xref: /openbmc/linux/arch/x86/kernel/paravirt.c (revision 34fa67e7)
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 #include <linux/static_call.h>
18 
19 #include <asm/bug.h>
20 #include <asm/paravirt.h>
21 #include <asm/debugreg.h>
22 #include <asm/desc.h>
23 #include <asm/setup.h>
24 #include <asm/time.h>
25 #include <asm/pgalloc.h>
26 #include <asm/irq.h>
27 #include <asm/delay.h>
28 #include <asm/fixmap.h>
29 #include <asm/apic.h>
30 #include <asm/tlbflush.h>
31 #include <asm/timer.h>
32 #include <asm/special_insns.h>
33 #include <asm/tlb.h>
34 #include <asm/io_bitmap.h>
35 
36 /*
37  * nop stub, which must not clobber anything *including the stack* to
38  * avoid confusing the entry prologues.
39  */
40 extern void _paravirt_nop(void);
41 asm (".pushsection .entry.text, \"ax\"\n"
42      ".global _paravirt_nop\n"
43      "_paravirt_nop:\n\t"
44      ASM_RET
45      ".size _paravirt_nop, . - _paravirt_nop\n\t"
46      ".type _paravirt_nop, @function\n\t"
47      ".popsection");
48 
49 /* stub always returning 0. */
50 asm (".pushsection .entry.text, \"ax\"\n"
51      ".global paravirt_ret0\n"
52      "paravirt_ret0:\n\t"
53      "xor %" _ASM_AX ", %" _ASM_AX ";\n\t"
54      ASM_RET
55      ".size paravirt_ret0, . - paravirt_ret0\n\t"
56      ".type paravirt_ret0, @function\n\t"
57      ".popsection");
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 noinstr void paravirt_BUG(void)
68 {
69 	BUG();
70 }
71 
72 struct branch {
73 	unsigned char opcode;
74 	u32 delta;
75 } __attribute__((packed));
76 
77 static unsigned paravirt_patch_call(void *insn_buff, const void *target,
78 				    unsigned long addr, unsigned len)
79 {
80 	const int call_len = 5;
81 	struct branch *b = insn_buff;
82 	unsigned long delta = (unsigned long)target - (addr+call_len);
83 
84 	if (len < call_len) {
85 		pr_warn("paravirt: Failed to patch indirect CALL at %ps\n", (void *)addr);
86 		/* Kernel might not be viable if patching fails, bail out: */
87 		BUG_ON(1);
88 	}
89 
90 	b->opcode = 0xe8; /* call */
91 	b->delta = delta;
92 	BUILD_BUG_ON(sizeof(*b) != call_len);
93 
94 	return call_len;
95 }
96 
97 #ifdef CONFIG_PARAVIRT_XXL
98 /* identity function, which can be inlined */
99 u64 notrace _paravirt_ident_64(u64 x)
100 {
101 	return x;
102 }
103 #endif
104 
105 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
106 
107 void __init native_pv_lock_init(void)
108 {
109 	if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
110 		static_branch_disable(&virt_spin_lock_key);
111 }
112 
113 unsigned int paravirt_patch(u8 type, void *insn_buff, unsigned long addr,
114 			    unsigned int len)
115 {
116 	/*
117 	 * Neat trick to map patch type back to the call within the
118 	 * corresponding structure.
119 	 */
120 	void *opfunc = *((void **)&pv_ops + type);
121 	unsigned ret;
122 
123 	if (opfunc == NULL)
124 		/* If there's no function, patch it with paravirt_BUG() */
125 		ret = paravirt_patch_call(insn_buff, paravirt_BUG, addr, len);
126 	else if (opfunc == _paravirt_nop)
127 		ret = 0;
128 	else
129 		/* Otherwise call the function. */
130 		ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
131 
132 	return ret;
133 }
134 
135 struct static_key paravirt_steal_enabled;
136 struct static_key paravirt_steal_rq_enabled;
137 
138 static u64 native_steal_clock(int cpu)
139 {
140 	return 0;
141 }
142 
143 DEFINE_STATIC_CALL(pv_steal_clock, native_steal_clock);
144 DEFINE_STATIC_CALL(pv_sched_clock, native_sched_clock);
145 
146 void paravirt_set_sched_clock(u64 (*func)(void))
147 {
148 	static_call_update(pv_sched_clock, func);
149 }
150 
151 /* These are in entry.S */
152 extern void native_iret(void);
153 
154 static struct resource reserve_ioports = {
155 	.start = 0,
156 	.end = IO_SPACE_LIMIT,
157 	.name = "paravirt-ioport",
158 	.flags = IORESOURCE_IO | IORESOURCE_BUSY,
159 };
160 
161 /*
162  * Reserve the whole legacy IO space to prevent any legacy drivers
163  * from wasting time probing for their hardware.  This is a fairly
164  * brute-force approach to disabling all non-virtual drivers.
165  *
166  * Note that this must be called very early to have any effect.
167  */
168 int paravirt_disable_iospace(void)
169 {
170 	return request_resource(&ioport_resource, &reserve_ioports);
171 }
172 
173 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
174 
175 static inline void enter_lazy(enum paravirt_lazy_mode mode)
176 {
177 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
178 
179 	this_cpu_write(paravirt_lazy_mode, mode);
180 }
181 
182 static void leave_lazy(enum paravirt_lazy_mode mode)
183 {
184 	BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
185 
186 	this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
187 }
188 
189 void paravirt_enter_lazy_mmu(void)
190 {
191 	enter_lazy(PARAVIRT_LAZY_MMU);
192 }
193 
194 void paravirt_leave_lazy_mmu(void)
195 {
196 	leave_lazy(PARAVIRT_LAZY_MMU);
197 }
198 
199 void paravirt_flush_lazy_mmu(void)
200 {
201 	preempt_disable();
202 
203 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
204 		arch_leave_lazy_mmu_mode();
205 		arch_enter_lazy_mmu_mode();
206 	}
207 
208 	preempt_enable();
209 }
210 
211 #ifdef CONFIG_PARAVIRT_XXL
212 void paravirt_start_context_switch(struct task_struct *prev)
213 {
214 	BUG_ON(preemptible());
215 
216 	if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
217 		arch_leave_lazy_mmu_mode();
218 		set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
219 	}
220 	enter_lazy(PARAVIRT_LAZY_CPU);
221 }
222 
223 void paravirt_end_context_switch(struct task_struct *next)
224 {
225 	BUG_ON(preemptible());
226 
227 	leave_lazy(PARAVIRT_LAZY_CPU);
228 
229 	if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
230 		arch_enter_lazy_mmu_mode();
231 }
232 
233 static noinstr unsigned long pv_native_read_cr2(void)
234 {
235 	return native_read_cr2();
236 }
237 
238 static noinstr void pv_native_write_cr2(unsigned long val)
239 {
240 	native_write_cr2(val);
241 }
242 
243 static noinstr unsigned long pv_native_get_debugreg(int regno)
244 {
245 	return native_get_debugreg(regno);
246 }
247 
248 static noinstr void pv_native_set_debugreg(int regno, unsigned long val)
249 {
250 	native_set_debugreg(regno, val);
251 }
252 
253 static noinstr void pv_native_irq_enable(void)
254 {
255 	native_irq_enable();
256 }
257 
258 static noinstr void pv_native_irq_disable(void)
259 {
260 	native_irq_disable();
261 }
262 #endif
263 
264 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
265 {
266 	if (in_interrupt())
267 		return PARAVIRT_LAZY_NONE;
268 
269 	return this_cpu_read(paravirt_lazy_mode);
270 }
271 
272 struct pv_info pv_info = {
273 	.name = "bare hardware",
274 #ifdef CONFIG_PARAVIRT_XXL
275 	.extra_user_64bit_cs = __USER_CS,
276 #endif
277 };
278 
279 /* 64-bit pagetable entries */
280 #define PTE_IDENT	__PV_IS_CALLEE_SAVE(_paravirt_ident_64)
281 
282 struct paravirt_patch_template pv_ops = {
283 	/* Cpu ops. */
284 	.cpu.io_delay		= native_io_delay,
285 
286 #ifdef CONFIG_PARAVIRT_XXL
287 	.cpu.cpuid		= native_cpuid,
288 	.cpu.get_debugreg	= pv_native_get_debugreg,
289 	.cpu.set_debugreg	= pv_native_set_debugreg,
290 	.cpu.read_cr0		= native_read_cr0,
291 	.cpu.write_cr0		= native_write_cr0,
292 	.cpu.write_cr4		= native_write_cr4,
293 	.cpu.wbinvd		= native_wbinvd,
294 	.cpu.read_msr		= native_read_msr,
295 	.cpu.write_msr		= native_write_msr,
296 	.cpu.read_msr_safe	= native_read_msr_safe,
297 	.cpu.write_msr_safe	= native_write_msr_safe,
298 	.cpu.read_pmc		= native_read_pmc,
299 	.cpu.load_tr_desc	= native_load_tr_desc,
300 	.cpu.set_ldt		= native_set_ldt,
301 	.cpu.load_gdt		= native_load_gdt,
302 	.cpu.load_idt		= native_load_idt,
303 	.cpu.store_tr		= native_store_tr,
304 	.cpu.load_tls		= native_load_tls,
305 	.cpu.load_gs_index	= native_load_gs_index,
306 	.cpu.write_ldt_entry	= native_write_ldt_entry,
307 	.cpu.write_gdt_entry	= native_write_gdt_entry,
308 	.cpu.write_idt_entry	= native_write_idt_entry,
309 
310 	.cpu.alloc_ldt		= paravirt_nop,
311 	.cpu.free_ldt		= paravirt_nop,
312 
313 	.cpu.load_sp0		= native_load_sp0,
314 
315 #ifdef CONFIG_X86_IOPL_IOPERM
316 	.cpu.invalidate_io_bitmap	= native_tss_invalidate_io_bitmap,
317 	.cpu.update_io_bitmap		= native_tss_update_io_bitmap,
318 #endif
319 
320 	.cpu.start_context_switch	= paravirt_nop,
321 	.cpu.end_context_switch		= paravirt_nop,
322 
323 	/* Irq ops. */
324 	.irq.save_fl		= __PV_IS_CALLEE_SAVE(native_save_fl),
325 	.irq.irq_disable	= __PV_IS_CALLEE_SAVE(pv_native_irq_disable),
326 	.irq.irq_enable		= __PV_IS_CALLEE_SAVE(pv_native_irq_enable),
327 	.irq.safe_halt		= native_safe_halt,
328 	.irq.halt		= native_halt,
329 #endif /* CONFIG_PARAVIRT_XXL */
330 
331 	/* Mmu ops. */
332 	.mmu.flush_tlb_user	= native_flush_tlb_local,
333 	.mmu.flush_tlb_kernel	= native_flush_tlb_global,
334 	.mmu.flush_tlb_one_user	= native_flush_tlb_one_user,
335 	.mmu.flush_tlb_multi	= native_flush_tlb_multi,
336 	.mmu.tlb_remove_table	=
337 			(void (*)(struct mmu_gather *, void *))tlb_remove_page,
338 
339 	.mmu.exit_mmap		= paravirt_nop,
340 	.mmu.notify_page_enc_status_changed	= paravirt_nop,
341 
342 #ifdef CONFIG_PARAVIRT_XXL
343 	.mmu.read_cr2		= __PV_IS_CALLEE_SAVE(pv_native_read_cr2),
344 	.mmu.write_cr2		= pv_native_write_cr2,
345 	.mmu.read_cr3		= __native_read_cr3,
346 	.mmu.write_cr3		= native_write_cr3,
347 
348 	.mmu.pgd_alloc		= __paravirt_pgd_alloc,
349 	.mmu.pgd_free		= paravirt_nop,
350 
351 	.mmu.alloc_pte		= paravirt_nop,
352 	.mmu.alloc_pmd		= paravirt_nop,
353 	.mmu.alloc_pud		= paravirt_nop,
354 	.mmu.alloc_p4d		= paravirt_nop,
355 	.mmu.release_pte	= paravirt_nop,
356 	.mmu.release_pmd	= paravirt_nop,
357 	.mmu.release_pud	= paravirt_nop,
358 	.mmu.release_p4d	= paravirt_nop,
359 
360 	.mmu.set_pte		= native_set_pte,
361 	.mmu.set_pmd		= native_set_pmd,
362 
363 	.mmu.ptep_modify_prot_start	= __ptep_modify_prot_start,
364 	.mmu.ptep_modify_prot_commit	= __ptep_modify_prot_commit,
365 
366 	.mmu.set_pud		= native_set_pud,
367 
368 	.mmu.pmd_val		= PTE_IDENT,
369 	.mmu.make_pmd		= PTE_IDENT,
370 
371 	.mmu.pud_val		= PTE_IDENT,
372 	.mmu.make_pud		= PTE_IDENT,
373 
374 	.mmu.set_p4d		= native_set_p4d,
375 
376 #if CONFIG_PGTABLE_LEVELS >= 5
377 	.mmu.p4d_val		= PTE_IDENT,
378 	.mmu.make_p4d		= PTE_IDENT,
379 
380 	.mmu.set_pgd		= native_set_pgd,
381 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
382 
383 	.mmu.pte_val		= PTE_IDENT,
384 	.mmu.pgd_val		= PTE_IDENT,
385 
386 	.mmu.make_pte		= PTE_IDENT,
387 	.mmu.make_pgd		= PTE_IDENT,
388 
389 	.mmu.dup_mmap		= paravirt_nop,
390 	.mmu.activate_mm	= paravirt_nop,
391 
392 	.mmu.lazy_mode = {
393 		.enter		= paravirt_nop,
394 		.leave		= paravirt_nop,
395 		.flush		= paravirt_nop,
396 	},
397 
398 	.mmu.set_fixmap		= native_set_fixmap,
399 #endif /* CONFIG_PARAVIRT_XXL */
400 
401 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
402 	/* Lock ops. */
403 #ifdef CONFIG_SMP
404 	.lock.queued_spin_lock_slowpath	= native_queued_spin_lock_slowpath,
405 	.lock.queued_spin_unlock	=
406 				PV_CALLEE_SAVE(__native_queued_spin_unlock),
407 	.lock.wait			= paravirt_nop,
408 	.lock.kick			= paravirt_nop,
409 	.lock.vcpu_is_preempted		=
410 				PV_CALLEE_SAVE(__native_vcpu_is_preempted),
411 #endif /* SMP */
412 #endif
413 };
414 
415 #ifdef CONFIG_PARAVIRT_XXL
416 NOKPROBE_SYMBOL(native_load_idt);
417 
418 void (*paravirt_iret)(void) = native_iret;
419 #endif
420 
421 EXPORT_SYMBOL(pv_ops);
422 EXPORT_SYMBOL_GPL(pv_info);
423