xref: /linux/arch/x86/kernel/paravirt.c (revision 6fdcba32711044c35c0e1b094cbd8f3f0b4472c9)
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 
17 #include <asm/bug.h>
18 #include <asm/paravirt.h>
19 #include <asm/debugreg.h>
20 #include <asm/desc.h>
21 #include <asm/setup.h>
22 #include <asm/pgtable.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 
34 /*
35  * nop stub, which must not clobber anything *including the stack* to
36  * avoid confusing the entry prologues.
37  */
38 extern void _paravirt_nop(void);
39 asm (".pushsection .entry.text, \"ax\"\n"
40      ".global _paravirt_nop\n"
41      "_paravirt_nop:\n\t"
42      "ret\n\t"
43      ".size _paravirt_nop, . - _paravirt_nop\n\t"
44      ".type _paravirt_nop, @function\n\t"
45      ".popsection");
46 
47 void __init default_banner(void)
48 {
49 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
50 	       pv_info.name);
51 }
52 
53 /* Undefined instruction for dealing with missing ops pointers. */
54 static const unsigned char ud2a[] = { 0x0f, 0x0b };
55 
56 struct branch {
57 	unsigned char opcode;
58 	u32 delta;
59 } __attribute__((packed));
60 
61 static unsigned paravirt_patch_call(void *insn_buff, const void *target,
62 				    unsigned long addr, unsigned len)
63 {
64 	const int call_len = 5;
65 	struct branch *b = insn_buff;
66 	unsigned long delta = (unsigned long)target - (addr+call_len);
67 
68 	if (len < call_len) {
69 		pr_warn("paravirt: Failed to patch indirect CALL at %ps\n", (void *)addr);
70 		/* Kernel might not be viable if patching fails, bail out: */
71 		BUG_ON(1);
72 	}
73 
74 	b->opcode = 0xe8; /* call */
75 	b->delta = delta;
76 	BUILD_BUG_ON(sizeof(*b) != call_len);
77 
78 	return call_len;
79 }
80 
81 #ifdef CONFIG_PARAVIRT_XXL
82 /* identity function, which can be inlined */
83 u64 notrace _paravirt_ident_64(u64 x)
84 {
85 	return x;
86 }
87 
88 static unsigned paravirt_patch_jmp(void *insn_buff, const void *target,
89 				   unsigned long addr, unsigned len)
90 {
91 	struct branch *b = insn_buff;
92 	unsigned long delta = (unsigned long)target - (addr+5);
93 
94 	if (len < 5) {
95 #ifdef CONFIG_RETPOLINE
96 		WARN_ONCE(1, "Failing to patch indirect JMP in %ps\n", (void *)addr);
97 #endif
98 		return len;	/* call too long for patch site */
99 	}
100 
101 	b->opcode = 0xe9;	/* jmp */
102 	b->delta = delta;
103 
104 	return 5;
105 }
106 #endif
107 
108 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
109 
110 void __init native_pv_lock_init(void)
111 {
112 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
113 		static_branch_disable(&virt_spin_lock_key);
114 }
115 
116 unsigned paravirt_patch_default(u8 type, void *insn_buff,
117 				unsigned long addr, unsigned len)
118 {
119 	/*
120 	 * Neat trick to map patch type back to the call within the
121 	 * corresponding structure.
122 	 */
123 	void *opfunc = *((void **)&pv_ops + type);
124 	unsigned ret;
125 
126 	if (opfunc == NULL)
127 		/* If there's no function, patch it with a ud2a (BUG) */
128 		ret = paravirt_patch_insns(insn_buff, len, ud2a, ud2a+sizeof(ud2a));
129 	else if (opfunc == _paravirt_nop)
130 		ret = 0;
131 
132 #ifdef CONFIG_PARAVIRT_XXL
133 	/* identity functions just return their single argument */
134 	else if (opfunc == _paravirt_ident_64)
135 		ret = paravirt_patch_ident_64(insn_buff, len);
136 
137 	else if (type == PARAVIRT_PATCH(cpu.iret) ||
138 		 type == PARAVIRT_PATCH(cpu.usergs_sysret64))
139 		/* If operation requires a jmp, then jmp */
140 		ret = paravirt_patch_jmp(insn_buff, opfunc, addr, len);
141 #endif
142 	else
143 		/* Otherwise call the function. */
144 		ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
145 
146 	return ret;
147 }
148 
149 unsigned paravirt_patch_insns(void *insn_buff, unsigned len,
150 			      const char *start, const char *end)
151 {
152 	unsigned insn_len = end - start;
153 
154 	/* Alternative instruction is too large for the patch site and we cannot continue: */
155 	BUG_ON(insn_len > len || start == NULL);
156 
157 	memcpy(insn_buff, start, insn_len);
158 
159 	return insn_len;
160 }
161 
162 static void native_flush_tlb(void)
163 {
164 	__native_flush_tlb();
165 }
166 
167 /*
168  * Global pages have to be flushed a bit differently. Not a real
169  * performance problem because this does not happen often.
170  */
171 static void native_flush_tlb_global(void)
172 {
173 	__native_flush_tlb_global();
174 }
175 
176 static void native_flush_tlb_one_user(unsigned long addr)
177 {
178 	__native_flush_tlb_one_user(addr);
179 }
180 
181 struct static_key paravirt_steal_enabled;
182 struct static_key paravirt_steal_rq_enabled;
183 
184 static u64 native_steal_clock(int cpu)
185 {
186 	return 0;
187 }
188 
189 /* These are in entry.S */
190 extern void native_iret(void);
191 extern void native_usergs_sysret64(void);
192 
193 static struct resource reserve_ioports = {
194 	.start = 0,
195 	.end = IO_SPACE_LIMIT,
196 	.name = "paravirt-ioport",
197 	.flags = IORESOURCE_IO | IORESOURCE_BUSY,
198 };
199 
200 /*
201  * Reserve the whole legacy IO space to prevent any legacy drivers
202  * from wasting time probing for their hardware.  This is a fairly
203  * brute-force approach to disabling all non-virtual drivers.
204  *
205  * Note that this must be called very early to have any effect.
206  */
207 int paravirt_disable_iospace(void)
208 {
209 	return request_resource(&ioport_resource, &reserve_ioports);
210 }
211 
212 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
213 
214 static inline void enter_lazy(enum paravirt_lazy_mode mode)
215 {
216 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
217 
218 	this_cpu_write(paravirt_lazy_mode, mode);
219 }
220 
221 static void leave_lazy(enum paravirt_lazy_mode mode)
222 {
223 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
224 
225 	this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
226 }
227 
228 void paravirt_enter_lazy_mmu(void)
229 {
230 	enter_lazy(PARAVIRT_LAZY_MMU);
231 }
232 
233 void paravirt_leave_lazy_mmu(void)
234 {
235 	leave_lazy(PARAVIRT_LAZY_MMU);
236 }
237 
238 void paravirt_flush_lazy_mmu(void)
239 {
240 	preempt_disable();
241 
242 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
243 		arch_leave_lazy_mmu_mode();
244 		arch_enter_lazy_mmu_mode();
245 	}
246 
247 	preempt_enable();
248 }
249 
250 #ifdef CONFIG_PARAVIRT_XXL
251 void paravirt_start_context_switch(struct task_struct *prev)
252 {
253 	BUG_ON(preemptible());
254 
255 	if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
256 		arch_leave_lazy_mmu_mode();
257 		set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
258 	}
259 	enter_lazy(PARAVIRT_LAZY_CPU);
260 }
261 
262 void paravirt_end_context_switch(struct task_struct *next)
263 {
264 	BUG_ON(preemptible());
265 
266 	leave_lazy(PARAVIRT_LAZY_CPU);
267 
268 	if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
269 		arch_enter_lazy_mmu_mode();
270 }
271 #endif
272 
273 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
274 {
275 	if (in_interrupt())
276 		return PARAVIRT_LAZY_NONE;
277 
278 	return this_cpu_read(paravirt_lazy_mode);
279 }
280 
281 struct pv_info pv_info = {
282 	.name = "bare hardware",
283 #ifdef CONFIG_PARAVIRT_XXL
284 	.kernel_rpl = 0,
285 	.shared_kernel_pmd = 1,	/* Only used when CONFIG_X86_PAE is set */
286 
287 #ifdef CONFIG_X86_64
288 	.extra_user_64bit_cs = __USER_CS,
289 #endif
290 #endif
291 };
292 
293 /* 64-bit pagetable entries */
294 #define PTE_IDENT	__PV_IS_CALLEE_SAVE(_paravirt_ident_64)
295 
296 struct paravirt_patch_template pv_ops = {
297 	/* Init ops. */
298 	.init.patch		= native_patch,
299 
300 	/* Time ops. */
301 	.time.sched_clock	= native_sched_clock,
302 	.time.steal_clock	= native_steal_clock,
303 
304 	/* Cpu ops. */
305 	.cpu.io_delay		= native_io_delay,
306 
307 #ifdef CONFIG_PARAVIRT_XXL
308 	.cpu.cpuid		= native_cpuid,
309 	.cpu.get_debugreg	= native_get_debugreg,
310 	.cpu.set_debugreg	= native_set_debugreg,
311 	.cpu.read_cr0		= native_read_cr0,
312 	.cpu.write_cr0		= native_write_cr0,
313 	.cpu.write_cr4		= native_write_cr4,
314 	.cpu.wbinvd		= native_wbinvd,
315 	.cpu.read_msr		= native_read_msr,
316 	.cpu.write_msr		= native_write_msr,
317 	.cpu.read_msr_safe	= native_read_msr_safe,
318 	.cpu.write_msr_safe	= native_write_msr_safe,
319 	.cpu.read_pmc		= native_read_pmc,
320 	.cpu.load_tr_desc	= native_load_tr_desc,
321 	.cpu.set_ldt		= native_set_ldt,
322 	.cpu.load_gdt		= native_load_gdt,
323 	.cpu.load_idt		= native_load_idt,
324 	.cpu.store_tr		= native_store_tr,
325 	.cpu.load_tls		= native_load_tls,
326 #ifdef CONFIG_X86_64
327 	.cpu.load_gs_index	= native_load_gs_index,
328 #endif
329 	.cpu.write_ldt_entry	= native_write_ldt_entry,
330 	.cpu.write_gdt_entry	= native_write_gdt_entry,
331 	.cpu.write_idt_entry	= native_write_idt_entry,
332 
333 	.cpu.alloc_ldt		= paravirt_nop,
334 	.cpu.free_ldt		= paravirt_nop,
335 
336 	.cpu.load_sp0		= native_load_sp0,
337 
338 #ifdef CONFIG_X86_64
339 	.cpu.usergs_sysret64	= native_usergs_sysret64,
340 #endif
341 	.cpu.iret		= native_iret,
342 	.cpu.swapgs		= native_swapgs,
343 
344 	.cpu.start_context_switch	= paravirt_nop,
345 	.cpu.end_context_switch		= paravirt_nop,
346 
347 	/* Irq ops. */
348 	.irq.save_fl		= __PV_IS_CALLEE_SAVE(native_save_fl),
349 	.irq.restore_fl		= __PV_IS_CALLEE_SAVE(native_restore_fl),
350 	.irq.irq_disable	= __PV_IS_CALLEE_SAVE(native_irq_disable),
351 	.irq.irq_enable		= __PV_IS_CALLEE_SAVE(native_irq_enable),
352 	.irq.safe_halt		= native_safe_halt,
353 	.irq.halt		= native_halt,
354 #endif /* CONFIG_PARAVIRT_XXL */
355 
356 	/* Mmu ops. */
357 	.mmu.flush_tlb_user	= native_flush_tlb,
358 	.mmu.flush_tlb_kernel	= native_flush_tlb_global,
359 	.mmu.flush_tlb_one_user	= native_flush_tlb_one_user,
360 	.mmu.flush_tlb_others	= native_flush_tlb_others,
361 	.mmu.tlb_remove_table	=
362 			(void (*)(struct mmu_gather *, void *))tlb_remove_page,
363 
364 	.mmu.exit_mmap		= paravirt_nop,
365 
366 #ifdef CONFIG_PARAVIRT_XXL
367 	.mmu.read_cr2		= __PV_IS_CALLEE_SAVE(native_read_cr2),
368 	.mmu.write_cr2		= native_write_cr2,
369 	.mmu.read_cr3		= __native_read_cr3,
370 	.mmu.write_cr3		= native_write_cr3,
371 
372 	.mmu.pgd_alloc		= __paravirt_pgd_alloc,
373 	.mmu.pgd_free		= paravirt_nop,
374 
375 	.mmu.alloc_pte		= paravirt_nop,
376 	.mmu.alloc_pmd		= paravirt_nop,
377 	.mmu.alloc_pud		= paravirt_nop,
378 	.mmu.alloc_p4d		= paravirt_nop,
379 	.mmu.release_pte	= paravirt_nop,
380 	.mmu.release_pmd	= paravirt_nop,
381 	.mmu.release_pud	= paravirt_nop,
382 	.mmu.release_p4d	= paravirt_nop,
383 
384 	.mmu.set_pte		= native_set_pte,
385 	.mmu.set_pte_at		= native_set_pte_at,
386 	.mmu.set_pmd		= native_set_pmd,
387 
388 	.mmu.ptep_modify_prot_start	= __ptep_modify_prot_start,
389 	.mmu.ptep_modify_prot_commit	= __ptep_modify_prot_commit,
390 
391 #if CONFIG_PGTABLE_LEVELS >= 3
392 #ifdef CONFIG_X86_PAE
393 	.mmu.set_pte_atomic	= native_set_pte_atomic,
394 	.mmu.pte_clear		= native_pte_clear,
395 	.mmu.pmd_clear		= native_pmd_clear,
396 #endif
397 	.mmu.set_pud		= native_set_pud,
398 
399 	.mmu.pmd_val		= PTE_IDENT,
400 	.mmu.make_pmd		= PTE_IDENT,
401 
402 #if CONFIG_PGTABLE_LEVELS >= 4
403 	.mmu.pud_val		= PTE_IDENT,
404 	.mmu.make_pud		= PTE_IDENT,
405 
406 	.mmu.set_p4d		= native_set_p4d,
407 
408 #if CONFIG_PGTABLE_LEVELS >= 5
409 	.mmu.p4d_val		= PTE_IDENT,
410 	.mmu.make_p4d		= PTE_IDENT,
411 
412 	.mmu.set_pgd		= native_set_pgd,
413 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
414 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */
415 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
416 
417 	.mmu.pte_val		= PTE_IDENT,
418 	.mmu.pgd_val		= PTE_IDENT,
419 
420 	.mmu.make_pte		= PTE_IDENT,
421 	.mmu.make_pgd		= PTE_IDENT,
422 
423 	.mmu.dup_mmap		= paravirt_nop,
424 	.mmu.activate_mm	= paravirt_nop,
425 
426 	.mmu.lazy_mode = {
427 		.enter		= paravirt_nop,
428 		.leave		= paravirt_nop,
429 		.flush		= paravirt_nop,
430 	},
431 
432 	.mmu.set_fixmap		= native_set_fixmap,
433 #endif /* CONFIG_PARAVIRT_XXL */
434 
435 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
436 	/* Lock ops. */
437 #ifdef CONFIG_SMP
438 	.lock.queued_spin_lock_slowpath	= native_queued_spin_lock_slowpath,
439 	.lock.queued_spin_unlock	=
440 				PV_CALLEE_SAVE(__native_queued_spin_unlock),
441 	.lock.wait			= paravirt_nop,
442 	.lock.kick			= paravirt_nop,
443 	.lock.vcpu_is_preempted		=
444 				PV_CALLEE_SAVE(__native_vcpu_is_preempted),
445 #endif /* SMP */
446 #endif
447 };
448 
449 #ifdef CONFIG_PARAVIRT_XXL
450 /* At this point, native_get/set_debugreg has real function entries */
451 NOKPROBE_SYMBOL(native_get_debugreg);
452 NOKPROBE_SYMBOL(native_set_debugreg);
453 NOKPROBE_SYMBOL(native_load_idt);
454 #endif
455 
456 EXPORT_SYMBOL(pv_ops);
457 EXPORT_SYMBOL_GPL(pv_info);
458