xref: /openbmc/linux/arch/x86/kernel/paravirt.c (revision 6c33a6f4)
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 #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  */
39 extern void _paravirt_nop(void);
40 asm (".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 
48 void __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. */
55 static const unsigned char ud2a[] = { 0x0f, 0x0b };
56 
57 struct branch {
58 	unsigned char opcode;
59 	u32 delta;
60 } __attribute__((packed));
61 
62 static 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 */
84 u64 notrace _paravirt_ident_64(u64 x)
85 {
86 	return x;
87 }
88 
89 static 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 
109 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
110 
111 void __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 
117 unsigned 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 
150 unsigned 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 
163 static void native_flush_tlb(void)
164 {
165 	__native_flush_tlb();
166 }
167 
168 /*
169  * Global pages have to be flushed a bit differently. Not a real
170  * performance problem because this does not happen often.
171  */
172 static void native_flush_tlb_global(void)
173 {
174 	__native_flush_tlb_global();
175 }
176 
177 static void native_flush_tlb_one_user(unsigned long addr)
178 {
179 	__native_flush_tlb_one_user(addr);
180 }
181 
182 struct static_key paravirt_steal_enabled;
183 struct static_key paravirt_steal_rq_enabled;
184 
185 static u64 native_steal_clock(int cpu)
186 {
187 	return 0;
188 }
189 
190 /* These are in entry.S */
191 extern void native_iret(void);
192 extern void native_usergs_sysret64(void);
193 
194 static struct resource reserve_ioports = {
195 	.start = 0,
196 	.end = IO_SPACE_LIMIT,
197 	.name = "paravirt-ioport",
198 	.flags = IORESOURCE_IO | IORESOURCE_BUSY,
199 };
200 
201 /*
202  * Reserve the whole legacy IO space to prevent any legacy drivers
203  * from wasting time probing for their hardware.  This is a fairly
204  * brute-force approach to disabling all non-virtual drivers.
205  *
206  * Note that this must be called very early to have any effect.
207  */
208 int paravirt_disable_iospace(void)
209 {
210 	return request_resource(&ioport_resource, &reserve_ioports);
211 }
212 
213 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
214 
215 static inline void enter_lazy(enum paravirt_lazy_mode mode)
216 {
217 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
218 
219 	this_cpu_write(paravirt_lazy_mode, mode);
220 }
221 
222 static void leave_lazy(enum paravirt_lazy_mode mode)
223 {
224 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
225 
226 	this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
227 }
228 
229 void paravirt_enter_lazy_mmu(void)
230 {
231 	enter_lazy(PARAVIRT_LAZY_MMU);
232 }
233 
234 void paravirt_leave_lazy_mmu(void)
235 {
236 	leave_lazy(PARAVIRT_LAZY_MMU);
237 }
238 
239 void paravirt_flush_lazy_mmu(void)
240 {
241 	preempt_disable();
242 
243 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
244 		arch_leave_lazy_mmu_mode();
245 		arch_enter_lazy_mmu_mode();
246 	}
247 
248 	preempt_enable();
249 }
250 
251 #ifdef CONFIG_PARAVIRT_XXL
252 void paravirt_start_context_switch(struct task_struct *prev)
253 {
254 	BUG_ON(preemptible());
255 
256 	if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
257 		arch_leave_lazy_mmu_mode();
258 		set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
259 	}
260 	enter_lazy(PARAVIRT_LAZY_CPU);
261 }
262 
263 void paravirt_end_context_switch(struct task_struct *next)
264 {
265 	BUG_ON(preemptible());
266 
267 	leave_lazy(PARAVIRT_LAZY_CPU);
268 
269 	if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
270 		arch_enter_lazy_mmu_mode();
271 }
272 #endif
273 
274 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
275 {
276 	if (in_interrupt())
277 		return PARAVIRT_LAZY_NONE;
278 
279 	return this_cpu_read(paravirt_lazy_mode);
280 }
281 
282 struct pv_info pv_info = {
283 	.name = "bare hardware",
284 #ifdef CONFIG_PARAVIRT_XXL
285 	.kernel_rpl = 0,
286 	.shared_kernel_pmd = 1,	/* Only used when CONFIG_X86_PAE is set */
287 
288 #ifdef CONFIG_X86_64
289 	.extra_user_64bit_cs = __USER_CS,
290 #endif
291 #endif
292 };
293 
294 /* 64-bit pagetable entries */
295 #define PTE_IDENT	__PV_IS_CALLEE_SAVE(_paravirt_ident_64)
296 
297 struct paravirt_patch_template pv_ops = {
298 	/* Init ops. */
299 	.init.patch		= native_patch,
300 
301 	/* Time ops. */
302 	.time.sched_clock	= native_sched_clock,
303 	.time.steal_clock	= native_steal_clock,
304 
305 	/* Cpu ops. */
306 	.cpu.io_delay		= native_io_delay,
307 
308 #ifdef CONFIG_PARAVIRT_XXL
309 	.cpu.cpuid		= native_cpuid,
310 	.cpu.get_debugreg	= native_get_debugreg,
311 	.cpu.set_debugreg	= native_set_debugreg,
312 	.cpu.read_cr0		= native_read_cr0,
313 	.cpu.write_cr0		= native_write_cr0,
314 	.cpu.write_cr4		= native_write_cr4,
315 	.cpu.wbinvd		= native_wbinvd,
316 	.cpu.read_msr		= native_read_msr,
317 	.cpu.write_msr		= native_write_msr,
318 	.cpu.read_msr_safe	= native_read_msr_safe,
319 	.cpu.write_msr_safe	= native_write_msr_safe,
320 	.cpu.read_pmc		= native_read_pmc,
321 	.cpu.load_tr_desc	= native_load_tr_desc,
322 	.cpu.set_ldt		= native_set_ldt,
323 	.cpu.load_gdt		= native_load_gdt,
324 	.cpu.load_idt		= native_load_idt,
325 	.cpu.store_tr		= native_store_tr,
326 	.cpu.load_tls		= native_load_tls,
327 #ifdef CONFIG_X86_64
328 	.cpu.load_gs_index	= native_load_gs_index,
329 #endif
330 	.cpu.write_ldt_entry	= native_write_ldt_entry,
331 	.cpu.write_gdt_entry	= native_write_gdt_entry,
332 	.cpu.write_idt_entry	= native_write_idt_entry,
333 
334 	.cpu.alloc_ldt		= paravirt_nop,
335 	.cpu.free_ldt		= paravirt_nop,
336 
337 	.cpu.load_sp0		= native_load_sp0,
338 
339 #ifdef CONFIG_X86_64
340 	.cpu.usergs_sysret64	= native_usergs_sysret64,
341 #endif
342 	.cpu.iret		= native_iret,
343 	.cpu.swapgs		= native_swapgs,
344 
345 #ifdef CONFIG_X86_IOPL_IOPERM
346 	.cpu.update_io_bitmap	= native_tss_update_io_bitmap,
347 #endif
348 
349 	.cpu.start_context_switch	= paravirt_nop,
350 	.cpu.end_context_switch		= paravirt_nop,
351 
352 	/* Irq ops. */
353 	.irq.save_fl		= __PV_IS_CALLEE_SAVE(native_save_fl),
354 	.irq.restore_fl		= __PV_IS_CALLEE_SAVE(native_restore_fl),
355 	.irq.irq_disable	= __PV_IS_CALLEE_SAVE(native_irq_disable),
356 	.irq.irq_enable		= __PV_IS_CALLEE_SAVE(native_irq_enable),
357 	.irq.safe_halt		= native_safe_halt,
358 	.irq.halt		= native_halt,
359 #endif /* CONFIG_PARAVIRT_XXL */
360 
361 	/* Mmu ops. */
362 	.mmu.flush_tlb_user	= native_flush_tlb,
363 	.mmu.flush_tlb_kernel	= native_flush_tlb_global,
364 	.mmu.flush_tlb_one_user	= native_flush_tlb_one_user,
365 	.mmu.flush_tlb_others	= native_flush_tlb_others,
366 	.mmu.tlb_remove_table	=
367 			(void (*)(struct mmu_gather *, void *))tlb_remove_page,
368 
369 	.mmu.exit_mmap		= paravirt_nop,
370 
371 #ifdef CONFIG_PARAVIRT_XXL
372 	.mmu.read_cr2		= __PV_IS_CALLEE_SAVE(native_read_cr2),
373 	.mmu.write_cr2		= native_write_cr2,
374 	.mmu.read_cr3		= __native_read_cr3,
375 	.mmu.write_cr3		= native_write_cr3,
376 
377 	.mmu.pgd_alloc		= __paravirt_pgd_alloc,
378 	.mmu.pgd_free		= paravirt_nop,
379 
380 	.mmu.alloc_pte		= paravirt_nop,
381 	.mmu.alloc_pmd		= paravirt_nop,
382 	.mmu.alloc_pud		= paravirt_nop,
383 	.mmu.alloc_p4d		= paravirt_nop,
384 	.mmu.release_pte	= paravirt_nop,
385 	.mmu.release_pmd	= paravirt_nop,
386 	.mmu.release_pud	= paravirt_nop,
387 	.mmu.release_p4d	= paravirt_nop,
388 
389 	.mmu.set_pte		= native_set_pte,
390 	.mmu.set_pte_at		= native_set_pte_at,
391 	.mmu.set_pmd		= native_set_pmd,
392 
393 	.mmu.ptep_modify_prot_start	= __ptep_modify_prot_start,
394 	.mmu.ptep_modify_prot_commit	= __ptep_modify_prot_commit,
395 
396 #if CONFIG_PGTABLE_LEVELS >= 3
397 #ifdef CONFIG_X86_PAE
398 	.mmu.set_pte_atomic	= native_set_pte_atomic,
399 	.mmu.pte_clear		= native_pte_clear,
400 	.mmu.pmd_clear		= native_pmd_clear,
401 #endif
402 	.mmu.set_pud		= native_set_pud,
403 
404 	.mmu.pmd_val		= PTE_IDENT,
405 	.mmu.make_pmd		= PTE_IDENT,
406 
407 #if CONFIG_PGTABLE_LEVELS >= 4
408 	.mmu.pud_val		= PTE_IDENT,
409 	.mmu.make_pud		= PTE_IDENT,
410 
411 	.mmu.set_p4d		= native_set_p4d,
412 
413 #if CONFIG_PGTABLE_LEVELS >= 5
414 	.mmu.p4d_val		= PTE_IDENT,
415 	.mmu.make_p4d		= PTE_IDENT,
416 
417 	.mmu.set_pgd		= native_set_pgd,
418 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
419 #endif /* CONFIG_PGTABLE_LEVELS >= 4 */
420 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
421 
422 	.mmu.pte_val		= PTE_IDENT,
423 	.mmu.pgd_val		= PTE_IDENT,
424 
425 	.mmu.make_pte		= PTE_IDENT,
426 	.mmu.make_pgd		= PTE_IDENT,
427 
428 	.mmu.dup_mmap		= paravirt_nop,
429 	.mmu.activate_mm	= paravirt_nop,
430 
431 	.mmu.lazy_mode = {
432 		.enter		= paravirt_nop,
433 		.leave		= paravirt_nop,
434 		.flush		= paravirt_nop,
435 	},
436 
437 	.mmu.set_fixmap		= native_set_fixmap,
438 #endif /* CONFIG_PARAVIRT_XXL */
439 
440 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
441 	/* Lock ops. */
442 #ifdef CONFIG_SMP
443 	.lock.queued_spin_lock_slowpath	= native_queued_spin_lock_slowpath,
444 	.lock.queued_spin_unlock	=
445 				PV_CALLEE_SAVE(__native_queued_spin_unlock),
446 	.lock.wait			= paravirt_nop,
447 	.lock.kick			= paravirt_nop,
448 	.lock.vcpu_is_preempted		=
449 				PV_CALLEE_SAVE(__native_vcpu_is_preempted),
450 #endif /* SMP */
451 #endif
452 };
453 
454 #ifdef CONFIG_PARAVIRT_XXL
455 /* At this point, native_get/set_debugreg has real function entries */
456 NOKPROBE_SYMBOL(native_get_debugreg);
457 NOKPROBE_SYMBOL(native_set_debugreg);
458 NOKPROBE_SYMBOL(native_load_idt);
459 #endif
460 
461 EXPORT_SYMBOL(pv_ops);
462 EXPORT_SYMBOL_GPL(pv_info);
463