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 #include <asm/gsseg.h>
36
37 /*
38 * nop stub, which must not clobber anything *including the stack* to
39 * avoid confusing the entry prologues.
40 */
41 DEFINE_PARAVIRT_ASM(_paravirt_nop, "", .entry.text);
42
43 /* stub always returning 0. */
44 DEFINE_PARAVIRT_ASM(paravirt_ret0, "xor %eax,%eax", .entry.text);
45
default_banner(void)46 void __init default_banner(void)
47 {
48 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
49 pv_info.name);
50 }
51
52 /* Undefined instruction for dealing with missing ops pointers. */
paravirt_BUG(void)53 noinstr void paravirt_BUG(void)
54 {
55 BUG();
56 }
57
paravirt_patch_call(void * insn_buff,const void * target,unsigned long addr,unsigned len)58 static unsigned paravirt_patch_call(void *insn_buff, const void *target,
59 unsigned long addr, unsigned len)
60 {
61 __text_gen_insn(insn_buff, CALL_INSN_OPCODE,
62 (void *)addr, target, CALL_INSN_SIZE);
63 return CALL_INSN_SIZE;
64 }
65
66 #ifdef CONFIG_PARAVIRT_XXL
67 DEFINE_PARAVIRT_ASM(_paravirt_ident_64, "mov %rdi, %rax", .text);
68 DEFINE_PARAVIRT_ASM(pv_native_save_fl, "pushf; pop %rax", .noinstr.text);
69 DEFINE_PARAVIRT_ASM(pv_native_irq_disable, "cli", .noinstr.text);
70 DEFINE_PARAVIRT_ASM(pv_native_irq_enable, "sti", .noinstr.text);
71 DEFINE_PARAVIRT_ASM(pv_native_read_cr2, "mov %cr2, %rax", .noinstr.text);
72 #endif
73
74 DEFINE_STATIC_KEY_FALSE(virt_spin_lock_key);
75
native_pv_lock_init(void)76 void __init native_pv_lock_init(void)
77 {
78 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
79 static_branch_enable(&virt_spin_lock_key);
80 }
81
native_tlb_remove_table(struct mmu_gather * tlb,void * table)82 static void native_tlb_remove_table(struct mmu_gather *tlb, void *table)
83 {
84 tlb_remove_page(tlb, table);
85 }
86
paravirt_patch(u8 type,void * insn_buff,unsigned long addr,unsigned int len)87 unsigned int paravirt_patch(u8 type, void *insn_buff, unsigned long addr,
88 unsigned int len)
89 {
90 /*
91 * Neat trick to map patch type back to the call within the
92 * corresponding structure.
93 */
94 void *opfunc = *((void **)&pv_ops + type);
95 unsigned ret;
96
97 if (opfunc == NULL)
98 /* If there's no function, patch it with paravirt_BUG() */
99 ret = paravirt_patch_call(insn_buff, paravirt_BUG, addr, len);
100 else if (opfunc == _paravirt_nop)
101 ret = 0;
102 else
103 /* Otherwise call the function. */
104 ret = paravirt_patch_call(insn_buff, opfunc, addr, len);
105
106 return ret;
107 }
108
109 struct static_key paravirt_steal_enabled;
110 struct static_key paravirt_steal_rq_enabled;
111
native_steal_clock(int cpu)112 static u64 native_steal_clock(int cpu)
113 {
114 return 0;
115 }
116
117 DEFINE_STATIC_CALL(pv_steal_clock, native_steal_clock);
118 DEFINE_STATIC_CALL(pv_sched_clock, native_sched_clock);
119
paravirt_set_sched_clock(u64 (* func)(void))120 void paravirt_set_sched_clock(u64 (*func)(void))
121 {
122 static_call_update(pv_sched_clock, func);
123 }
124
125 /* These are in entry.S */
126 static struct resource reserve_ioports = {
127 .start = 0,
128 .end = IO_SPACE_LIMIT,
129 .name = "paravirt-ioport",
130 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
131 };
132
133 /*
134 * Reserve the whole legacy IO space to prevent any legacy drivers
135 * from wasting time probing for their hardware. This is a fairly
136 * brute-force approach to disabling all non-virtual drivers.
137 *
138 * Note that this must be called very early to have any effect.
139 */
paravirt_disable_iospace(void)140 int paravirt_disable_iospace(void)
141 {
142 return request_resource(&ioport_resource, &reserve_ioports);
143 }
144
145 #ifdef CONFIG_PARAVIRT_XXL
pv_native_write_cr2(unsigned long val)146 static noinstr void pv_native_write_cr2(unsigned long val)
147 {
148 native_write_cr2(val);
149 }
150
pv_native_get_debugreg(int regno)151 static noinstr unsigned long pv_native_get_debugreg(int regno)
152 {
153 return native_get_debugreg(regno);
154 }
155
pv_native_set_debugreg(int regno,unsigned long val)156 static noinstr void pv_native_set_debugreg(int regno, unsigned long val)
157 {
158 native_set_debugreg(regno, val);
159 }
160
pv_native_wbinvd(void)161 noinstr void pv_native_wbinvd(void)
162 {
163 native_wbinvd();
164 }
165
pv_native_safe_halt(void)166 static noinstr void pv_native_safe_halt(void)
167 {
168 native_safe_halt();
169 }
170 #endif
171
172 struct pv_info pv_info = {
173 .name = "bare hardware",
174 #ifdef CONFIG_PARAVIRT_XXL
175 .extra_user_64bit_cs = __USER_CS,
176 #endif
177 };
178
179 /* 64-bit pagetable entries */
180 #define PTE_IDENT __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
181
182 struct paravirt_patch_template pv_ops = {
183 /* Cpu ops. */
184 .cpu.io_delay = native_io_delay,
185
186 #ifdef CONFIG_PARAVIRT_XXL
187 .cpu.cpuid = native_cpuid,
188 .cpu.get_debugreg = pv_native_get_debugreg,
189 .cpu.set_debugreg = pv_native_set_debugreg,
190 .cpu.read_cr0 = native_read_cr0,
191 .cpu.write_cr0 = native_write_cr0,
192 .cpu.write_cr4 = native_write_cr4,
193 .cpu.wbinvd = pv_native_wbinvd,
194 .cpu.read_msr = native_read_msr,
195 .cpu.write_msr = native_write_msr,
196 .cpu.read_msr_safe = native_read_msr_safe,
197 .cpu.write_msr_safe = native_write_msr_safe,
198 .cpu.read_pmc = native_read_pmc,
199 .cpu.load_tr_desc = native_load_tr_desc,
200 .cpu.set_ldt = native_set_ldt,
201 .cpu.load_gdt = native_load_gdt,
202 .cpu.load_idt = native_load_idt,
203 .cpu.store_tr = native_store_tr,
204 .cpu.load_tls = native_load_tls,
205 .cpu.load_gs_index = native_load_gs_index,
206 .cpu.write_ldt_entry = native_write_ldt_entry,
207 .cpu.write_gdt_entry = native_write_gdt_entry,
208 .cpu.write_idt_entry = native_write_idt_entry,
209
210 .cpu.alloc_ldt = paravirt_nop,
211 .cpu.free_ldt = paravirt_nop,
212
213 .cpu.load_sp0 = native_load_sp0,
214
215 #ifdef CONFIG_X86_IOPL_IOPERM
216 .cpu.invalidate_io_bitmap = native_tss_invalidate_io_bitmap,
217 .cpu.update_io_bitmap = native_tss_update_io_bitmap,
218 #endif
219
220 .cpu.start_context_switch = paravirt_nop,
221 .cpu.end_context_switch = paravirt_nop,
222
223 /* Irq ops. */
224 .irq.save_fl = __PV_IS_CALLEE_SAVE(pv_native_save_fl),
225 .irq.irq_disable = __PV_IS_CALLEE_SAVE(pv_native_irq_disable),
226 .irq.irq_enable = __PV_IS_CALLEE_SAVE(pv_native_irq_enable),
227 .irq.safe_halt = pv_native_safe_halt,
228 .irq.halt = native_halt,
229 #endif /* CONFIG_PARAVIRT_XXL */
230
231 /* Mmu ops. */
232 .mmu.flush_tlb_user = native_flush_tlb_local,
233 .mmu.flush_tlb_kernel = native_flush_tlb_global,
234 .mmu.flush_tlb_one_user = native_flush_tlb_one_user,
235 .mmu.flush_tlb_multi = native_flush_tlb_multi,
236 .mmu.tlb_remove_table = native_tlb_remove_table,
237
238 .mmu.exit_mmap = paravirt_nop,
239 .mmu.notify_page_enc_status_changed = paravirt_nop,
240
241 #ifdef CONFIG_PARAVIRT_XXL
242 .mmu.read_cr2 = __PV_IS_CALLEE_SAVE(pv_native_read_cr2),
243 .mmu.write_cr2 = pv_native_write_cr2,
244 .mmu.read_cr3 = __native_read_cr3,
245 .mmu.write_cr3 = native_write_cr3,
246
247 .mmu.pgd_alloc = __paravirt_pgd_alloc,
248 .mmu.pgd_free = paravirt_nop,
249
250 .mmu.alloc_pte = paravirt_nop,
251 .mmu.alloc_pmd = paravirt_nop,
252 .mmu.alloc_pud = paravirt_nop,
253 .mmu.alloc_p4d = paravirt_nop,
254 .mmu.release_pte = paravirt_nop,
255 .mmu.release_pmd = paravirt_nop,
256 .mmu.release_pud = paravirt_nop,
257 .mmu.release_p4d = paravirt_nop,
258
259 .mmu.set_pte = native_set_pte,
260 .mmu.set_pmd = native_set_pmd,
261
262 .mmu.ptep_modify_prot_start = __ptep_modify_prot_start,
263 .mmu.ptep_modify_prot_commit = __ptep_modify_prot_commit,
264
265 .mmu.set_pud = native_set_pud,
266
267 .mmu.pmd_val = PTE_IDENT,
268 .mmu.make_pmd = PTE_IDENT,
269
270 .mmu.pud_val = PTE_IDENT,
271 .mmu.make_pud = PTE_IDENT,
272
273 .mmu.set_p4d = native_set_p4d,
274
275 #if CONFIG_PGTABLE_LEVELS >= 5
276 .mmu.p4d_val = PTE_IDENT,
277 .mmu.make_p4d = PTE_IDENT,
278
279 .mmu.set_pgd = native_set_pgd,
280 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
281
282 .mmu.pte_val = PTE_IDENT,
283 .mmu.pgd_val = PTE_IDENT,
284
285 .mmu.make_pte = PTE_IDENT,
286 .mmu.make_pgd = PTE_IDENT,
287
288 .mmu.enter_mmap = paravirt_nop,
289
290 .mmu.lazy_mode = {
291 .enter = paravirt_nop,
292 .leave = paravirt_nop,
293 .flush = paravirt_nop,
294 },
295
296 .mmu.set_fixmap = native_set_fixmap,
297 #endif /* CONFIG_PARAVIRT_XXL */
298
299 #if defined(CONFIG_PARAVIRT_SPINLOCKS)
300 /* Lock ops. */
301 #ifdef CONFIG_SMP
302 .lock.queued_spin_lock_slowpath = native_queued_spin_lock_slowpath,
303 .lock.queued_spin_unlock =
304 PV_CALLEE_SAVE(__native_queued_spin_unlock),
305 .lock.wait = paravirt_nop,
306 .lock.kick = paravirt_nop,
307 .lock.vcpu_is_preempted =
308 PV_CALLEE_SAVE(__native_vcpu_is_preempted),
309 #endif /* SMP */
310 #endif
311 };
312
313 #ifdef CONFIG_PARAVIRT_XXL
314 NOKPROBE_SYMBOL(native_load_idt);
315 #endif
316
317 EXPORT_SYMBOL(pv_ops);
318 EXPORT_SYMBOL_GPL(pv_info);
319