|
1 /* World's simplest hypervisor, to test paravirt_ops and show
2 * unbelievers that virtualization is the future. Plus, it's fun! */
3 #include <linux/module.h>
4 #include <linux/stringify.h>
5 #include <linux/stddef.h>
6 #include <linux/io.h>
7 #include <linux/mm.h>
8 #include <linux/vmalloc.h>
9 #include <linux/cpu.h>
10 #include <linux/freezer.h>
11 #include <asm/paravirt.h>
12 #include <asm/desc.h>
13 #include <asm/pgtable.h>
14 #include <asm/uaccess.h>
15 #include <asm/poll.h>
16 #include <asm/highmem.h>
17 #include <asm/asm-offsets.h>
18 #include <asm/i387.h>
19 #include "lg.h"
20
21 /* Found in switcher.S */
22 extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
23 extern unsigned long default_idt_entries[];
24
25 /* Every guest maps the core switcher code. */
26 #define SHARED_SWITCHER_PAGES \
27 DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
28 /* Pages for switcher itself, then two pages per cpu */
29 #define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)
30
31 /* We map at -4M for ease of mapping into the guest (one PTE page). */
32 #define SWITCHER_ADDR 0xFFC00000
33
//struct vm_struct {
/* keep next,addr,size together to speedup lookups */
// struct vm_struct *next;
// void *addr;
// unsigned long size;
// unsigned long flags;
// struct page **pages;
// unsigned int nr_pages;
// unsigned long phys_addr;
//};
34 static struct vm_struct *switcher_vma;
35 static struct page **switcher_page;
36
37 static int cpu_had_pge;
38 static struct {
39 unsigned long offset;
40 unsigned short segment;
41 } lguest_entry;
42
43 /* This One Big lock protects all inter-guest data structures. */
// mutex.h
// #define __MUTEX_INITIALIZER(lockname) \
// { .count = ATOMIC_INIT(1) \
// , .wait_lock = __SPIN_LOCK_UNLOCKED(lockname.wait_lock) \
// , .wait_list = LIST_HEAD_INIT(lockname.wait_list) \
// __DEBUG_MUTEX_INITIALIZER(lockname) \
// __DEP_MAP_MUTEX_INITIALIZER(lockname) }
//#define DEFINE_MUTEX(mutexname) \
// struct mutex mutexname = __MUTEX_INITIALIZER(mutexname)
44 DEFINE_MUTEX(lguest_lock);
45 static DEFINE_PER_CPU(struct lguest *, last_guest);
46
47 /* FIXME: Make dynamic. */
48 #define MAX_LGUEST_GUESTS 16
49 struct lguest lguests[MAX_LGUEST_GUESTS];
50
51 /* Offset from where switcher.S was compiled to where we've copied it */
/* SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE:存放struct lguest_pages 数据的起始地址*/
52 static unsigned long switcher_offset(void)
53 {
54 return SWITCHER_ADDR - (unsigned long)start_switcher_text;
55 }
56
57 /* This cpu's struct lguest_pages. */
58 static struct lguest_pages *lguest_pages(unsigned int cpu)
59 {
60 return &(((struct lguest_pages *)
61 (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
62 }
63
64 static __init int map_switcher(void)
65 {
66 int i, err;
67 struct page **pagep;
68
69 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
70 GFP_KERNEL);
71 if (!switcher_page) {
72 err = -ENOMEM;
73 goto out;
74 }
75
/* get_zeroed_page:返回一个指向新页的指针并且用零填充了该页. */
/* GFP_KERNEL，它的意思是该内存分配是由运行在内核模式的进程调用的，
* 即当内存低于min_free_pages的时候可以让该进程进入睡眠； */
76 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
77 unsigned long addr = get_zeroed_page(GFP_KERNEL);
78 if (!addr) {
79 err = -ENOMEM;
80 goto free_some_pages;
81 }
82 switcher_page[i] = virt_to_page(addr);
83 }
84
/**
* get_vm_area - reserve a contiguous kernel virtual area
* @size: size of the area
* @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
*
* Search an area of @size in the kernel virtual mapping area,
* and reserved it for out purposes. Returns the area descriptor
* on success or %NULL on failure.
*/
85 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
86 VM_ALLOC, SWITCHER_ADDR, VMALLOC_END);
87 if (!switcher_vma) {
88 err = -ENOMEM;
89 printk("lguest: could not map switcher pages high\n");
90 goto free_pages;
91 }
92
93 pagep = switcher_page;
/* map_vm_area建立内核使用的页表入口，标示已经分配的非连续内存空间的
* 每个页面已经和vmalloc分配的连续线性地址联系起来。
*/
94 err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
95 if (err) {
96 printk("lguest: map_vm_area failed: %i\n", err);
97 goto free_vma;
98 }
99 memcpy(switcher_vma->addr, start_switcher_text,
100 end_switcher_text - start_switcher_text);
101
102 /* Fix up IDT entries to point into copied text. */
103 for (i = 0; i < IDT_ENTRIES; i++)
104 default_idt_entries[i] += switcher_offset();
105
106 for_each_possible_cpu(i) {
107 struct lguest_pages *pages = lguest_pages(i);
108 struct lguest_ro_state *state = &pages->state;
109
110 /* These fields are static: rest done in copy_in_guest_info */
111 state->host_gdt_desc.size = GDT_SIZE-1;
112 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
113 store_idt(&state->host_idt_desc);
114 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
115 state->guest_idt_desc.address = (long)&state->guest_idt;
116 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
117 state->guest_gdt_desc.address = (long)&state->guest_gdt;
118 state->guest_tss.esp0 = (long)(&pages->regs + 1);
119 state->guest_tss.ss0 = LGUEST_DS;
120 /* No I/O for you! */
121 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
122 setup_default_gdt_entries(state);
123 setup_default_idt_entries(state, default_idt_entries);
124
125 /* Setup LGUEST segments on all cpus */
126 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
127 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
128 }
129
130 /* Initialize entry point into switcher. */
// switch_to_guest 在调整后的实际地址
131 lguest_entry.offset = (long)switch_to_guest + switcher_offset();
132 lguest_entry.segment = LGUEST_CS;
133
134 printk(KERN_INFO "lguest: mapped switcher at %p\n",
135 switcher_vma->addr);
136 return 0;
137
138 free_vma:
139 vunmap(switcher_vma->addr);
140 free_pages:
141 i = TOTAL_SWITCHER_PAGES;
142 free_some_pages:
143 for (--i; i >= 0; i--)
144 __free_pages(switcher_page[i], 0);
145 kfree(switcher_page);
146 out:
147 return err;
148 }
149
150 static void unmap_switcher(void)
151 {
152 unsigned int i;
153
154 vunmap(switcher_vma->addr);
155 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
156 __free_pages(switcher_page[i], 0);
157 }
158
159 /* IN/OUT insns: enough to get us past boot-time probing. */
160 static int emulate_insn(struct lguest *lg)
161 {
162 u8 insn;
163 unsigned int insnlen = 0, in = 0, shift = 0;
164 unsigned long physaddr = guest_pa(lg, lg->regs->eip);
165
166 /* This only works for addresses in linear mapping... */
167 if (lg->regs->eip < lg->page_offset)
168 return 0;
169 lgread(lg, &insn, physaddr, 1);
170
171 /* Operand size prefix means it's actually for ax. */
172 if (insn == 0x66) {
173 shift = 16;
174 insnlen = 1;
175 lgread(lg, &insn, physaddr + insnlen, 1);
176 }
177
178 switch (insn & 0xFE) {
179 case 0xE4: /* in ,%al */
180 insnlen += 2;
181 in = 1;
182 break;
183 case 0xEC: /* in (%dx),%al */
184 insnlen += 1;
185 in = 1;
186 break;
187 case 0xE6: /* out %al, */
188 insnlen += 2;
189 break;
190 case 0xEE: /* out %al,(%dx) */
191 insnlen += 1;
192 break;
193 default:
194 return 0;
195 }
196
197 if (in) {
198 /* Lower bit tells is whether it's a 16 or 32 bit access */
199 if (insn & 0x1)
200 lg->regs->eax = 0xFFFFFFFF;
201 else
202 lg->regs->eax |= (0xFFFF << shift);
203 }
204 lg->regs->eip += insnlen;
205 return 1;
206 }
207
208 int lguest_address_ok(const struct lguest *lg,
209 unsigned long addr, unsigned long len)
210 {
211 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
212 }
213
214 /* Just like get_user, but don't let guest access lguest binary. */
215 u32 lgread_u32(struct lguest *lg, unsigned long addr)
216 {
217 u32 val = 0;
218
219 /* Don't let them access lguest binary */
220 if (!lguest_address_ok(lg, addr, sizeof(val))
221 || get_user(val, (u32 __user *)addr) != 0)
222 kill_guest(lg, "bad read address %#lx", addr);
223 return val;
224 }
225
226 void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val)
227 {
228 if (!lguest_address_ok(lg, addr, sizeof(val))
229 || put_user(val, (u32 __user *)addr) != 0)
230 kill_guest(lg, "bad write address %#lx", addr);
231 }
232
233 vid lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
234 {
235 if (!lguest_address_ok(lg, addr, bytes)
236 || copy_from_user(b, (void __user *)addr, bytes) != 0) {
237 /* copy_from_user should do this, but as we rely on it... */
238 memset(b, 0, bytes);
239 kill_guest(lg, "bad read address %#lx len %u", addr, bytes);
240 }
241 }
242
243 void lgwrite(struct lguest *lg, unsigned long addr, const void *b,
244 unsigned bytes)
245 {
246 if (!lguest_address_ok(lg, addr, bytes)
247 || copy_to_user((void __user *)addr, b, bytes) != 0)
248 kill_guest(lg, "bad write address %#lx len %u", addr, bytes);
249 }
250
251 static void set_ts(void)
252 {
253 u32 cr0;
254
255 cr0 = read_cr0();
256 if (!(cr0 & 8))
257 write_cr0(cr0|8);
258 }
259
260static void copy_in_guest_info(struct lguest *lg,struct lguest_pages *pages)
261 {
262 if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
263 __get_cpu_var(last_guest) = lg;
264 lg->last_pages = pages;
265 lg->changed = CHANGED_ALL;
266 }
267
268 /* These are pretty cheap, so we do them unconditionally. */
/* __pa(x)代表取内核地址x的物理地址，内核地址范围是G-4G，因此__pa(x)不会溢出. */
269 pages->state.host_cr3 = __pa(current->mm->pgd);
270 map_switcher_in_guest(lg, pages); // 映射到最后4M
271 pages->state.guest_tss.esp1 = lg->esp1;
272 pages->state.guest_tss.ss1 = lg->ss1;
273
274 /* Copy direct trap entries. */
275 if (lg->changed & CHANGED_IDT)
276 copy_traps(lg, pages->state.guest_idt, default_idt_entries);
277
278 /* Copy all GDT entries but the TSS. */
279 if (lg->changed & CHANGED_GDT)
280 copy_gdt(lg, pages->state.guest_gdt);
281 /* If only the TLS entries have changed, copy them. */
282 else if (lg->changed & CHANGED_GDT_TLS)
283 copy_gdt_tls(lg, pages->state.guest_gdt);
284
285 lg->changed = 0;
286 }
287
288 static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
289 {
290 unsigned int clobber;
291
292 copy_in_guest_info(lg, pages);
293
294 /* Put eflags on stack, lcall does rest: suitable for iret return.*/
// 调用switch_to_guest
295 asm volatile("pushf; lcall *lguest_entry"
296 : "=a"(clobber), "=b"(clobber)
297 : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
298 : "memory", "%edx", "%ecx", "%edi", "%esi");
299 }
300
301 int run_guest(struct lguest *lg, unsigned long __user *user)
302 {
303 while (!lg->dead) {
304 unsigned int cr2 = 0; /* Damn gcc */
305
306 /* Hypercalls first: we might have been out to userspace */
307 do_hypercalls(lg);
308 if (lg->dma_is_pending) {
309 if (put_user(lg->pending_dma, user) ||
310 put_user(lg->pending_key, user+1))
311 return -EFAULT;
312 return sizeof(unsigned long)*2;
313 }
314
315 if (signal_pending(current))
316 return -ERESTARTSYS;
317
318 /* If Waker set break_out, return to Launcher. */
319 if (lg->break_out)
320 return -EAGAIN;
321
322 maybe_do_interrupt(lg);
323
324 try_to_freeze();
325
326 if (lg->dead)
327 break;
328
329 if (lg->halted) {
330 set_current_state(TASK_INTERRUPTIBLE);
331 schedule();
332 continue;
333 }
334
335 local_irq_disable();
336
337 /* Even if *we* don't want FPU trap, guest might... */
338 if (lg->ts)
339 set_ts();
340
341 /* Don't let Guest do SYSENTER: we can't handle it. */
342 if (boot_cpu_has(X86_FEATURE_SEP))
343 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
344
345 run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
346
347 /* Save cr2 now if we page-faulted. */
348 if (lg->regs->trapnum == 14)
349 cr2 = read_cr2();
350 else if (lg->regs->trapnum == 7)
351 math_state_restore();
352
353 if (boot_cpu_has(X86_FEATURE_SEP))
354 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
355 local_irq_enable();
356
357 switch (lg->regs->trapnum) {
358 case 13: /* We've intercepted a GPF. */
359 if (lg->regs->errcode == 0) {
360 if (emulate_insn(lg))
361 continue;
362 }
363 break;
364 case 14: /* We've intercepted a page fault. */
365 if (demand_page(lg, cr2, lg->regs->errcode))
366 continue;
367
368 /* If lguest_data is NULL, this won't hurt. */
369 if (put_user(cr2, &lg->lguest_data->cr2))
370 kill_guest(lg, "Writing cr2");
371 break;
372 case 7:/* We've intercepted a Device Not Available fault.*/
373 /* If they don't want to know, just absorb it. */
374 if (!lg->ts)
375 continue;
376 break;
377 case 32 ... 255: /* Real interrupt, fall thru */
378 cond_resched();
379 case LGUEST_TRAP_ENTRY: /* Handled at top of loop */
380 continue;
381 }
382
383 if (deliver_trap(lg, lg->regs->trapnum))
384 continue;
385
386 kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
387 lg->regs->trapnum, lg->regs->eip,
388 lg->regs->trapnum == 14 ? cr2 : lg->regs->errcode);
389 }
390 return -ENOENT;
391 }
392
393 int find_free_guest(void)
394 {
395 unsigned int i;
396 for (i = 0; i < MAX_LGUEST_GUESTS; i++)
397 if (!lguests[i].tsk)
398 return i;
399 return -1;
400 }
401
402 static void adjust_pge(void *on)
403 {
404 if (on)
405 write_cr4(read_cr4() | X86_CR4_PGE);
406 else
407 write_cr4(read_cr4() & ~X86_CR4_PGE);
408 }
409
410 static int __init init(void)
411 {
412 int err;
413
414 if (paravirt_enabled()) {
415 printk("lguest is afraid of %s\n", paravirt_ops.name);
416 return -EPERM;
417 }
418
419 err = map_switcher();
420 if (err)
421 return err;
422
423 err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
424 if (err) {
425 unmap_switcher();
426 return err;
427 }
428 lguest_io_init();
429
430 err = lguest_device_init();
431 if (err) {
432 free_pagetables();
433 unmap_switcher();
434 return err;
435 }
436 lock_cpu_hotplug();
437 if (cpu_has_pge) { /* We have a broader idea of "global". */
438 cpu_had_pge = 1;
439 on_each_cpu(adjust_pge, (void *)0, 0, 1);
440 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
441 }
442 unlock_cpu_hotplug();
443 return 0;
444 }
445
446 static void __exit fini(void)
447 {
448 lguest_device_remove();
449 free_pagetables();
450 unmap_switcher();
451 lock_cpu_hotplug();
452 if (cpu_had_pge) {
453 set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
454 on_each_cpu(adjust_pge, (void *)1, 0, 1);
455 }
456 unlock_cpu_hotplug();
457 }
458
/**
* module_init() - driver initialization entry point
* @x: function to be run at kernel boot time or module insertion
*
* module_init() will add the driver initialization routine in
* the "__initcall.int" code segment if the driver is checked as
* "y" or static, or else it will wrap the driver initialization
* routine with init_module() which is used by insmod and
* modprobe when the driver is used as a module.
*/
459 module_init(init);
/**
* module_exit() - driver exit entry point
* @x: function to be run when driver is removed
*
* module_exit() will wrap the driver clean-up code
* with cleanup_module() when used with rmmod when
* the driver is a module. If the driver is statically
* compiled into the kernel, module_exit() has no effect.
*/
460 module_exit(fini);
/*
* The following license idents are currently accepted as indicating free
* software modules
*
* "GPL" [GNU Public License v2 or later]
* "GPL v2" [GNU Public License v2]
* "GPL and additional rights" [GNU Public License v2 rights and more]
* "Dual BSD/GPL" [GNU Public License v2
* or BSD license choice]
* "Dual MIT/GPL" [GNU Public License v2
* or MIT license choice]
* "Dual MPL/GPL" [GNU Public License v2
* or Mozilla license choice]
*
* The following other idents are available
*
* "Proprietary" [Non free products]
*
* There are dual licensed components, but when running with Linux it is the
* GPL that is relevant so this is a non issue. Similarly LGPL linked with GPL
* is a GPL combined work.
*
* This exists for several reasons
* 1. So modinfo can show license info for users wanting to vet their setup
* is free
* 2. So the community can ignore bug reports including proprietary modules
* 3. So vendors can do likewise based on their own policies
*/
461 MODULE_LICENSE("GPL");
/* Author, ideally of form NAME[, NAME]*[ and NAME] */
462 MODULE_AUTHOR("Rusty Russell ");
|
