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_ENDBR
45 ASM_RET
46 ".size _paravirt_nop, . - _paravirt_nop\n\t"
47 ".type _paravirt_nop, @function\n\t"
48 ".popsection");
49
50 /* stub always returning 0. */
51 asm (".pushsection .entry.text, \"ax\"\n"
52 ".global paravirt_ret0\n"
53 "paravirt_ret0:\n\t"
54 ASM_ENDBR
55 "xor %" _ASM_AX ", %" _ASM_AX ";\n\t"
56 ASM_RET
57 ".size paravirt_ret0, . - paravirt_ret0\n\t"
58 ".type paravirt_ret0, @function\n\t"
59 ".popsection");
60
61
default_banner(void)62 void __init default_banner(void)
63 {
64 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
65 pv_info.name);
66 }
67
68 /* Undefined instruction for dealing with missing ops pointers. */
paravirt_BUG(void)69 noinstr void paravirt_BUG(void)
70 {
71 BUG();
72 }
73
paravirt_patch_call(void * insn_buff,const void * target,unsigned long addr,unsigned len)74 static unsigned paravirt_patch_call(void *insn_buff, const void *target,
75 unsigned long addr, unsigned len)
76 {
77 __text_gen_insn(insn_buff, CALL_INSN_OPCODE,
78 (void *)addr, target, CALL_INSN_SIZE);
79 return CALL_INSN_SIZE;
80 }
81
82 #ifdef CONFIG_PARAVIRT_XXL
83 /* identity function, which can be inlined */
_paravirt_ident_64(u64 x)84 u64 notrace _paravirt_ident_64(u64 x)
85 {
86 return x;
87 }
88 #endif
89
90 DEFINE_STATIC_KEY_TRUE(virt_spin_lock_key);
91
native_pv_lock_init(void)92 void __init native_pv_lock_init(void)
93 {
94 if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
95 static_branch_disable(&virt_spin_lock_key);
96 }
97
paravirt_patch(u8 type,void * insn_buff,unsigned long addr,unsigned int len)98 unsigned int paravirt_patch(u8 type, void *insn_buff, unsigned long addr,
99 unsigned int len)
100 {
101 /*
102 * Neat trick to map patch type back to the call within the
103 * corresponding structure.
104 */
105 void *opfunc = *((void **)&pv_ops + type);
106 unsigned ret;
107
108 if (opfunc == NULL)
109 /* If there's no function, patch it with paravirt_BUG() */
110 ret = paravirt_patch_call(insn_buff, paravirt_BUG, addr, len);
111 else if (opfunc == _paravirt_nop)
112 ret = 0;
113 else
114 /* Otherwise call the function. */
115 ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
116
117 return ret;
118 }
119
120 struct static_key paravirt_steal_enabled;
121 struct static_key paravirt_steal_rq_enabled;
122
native_steal_clock(int cpu)123 static u64 native_steal_clock(int cpu)
124 {
125 return 0;
126 }
127
128 DEFINE_STATIC_CALL(pv_steal_clock, native_steal_clock);
129 DEFINE_STATIC_CALL(pv_sched_clock, native_sched_clock);
130
paravirt_set_sched_clock(u64 (* func)(void))131 void paravirt_set_sched_clock(u64 (*func)(void))
132 {
133 static_call_update(pv_sched_clock, func);
134 }
135
136 /* These are in entry.S */
137 static struct resource reserve_ioports = {
138 .start = 0,
139 .end = IO_SPACE_LIMIT,
140 .name = "paravirt-ioport",
141 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
142 };
143
144 /*
145 * Reserve the whole legacy IO space to prevent any legacy drivers
146 * from wasting time probing for their hardware. This is a fairly
147 * brute-force approach to disabling all non-virtual drivers.
148 *
149 * Note that this must be called very early to have any effect.
150 */
paravirt_disable_iospace(void)151 int paravirt_disable_iospace(void)
152 {
153 return request_resource(&ioport_resource, &reserve_ioports);
154 }
155
156 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
157
enter_lazy(enum paravirt_lazy_mode mode)158 static inline void enter_lazy(enum paravirt_lazy_mode mode)
159 {
160 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
161
162 this_cpu_write(paravirt_lazy_mode, mode);
163 }
164
leave_lazy(enum paravirt_lazy_mode mode)165 static void leave_lazy(enum paravirt_lazy_mode mode)
166 {
167 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
168
169 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
170 }
171
paravirt_enter_lazy_mmu(void)172 void paravirt_enter_lazy_mmu(void)
173 {
174 enter_lazy(PARAVIRT_LAZY_MMU);
175 }
176
paravirt_leave_lazy_mmu(void)177 void paravirt_leave_lazy_mmu(void)
178 {
179 leave_lazy(PARAVIRT_LAZY_MMU);
180 }
181
paravirt_flush_lazy_mmu(void)182 void paravirt_flush_lazy_mmu(void)
183 {
184 preempt_disable();
185
186 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
187 arch_leave_lazy_mmu_mode();
188 arch_enter_lazy_mmu_mode();
189 }
190
191 preempt_enable();
192 }
193
194 #ifdef CONFIG_PARAVIRT_XXL
paravirt_start_context_switch(struct task_struct * prev)195 void paravirt_start_context_switch(struct task_struct *prev)
196 {
197 BUG_ON(preemptible());
198
199 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
200 arch_leave_lazy_mmu_mode();
201 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
202 }
203 enter_lazy(PARAVIRT_LAZY_CPU);
204 }
205
paravirt_end_context_switch(struct task_struct * next)206 void paravirt_end_context_switch(struct task_struct *next)
207 {
208 BUG_ON(preemptible());
209
210 leave_lazy(PARAVIRT_LAZY_CPU);
211
212 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
213 arch_enter_lazy_mmu_mode();
214 }
215
pv_native_read_cr2(void)216 static noinstr unsigned long pv_native_read_cr2(void)
217 {
218 return native_read_cr2();
219 }
220
pv_native_write_cr2(unsigned long val)221 static noinstr void pv_native_write_cr2(unsigned long val)
222 {
223 native_write_cr2(val);
224 }
225
pv_native_get_debugreg(int regno)226 static noinstr unsigned long pv_native_get_debugreg(int regno)
227 {
228 return native_get_debugreg(regno);
229 }
230
pv_native_set_debugreg(int regno,unsigned long val)231 static noinstr void pv_native_set_debugreg(int regno, unsigned long val)
232 {
233 native_set_debugreg(regno, val);
234 }
235
pv_native_irq_enable(void)236 static noinstr void pv_native_irq_enable(void)
237 {
238 native_irq_enable();
239 }
240
pv_native_irq_disable(void)241 static noinstr void pv_native_irq_disable(void)
242 {
243 native_irq_disable();
244 }
245 #endif
246
paravirt_get_lazy_mode(void)247 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
248 {
249 if (in_interrupt())
250 return PARAVIRT_LAZY_NONE;
251
252 return this_cpu_read(paravirt_lazy_mode);
253 }
254
255 struct pv_info pv_info = {
256 .name = "bare hardware",
257 #ifdef CONFIG_PARAVIRT_XXL
258 .extra_user_64bit_cs = __USER_CS,
259 #endif
260 };
261
262 /* 64-bit pagetable entries */
263 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
264
265 struct paravirt_patch_template pv_ops = {
266 /* Cpu ops. */
267 .cpu.io_delay = native_io_delay,
268
269 #ifdef CONFIG_PARAVIRT_XXL
270 .cpu.cpuid = native_cpuid,
271 .cpu.get_debugreg = pv_native_get_debugreg,
272 .cpu.set_debugreg = pv_native_set_debugreg,
273 .cpu.read_cr0 = native_read_cr0,
274 .cpu.write_cr0 = native_write_cr0,
275 .cpu.write_cr4 = native_write_cr4,
276 .cpu.wbinvd = native_wbinvd,
277 .cpu.read_msr = native_read_msr,
278 .cpu.write_msr = native_write_msr,
279 .cpu.read_msr_safe = native_read_msr_safe,
280 .cpu.write_msr_safe = native_write_msr_safe,
281 .cpu.read_pmc = native_read_pmc,
282 .cpu.load_tr_desc = native_load_tr_desc,
283 .cpu.set_ldt = native_set_ldt,
284 .cpu.load_gdt = native_load_gdt,
285 .cpu.load_idt = native_load_idt,
286 .cpu.store_tr = native_store_tr,
287 .cpu.load_tls = native_load_tls,
288 .cpu.load_gs_index = native_load_gs_index,
289 .cpu.write_ldt_entry = native_write_ldt_entry,
290 .cpu.write_gdt_entry = native_write_gdt_entry,
291 .cpu.write_idt_entry = native_write_idt_entry,
292
293 .cpu.alloc_ldt = paravirt_nop,
294 .cpu.free_ldt = paravirt_nop,
295
296 .cpu.load_sp0 = native_load_sp0,
297
298 #ifdef CONFIG_X86_IOPL_IOPERM
299 .cpu.invalidate_io_bitmap = native_tss_invalidate_io_bitmap,
300 .cpu.update_io_bitmap = native_tss_update_io_bitmap,
301 #endif
302
303 .cpu.start_context_switch = paravirt_nop,
304 .cpu.end_context_switch = paravirt_nop,
305
306 /* Irq ops. */
307 .irq.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
308 .irq.irq_disable = __PV_IS_CALLEE_SAVE(pv_native_irq_disable),
309 .irq.irq_enable = __PV_IS_CALLEE_SAVE(pv_native_irq_enable),
310 .irq.safe_halt = native_safe_halt,
311 .irq.halt = native_halt,
312 #endif /* CONFIG_PARAVIRT_XXL */
313
314 /* Mmu ops. */
315 .mmu.flush_tlb_user = native_flush_tlb_local,
316 .mmu.flush_tlb_kernel = native_flush_tlb_global,
317 .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
318 .mmu.flush_tlb_multi = native_flush_tlb_multi,
319 .mmu.tlb_remove_table =
320 (void (*)(struct mmu_gather *, void *))tlb_remove_page,
321
322 .mmu.exit_mmap = paravirt_nop,
323 .mmu.notify_page_enc_status_changed = paravirt_nop,
324
325 #ifdef CONFIG_PARAVIRT_XXL
326 .mmu.read_cr2 = __PV_IS_CALLEE_SAVE(pv_native_read_cr2),
327 .mmu.write_cr2 = pv_native_write_cr2,
328 .mmu.read_cr3 = __native_read_cr3,
329 .mmu.write_cr3 = native_write_cr3,
330
331 .mmu.pgd_alloc = __paravirt_pgd_alloc,
332 .mmu.pgd_free = paravirt_nop,
333
334 .mmu.alloc_pte = paravirt_nop,
335 .mmu.alloc_pmd = paravirt_nop,
336 .mmu.alloc_pud = paravirt_nop,
337 .mmu.alloc_p4d = paravirt_nop,
338 .mmu.release_pte = paravirt_nop,
339 .mmu.release_pmd = paravirt_nop,
340 .mmu.release_pud = paravirt_nop,
341 .mmu.release_p4d = paravirt_nop,
342
343 .mmu.set_pte = native_set_pte,
344 .mmu.set_pmd = native_set_pmd,
345
346 .mmu.ptep_modify_prot_start = __ptep_modify_prot_start,
347 .mmu.ptep_modify_prot_commit = __ptep_modify_prot_commit,
348
349 .mmu.set_pud = native_set_pud,
350
351 .mmu.pmd_val = PTE_IDENT,
352 .mmu.make_pmd = PTE_IDENT,
353
354 .mmu.pud_val = PTE_IDENT,
355 .mmu.make_pud = PTE_IDENT,
356
357 .mmu.set_p4d = native_set_p4d,
358
359 #if CONFIG_PGTABLE_LEVELS >= 5
360 .mmu.p4d_val = PTE_IDENT,
361 .mmu.make_p4d = PTE_IDENT,
362
363 .mmu.set_pgd = native_set_pgd,
364 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
365
366 .mmu.pte_val = PTE_IDENT,
367 .mmu.pgd_val = PTE_IDENT,
368
369 .mmu.make_pte = PTE_IDENT,
370 .mmu.make_pgd = PTE_IDENT,
371
372 .mmu.dup_mmap = paravirt_nop,
373 .mmu.activate_mm = paravirt_nop,
374
375 .mmu.lazy_mode = {
376 .enter = paravirt_nop,
377 .leave = paravirt_nop,
378 .flush = paravirt_nop,
379 },
380
381 .mmu.set_fixmap = native_set_fixmap,
382 #endif /* CONFIG_PARAVIRT_XXL */
383
384 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
385 /* Lock ops. */
386 #ifdef CONFIG_SMP
387 .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
388 .lock.queued_spin_unlock =
389 PV_CALLEE_SAVE(__native_queued_spin_unlock),
390 .lock.wait = paravirt_nop,
391 .lock.kick = paravirt_nop,
392 .lock.vcpu_is_preempted =
393 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
394 #endif /* SMP */
395 #endif
396 };
397
398 #ifdef CONFIG_PARAVIRT_XXL
399 NOKPROBE_SYMBOL(native_load_idt);
400 #endif
401
402 EXPORT_SYMBOL(pv_ops);
403 EXPORT_SYMBOL_GPL(pv_info);
404