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
18 #include <asm/bug.h>
19 #include <asm/paravirt.h>
20 #include <asm/debugreg.h>
21 #include <asm/desc.h>
22 #include <asm/setup.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 ASM_RET
44 ".size _paravirt_nop, . - _paravirt_nop\n\t"
45 ".type _paravirt_nop, @function\n\t"
46 ".popsection");
47
default_banner(void)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
paravirt_patch_call(void * insn_buff,const void * target,unsigned long addr,unsigned len)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 */
_paravirt_ident_64(u64 x)84 u64 notrace _paravirt_ident_64(u64 x)
85 {
86 return x;
87 }
88
paravirt_patch_jmp(void * insn_buff,const void * target,unsigned long addr,unsigned len)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
native_pv_lock_init(void)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
paravirt_patch_default(u8 type,void * insn_buff,unsigned long addr,unsigned len)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
paravirt_patch_insns(void * insn_buff,unsigned len,const char * start,const char * end)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 struct static_key paravirt_steal_enabled;
164 struct static_key paravirt_steal_rq_enabled;
165
native_steal_clock(int cpu)166 static u64 native_steal_clock(int cpu)
167 {
168 return 0;
169 }
170
171 /* These are in entry.S */
172 extern void native_iret(void);
173 extern void native_usergs_sysret64(void);
174
175 static struct resource reserve_ioports = {
176 .start = 0,
177 .end = IO_SPACE_LIMIT,
178 .name = "paravirt-ioport",
179 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
180 };
181
182 /*
183 * Reserve the whole legacy IO space to prevent any legacy drivers
184 * from wasting time probing for their hardware. This is a fairly
185 * brute-force approach to disabling all non-virtual drivers.
186 *
187 * Note that this must be called very early to have any effect.
188 */
paravirt_disable_iospace(void)189 int paravirt_disable_iospace(void)
190 {
191 return request_resource(&ioport_resource, &reserve_ioports);
192 }
193
194 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
195
enter_lazy(enum paravirt_lazy_mode mode)196 static inline void enter_lazy(enum paravirt_lazy_mode mode)
197 {
198 BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
199
200 this_cpu_write(paravirt_lazy_mode, mode);
201 }
202
leave_lazy(enum paravirt_lazy_mode mode)203 static void leave_lazy(enum paravirt_lazy_mode mode)
204 {
205 BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
206
207 this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
208 }
209
paravirt_enter_lazy_mmu(void)210 void paravirt_enter_lazy_mmu(void)
211 {
212 enter_lazy(PARAVIRT_LAZY_MMU);
213 }
214
paravirt_leave_lazy_mmu(void)215 void paravirt_leave_lazy_mmu(void)
216 {
217 leave_lazy(PARAVIRT_LAZY_MMU);
218 }
219
paravirt_flush_lazy_mmu(void)220 void paravirt_flush_lazy_mmu(void)
221 {
222 preempt_disable();
223
224 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
225 arch_leave_lazy_mmu_mode();
226 arch_enter_lazy_mmu_mode();
227 }
228
229 preempt_enable();
230 }
231
232 #ifdef CONFIG_PARAVIRT_XXL
paravirt_start_context_switch(struct task_struct * prev)233 void paravirt_start_context_switch(struct task_struct *prev)
234 {
235 BUG_ON(preemptible());
236
237 if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
238 arch_leave_lazy_mmu_mode();
239 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
240 }
241 enter_lazy(PARAVIRT_LAZY_CPU);
242 }
243
paravirt_end_context_switch(struct task_struct * next)244 void paravirt_end_context_switch(struct task_struct *next)
245 {
246 BUG_ON(preemptible());
247
248 leave_lazy(PARAVIRT_LAZY_CPU);
249
250 if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
251 arch_enter_lazy_mmu_mode();
252 }
253 #endif
254
paravirt_get_lazy_mode(void)255 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
256 {
257 if (in_interrupt())
258 return PARAVIRT_LAZY_NONE;
259
260 return this_cpu_read(paravirt_lazy_mode);
261 }
262
263 struct pv_info pv_info = {
264 .name = "bare hardware",
265 #ifdef CONFIG_PARAVIRT_XXL
266 .extra_user_64bit_cs = __USER_CS,
267 #endif
268 };
269
270 /* 64-bit pagetable entries */
271 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
272
273 struct paravirt_patch_template pv_ops = {
274 /* Init ops. */
275 .init.patch = native_patch,
276
277 /* Time ops. */
278 .time.sched_clock = native_sched_clock,
279 .time.steal_clock = native_steal_clock,
280
281 /* Cpu ops. */
282 .cpu.io_delay = native_io_delay,
283
284 #ifdef CONFIG_PARAVIRT_XXL
285 .cpu.cpuid = native_cpuid,
286 .cpu.get_debugreg = native_get_debugreg,
287 .cpu.set_debugreg = native_set_debugreg,
288 .cpu.read_cr0 = native_read_cr0,
289 .cpu.write_cr0 = native_write_cr0,
290 .cpu.write_cr4 = native_write_cr4,
291 .cpu.wbinvd = native_wbinvd,
292 .cpu.read_msr = native_read_msr,
293 .cpu.write_msr = native_write_msr,
294 .cpu.read_msr_safe = native_read_msr_safe,
295 .cpu.write_msr_safe = native_write_msr_safe,
296 .cpu.read_pmc = native_read_pmc,
297 .cpu.load_tr_desc = native_load_tr_desc,
298 .cpu.set_ldt = native_set_ldt,
299 .cpu.load_gdt = native_load_gdt,
300 .cpu.load_idt = native_load_idt,
301 .cpu.store_tr = native_store_tr,
302 .cpu.load_tls = native_load_tls,
303 .cpu.load_gs_index = native_load_gs_index,
304 .cpu.write_ldt_entry = native_write_ldt_entry,
305 .cpu.write_gdt_entry = native_write_gdt_entry,
306 .cpu.write_idt_entry = native_write_idt_entry,
307
308 .cpu.alloc_ldt = paravirt_nop,
309 .cpu.free_ldt = paravirt_nop,
310
311 .cpu.load_sp0 = native_load_sp0,
312
313 .cpu.usergs_sysret64 = native_usergs_sysret64,
314 .cpu.iret = native_iret,
315
316 #ifdef CONFIG_X86_IOPL_IOPERM
317 .cpu.invalidate_io_bitmap = native_tss_invalidate_io_bitmap,
318 .cpu.update_io_bitmap = native_tss_update_io_bitmap,
319 #endif
320
321 .cpu.start_context_switch = paravirt_nop,
322 .cpu.end_context_switch = paravirt_nop,
323
324 /* Irq ops. */
325 .irq.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
326 .irq.restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
327 .irq.irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
328 .irq.irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
329 .irq.safe_halt = native_safe_halt,
330 .irq.halt = native_halt,
331 #endif /* CONFIG_PARAVIRT_XXL */
332
333 /* Mmu ops. */
334 .mmu.flush_tlb_user = native_flush_tlb_local,
335 .mmu.flush_tlb_kernel = native_flush_tlb_global,
336 .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
337 .mmu.flush_tlb_others = native_flush_tlb_others,
338 .mmu.tlb_remove_table =
339 (void (*)(struct mmu_gather *, void *))tlb_remove_page,
340
341 .mmu.exit_mmap = paravirt_nop,
342
343 #ifdef CONFIG_PARAVIRT_XXL
344 .mmu.read_cr2 = __PV_IS_CALLEE_SAVE(native_read_cr2),
345 .mmu.write_cr2 = native_write_cr2,
346 .mmu.read_cr3 = __native_read_cr3,
347 .mmu.write_cr3 = native_write_cr3,
348
349 .mmu.pgd_alloc = __paravirt_pgd_alloc,
350 .mmu.pgd_free = paravirt_nop,
351
352 .mmu.alloc_pte = paravirt_nop,
353 .mmu.alloc_pmd = paravirt_nop,
354 .mmu.alloc_pud = paravirt_nop,
355 .mmu.alloc_p4d = paravirt_nop,
356 .mmu.release_pte = paravirt_nop,
357 .mmu.release_pmd = paravirt_nop,
358 .mmu.release_pud = paravirt_nop,
359 .mmu.release_p4d = paravirt_nop,
360
361 .mmu.set_pte = native_set_pte,
362 .mmu.set_pmd = native_set_pmd,
363
364 .mmu.ptep_modify_prot_start = __ptep_modify_prot_start,
365 .mmu.ptep_modify_prot_commit = __ptep_modify_prot_commit,
366
367 .mmu.set_pud = native_set_pud,
368
369 .mmu.pmd_val = PTE_IDENT,
370 .mmu.make_pmd = PTE_IDENT,
371
372 .mmu.pud_val = PTE_IDENT,
373 .mmu.make_pud = PTE_IDENT,
374
375 .mmu.set_p4d = native_set_p4d,
376
377 #if CONFIG_PGTABLE_LEVELS >= 5
378 .mmu.p4d_val = PTE_IDENT,
379 .mmu.make_p4d = PTE_IDENT,
380
381 .mmu.set_pgd = native_set_pgd,
382 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
383
384 .mmu.pte_val = PTE_IDENT,
385 .mmu.pgd_val = PTE_IDENT,
386
387 .mmu.make_pte = PTE_IDENT,
388 .mmu.make_pgd = PTE_IDENT,
389
390 .mmu.dup_mmap = paravirt_nop,
391 .mmu.activate_mm = paravirt_nop,
392
393 .mmu.lazy_mode = {
394 .enter = paravirt_nop,
395 .leave = paravirt_nop,
396 .flush = paravirt_nop,
397 },
398
399 .mmu.set_fixmap = native_set_fixmap,
400 #endif /* CONFIG_PARAVIRT_XXL */
401
402 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
403 /* Lock ops. */
404 #ifdef CONFIG_SMP
405 .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
406 .lock.queued_spin_unlock =
407 PV_CALLEE_SAVE(__native_queued_spin_unlock),
408 .lock.wait = paravirt_nop,
409 .lock.kick = paravirt_nop,
410 .lock.vcpu_is_preempted =
411 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
412 #endif /* SMP */
413 #endif
414 };
415
416 #ifdef CONFIG_PARAVIRT_XXL
417 /* At this point, native_get/set_debugreg has real function entries */
418 NOKPROBE_SYMBOL(native_get_debugreg);
419 NOKPROBE_SYMBOL(native_set_debugreg);
420 NOKPROBE_SYMBOL(native_load_idt);
421 #endif
422
423 EXPORT_SYMBOL(pv_ops);
424 EXPORT_SYMBOL_GPL(pv_info);
425