xref: /openbmc/linux/arch/x86/kernel/paravirt.c (revision 643d1f7f)
1 /*  Paravirtualization interfaces
2     Copyright (C) 2006 Rusty Russell IBM Corporation
3 
4     This program is free software; you can redistribute it and/or modify
5     it under the terms of the GNU General Public License as published by
6     the Free Software Foundation; either version 2 of the License, or
7     (at your option) any later version.
8 
9     This program is distributed in the hope that it will be useful,
10     but WITHOUT ANY WARRANTY; without even the implied warranty of
11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12     GNU General Public License for more details.
13 
14     You should have received a copy of the GNU General Public License
15     along with this program; if not, write to the Free Software
16     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17 
18     2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
19 */
20 
21 #include <linux/errno.h>
22 #include <linux/module.h>
23 #include <linux/efi.h>
24 #include <linux/bcd.h>
25 #include <linux/highmem.h>
26 
27 #include <asm/bug.h>
28 #include <asm/paravirt.h>
29 #include <asm/desc.h>
30 #include <asm/setup.h>
31 #include <asm/arch_hooks.h>
32 #include <asm/time.h>
33 #include <asm/irq.h>
34 #include <asm/delay.h>
35 #include <asm/fixmap.h>
36 #include <asm/apic.h>
37 #include <asm/tlbflush.h>
38 #include <asm/timer.h>
39 
40 /* nop stub */
41 void _paravirt_nop(void)
42 {
43 }
44 
45 static void __init default_banner(void)
46 {
47 	printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
48 	       pv_info.name);
49 }
50 
51 char *memory_setup(void)
52 {
53 	return pv_init_ops.memory_setup();
54 }
55 
56 /* Simple instruction patching code. */
57 #define DEF_NATIVE(ops, name, code)					\
58 	extern const char start_##ops##_##name[], end_##ops##_##name[];	\
59 	asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
60 
61 /* Undefined instruction for dealing with missing ops pointers. */
62 static const unsigned char ud2a[] = { 0x0f, 0x0b };
63 
64 unsigned paravirt_patch_nop(void)
65 {
66 	return 0;
67 }
68 
69 unsigned paravirt_patch_ignore(unsigned len)
70 {
71 	return len;
72 }
73 
74 struct branch {
75 	unsigned char opcode;
76 	u32 delta;
77 } __attribute__((packed));
78 
79 unsigned paravirt_patch_call(void *insnbuf,
80 			     const void *target, u16 tgt_clobbers,
81 			     unsigned long addr, u16 site_clobbers,
82 			     unsigned len)
83 {
84 	struct branch *b = insnbuf;
85 	unsigned long delta = (unsigned long)target - (addr+5);
86 
87 	if (tgt_clobbers & ~site_clobbers)
88 		return len;	/* target would clobber too much for this site */
89 	if (len < 5)
90 		return len;	/* call too long for patch site */
91 
92 	b->opcode = 0xe8; /* call */
93 	b->delta = delta;
94 	BUILD_BUG_ON(sizeof(*b) != 5);
95 
96 	return 5;
97 }
98 
99 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
100 			    unsigned long addr, unsigned len)
101 {
102 	struct branch *b = insnbuf;
103 	unsigned long delta = (unsigned long)target - (addr+5);
104 
105 	if (len < 5)
106 		return len;	/* call too long for patch site */
107 
108 	b->opcode = 0xe9;	/* jmp */
109 	b->delta = delta;
110 
111 	return 5;
112 }
113 
114 /* Neat trick to map patch type back to the call within the
115  * corresponding structure. */
116 static void *get_call_destination(u8 type)
117 {
118 	struct paravirt_patch_template tmpl = {
119 		.pv_init_ops = pv_init_ops,
120 		.pv_time_ops = pv_time_ops,
121 		.pv_cpu_ops = pv_cpu_ops,
122 		.pv_irq_ops = pv_irq_ops,
123 		.pv_apic_ops = pv_apic_ops,
124 		.pv_mmu_ops = pv_mmu_ops,
125 	};
126 	return *((void **)&tmpl + type);
127 }
128 
129 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
130 				unsigned long addr, unsigned len)
131 {
132 	void *opfunc = get_call_destination(type);
133 	unsigned ret;
134 
135 	if (opfunc == NULL)
136 		/* If there's no function, patch it with a ud2a (BUG) */
137 		ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
138 	else if (opfunc == paravirt_nop)
139 		/* If the operation is a nop, then nop the callsite */
140 		ret = paravirt_patch_nop();
141 	else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
142 		 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_syscall_ret))
143 		/* If operation requires a jmp, then jmp */
144 		ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
145 	else
146 		/* Otherwise call the function; assume target could
147 		   clobber any caller-save reg */
148 		ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
149 					  addr, clobbers, len);
150 
151 	return ret;
152 }
153 
154 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
155 			      const char *start, const char *end)
156 {
157 	unsigned insn_len = end - start;
158 
159 	if (insn_len > len || start == NULL)
160 		insn_len = len;
161 	else
162 		memcpy(insnbuf, start, insn_len);
163 
164 	return insn_len;
165 }
166 
167 void init_IRQ(void)
168 {
169 	pv_irq_ops.init_IRQ();
170 }
171 
172 static void native_flush_tlb(void)
173 {
174 	__native_flush_tlb();
175 }
176 
177 /*
178  * Global pages have to be flushed a bit differently. Not a real
179  * performance problem because this does not happen often.
180  */
181 static void native_flush_tlb_global(void)
182 {
183 	__native_flush_tlb_global();
184 }
185 
186 static void native_flush_tlb_single(unsigned long addr)
187 {
188 	__native_flush_tlb_single(addr);
189 }
190 
191 /* These are in entry.S */
192 extern void native_iret(void);
193 extern void native_irq_enable_syscall_ret(void);
194 
195 static int __init print_banner(void)
196 {
197 	pv_init_ops.banner();
198 	return 0;
199 }
200 core_initcall(print_banner);
201 
202 static struct resource reserve_ioports = {
203 	.start = 0,
204 	.end = IO_SPACE_LIMIT,
205 	.name = "paravirt-ioport",
206 	.flags = IORESOURCE_IO | IORESOURCE_BUSY,
207 };
208 
209 static struct resource reserve_iomem = {
210 	.start = 0,
211 	.end = -1,
212 	.name = "paravirt-iomem",
213 	.flags = IORESOURCE_MEM | IORESOURCE_BUSY,
214 };
215 
216 /*
217  * Reserve the whole legacy IO space to prevent any legacy drivers
218  * from wasting time probing for their hardware.  This is a fairly
219  * brute-force approach to disabling all non-virtual drivers.
220  *
221  * Note that this must be called very early to have any effect.
222  */
223 int paravirt_disable_iospace(void)
224 {
225 	int ret;
226 
227 	ret = request_resource(&ioport_resource, &reserve_ioports);
228 	if (ret == 0) {
229 		ret = request_resource(&iomem_resource, &reserve_iomem);
230 		if (ret)
231 			release_resource(&reserve_ioports);
232 	}
233 
234 	return ret;
235 }
236 
237 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
238 
239 static inline void enter_lazy(enum paravirt_lazy_mode mode)
240 {
241 	BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
242 	BUG_ON(preemptible());
243 
244 	__get_cpu_var(paravirt_lazy_mode) = mode;
245 }
246 
247 void paravirt_leave_lazy(enum paravirt_lazy_mode mode)
248 {
249 	BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode);
250 	BUG_ON(preemptible());
251 
252 	__get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
253 }
254 
255 void paravirt_enter_lazy_mmu(void)
256 {
257 	enter_lazy(PARAVIRT_LAZY_MMU);
258 }
259 
260 void paravirt_leave_lazy_mmu(void)
261 {
262 	paravirt_leave_lazy(PARAVIRT_LAZY_MMU);
263 }
264 
265 void paravirt_enter_lazy_cpu(void)
266 {
267 	enter_lazy(PARAVIRT_LAZY_CPU);
268 }
269 
270 void paravirt_leave_lazy_cpu(void)
271 {
272 	paravirt_leave_lazy(PARAVIRT_LAZY_CPU);
273 }
274 
275 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
276 {
277 	return __get_cpu_var(paravirt_lazy_mode);
278 }
279 
280 struct pv_info pv_info = {
281 	.name = "bare hardware",
282 	.paravirt_enabled = 0,
283 	.kernel_rpl = 0,
284 	.shared_kernel_pmd = 1,	/* Only used when CONFIG_X86_PAE is set */
285 };
286 
287 struct pv_init_ops pv_init_ops = {
288 	.patch = native_patch,
289 	.banner = default_banner,
290 	.arch_setup = paravirt_nop,
291 	.memory_setup = machine_specific_memory_setup,
292 };
293 
294 struct pv_time_ops pv_time_ops = {
295 	.time_init = hpet_time_init,
296 	.get_wallclock = native_get_wallclock,
297 	.set_wallclock = native_set_wallclock,
298 	.sched_clock = native_sched_clock,
299 	.get_cpu_khz = native_calculate_cpu_khz,
300 };
301 
302 struct pv_irq_ops pv_irq_ops = {
303 	.init_IRQ = native_init_IRQ,
304 	.save_fl = native_save_fl,
305 	.restore_fl = native_restore_fl,
306 	.irq_disable = native_irq_disable,
307 	.irq_enable = native_irq_enable,
308 	.safe_halt = native_safe_halt,
309 	.halt = native_halt,
310 };
311 
312 struct pv_cpu_ops pv_cpu_ops = {
313 	.cpuid = native_cpuid,
314 	.get_debugreg = native_get_debugreg,
315 	.set_debugreg = native_set_debugreg,
316 	.clts = native_clts,
317 	.read_cr0 = native_read_cr0,
318 	.write_cr0 = native_write_cr0,
319 	.read_cr4 = native_read_cr4,
320 	.read_cr4_safe = native_read_cr4_safe,
321 	.write_cr4 = native_write_cr4,
322 #ifdef CONFIG_X86_64
323 	.read_cr8 = native_read_cr8,
324 	.write_cr8 = native_write_cr8,
325 #endif
326 	.wbinvd = native_wbinvd,
327 	.read_msr = native_read_msr_safe,
328 	.write_msr = native_write_msr_safe,
329 	.read_tsc = native_read_tsc,
330 	.read_pmc = native_read_pmc,
331 	.read_tscp = native_read_tscp,
332 	.load_tr_desc = native_load_tr_desc,
333 	.set_ldt = native_set_ldt,
334 	.load_gdt = native_load_gdt,
335 	.load_idt = native_load_idt,
336 	.store_gdt = native_store_gdt,
337 	.store_idt = native_store_idt,
338 	.store_tr = native_store_tr,
339 	.load_tls = native_load_tls,
340 	.write_ldt_entry = native_write_ldt_entry,
341 	.write_gdt_entry = native_write_gdt_entry,
342 	.write_idt_entry = native_write_idt_entry,
343 	.load_sp0 = native_load_sp0,
344 
345 	.irq_enable_syscall_ret = native_irq_enable_syscall_ret,
346 	.iret = native_iret,
347 	.swapgs = native_swapgs,
348 
349 	.set_iopl_mask = native_set_iopl_mask,
350 	.io_delay = native_io_delay,
351 
352 	.lazy_mode = {
353 		.enter = paravirt_nop,
354 		.leave = paravirt_nop,
355 	},
356 };
357 
358 struct pv_apic_ops pv_apic_ops = {
359 #ifdef CONFIG_X86_LOCAL_APIC
360 	.apic_write = native_apic_write,
361 	.apic_write_atomic = native_apic_write_atomic,
362 	.apic_read = native_apic_read,
363 	.setup_boot_clock = setup_boot_APIC_clock,
364 	.setup_secondary_clock = setup_secondary_APIC_clock,
365 	.startup_ipi_hook = paravirt_nop,
366 #endif
367 };
368 
369 struct pv_mmu_ops pv_mmu_ops = {
370 #ifndef CONFIG_X86_64
371 	.pagetable_setup_start = native_pagetable_setup_start,
372 	.pagetable_setup_done = native_pagetable_setup_done,
373 #endif
374 
375 	.read_cr2 = native_read_cr2,
376 	.write_cr2 = native_write_cr2,
377 	.read_cr3 = native_read_cr3,
378 	.write_cr3 = native_write_cr3,
379 
380 	.flush_tlb_user = native_flush_tlb,
381 	.flush_tlb_kernel = native_flush_tlb_global,
382 	.flush_tlb_single = native_flush_tlb_single,
383 	.flush_tlb_others = native_flush_tlb_others,
384 
385 	.alloc_pt = paravirt_nop,
386 	.alloc_pd = paravirt_nop,
387 	.alloc_pd_clone = paravirt_nop,
388 	.release_pt = paravirt_nop,
389 	.release_pd = paravirt_nop,
390 
391 	.set_pte = native_set_pte,
392 	.set_pte_at = native_set_pte_at,
393 	.set_pmd = native_set_pmd,
394 	.pte_update = paravirt_nop,
395 	.pte_update_defer = paravirt_nop,
396 
397 #ifdef CONFIG_HIGHPTE
398 	.kmap_atomic_pte = kmap_atomic,
399 #endif
400 
401 #if PAGETABLE_LEVELS >= 3
402 #ifdef CONFIG_X86_PAE
403 	.set_pte_atomic = native_set_pte_atomic,
404 	.set_pte_present = native_set_pte_present,
405 	.pte_clear = native_pte_clear,
406 	.pmd_clear = native_pmd_clear,
407 #endif
408 	.set_pud = native_set_pud,
409 	.pmd_val = native_pmd_val,
410 	.make_pmd = native_make_pmd,
411 
412 #if PAGETABLE_LEVELS == 4
413 	.pud_val = native_pud_val,
414 	.make_pud = native_make_pud,
415 	.set_pgd = native_set_pgd,
416 #endif
417 #endif /* PAGETABLE_LEVELS >= 3 */
418 
419 	.pte_val = native_pte_val,
420 	.pgd_val = native_pgd_val,
421 
422 	.make_pte = native_make_pte,
423 	.make_pgd = native_make_pgd,
424 
425 	.dup_mmap = paravirt_nop,
426 	.exit_mmap = paravirt_nop,
427 	.activate_mm = paravirt_nop,
428 
429 	.lazy_mode = {
430 		.enter = paravirt_nop,
431 		.leave = paravirt_nop,
432 	},
433 };
434 
435 EXPORT_SYMBOL_GPL(pv_time_ops);
436 EXPORT_SYMBOL    (pv_cpu_ops);
437 EXPORT_SYMBOL    (pv_mmu_ops);
438 EXPORT_SYMBOL_GPL(pv_apic_ops);
439 EXPORT_SYMBOL_GPL(pv_info);
440 EXPORT_SYMBOL    (pv_irq_ops);
441