte_atomic = native_set_pte_atomic,
451 	.pte_clear = native_pte_clear,
452 	.pmd_clear = native_pmd_clear,
453 #endif
454 	.set_pud = native_set_pud,
455 
456 	.pmd_val = PTE_IDENT,
457 	.make_pmd = PTE_IDENT,
458 
459 #if CONFIG_PGTABLE_LEVELS == 4
460 	.pud_val = PTE_IDENT,
461 	.make_pud = PTE_IDENT,
462 
463 	.set_pgd = native_set_pgd,
464 #endif
465 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
466 
467 	.pte_val = PTE_IDENT,
468 	.pgd_val = PTE_IDENT,
469 
470 	.make_pte = PTE_IDENT,
471 	.make_pgd = PTE_IDENT,
472 
473 	.dup_mmap = paravirt_nop,
474 	.exit_mmap = paravirt_nop,
475 	.activate_mm = paravirt_nop,
476 
477 	.lazy_mode = {
478 		.enter = paravirt_nop,
479 		.leave = paravirt_nop,
480 		.flush = paravirt_nop,
481 	},
482 
483 	.set_fixmap = native_set_fixmap,
484 };
485 
486 EXPORT_SYMBOL_GPL(pv_time_ops);
487 EXPORT_SYMBOL    (pv_cpu_ops);
488 EXPORT_SYMBOL    (pv_mmu_ops);
489 EXPORT_SYMBOL_GPL(pv_apic_ops);
490 EXPORT_SYMBOL_GPL(pv_info);
491 EXPORT_SYMBOL    (pv_irq_ops);
492