]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/lguest/core.c
lguest: Don't need comment terminator before disk section.
[mirror_ubuntu-artful-kernel.git] / drivers / lguest / core.c
CommitLineData
f938d2c8
RR
1/*P:400 This contains run_guest() which actually calls into the Host<->Guest
2 * Switcher and analyzes the return, such as determining if the Guest wants the
3 * Host to do something. This file also contains useful helper routines, and a
4 * couple of non-obvious setup and teardown pieces which were implemented after
5 * days of debugging pain. :*/
d7e28ffe
RR
6#include <linux/module.h>
7#include <linux/stringify.h>
8#include <linux/stddef.h>
9#include <linux/io.h>
10#include <linux/mm.h>
11#include <linux/vmalloc.h>
12#include <linux/cpu.h>
13#include <linux/freezer.h>
625efab1 14#include <linux/highmem.h>
d7e28ffe 15#include <asm/paravirt.h>
d7e28ffe
RR
16#include <asm/pgtable.h>
17#include <asm/uaccess.h>
18#include <asm/poll.h>
d7e28ffe 19#include <asm/asm-offsets.h>
d7e28ffe
RR
20#include "lg.h"
21
d7e28ffe
RR
22
23static struct vm_struct *switcher_vma;
24static struct page **switcher_page;
25
d7e28ffe
RR
26/* This One Big lock protects all inter-guest data structures. */
27DEFINE_MUTEX(lguest_lock);
d7e28ffe 28
bff672e6
RR
29/*H:010 We need to set up the Switcher at a high virtual address. Remember the
30 * Switcher is a few hundred bytes of assembler code which actually changes the
31 * CPU to run the Guest, and then changes back to the Host when a trap or
32 * interrupt happens.
33 *
34 * The Switcher code must be at the same virtual address in the Guest as the
35 * Host since it will be running as the switchover occurs.
36 *
37 * Trying to map memory at a particular address is an unusual thing to do, so
625efab1 38 * it's not a simple one-liner. */
d7e28ffe
RR
39static __init int map_switcher(void)
40{
41 int i, err;
42 struct page **pagep;
43
bff672e6
RR
44 /*
45 * Map the Switcher in to high memory.
46 *
47 * It turns out that if we choose the address 0xFFC00000 (4MB under the
48 * top virtual address), it makes setting up the page tables really
49 * easy.
50 */
51
52 /* We allocate an array of "struct page"s. map_vm_area() wants the
53 * pages in this form, rather than just an array of pointers. */
d7e28ffe
RR
54 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
55 GFP_KERNEL);
56 if (!switcher_page) {
57 err = -ENOMEM;
58 goto out;
59 }
60
bff672e6
RR
61 /* Now we actually allocate the pages. The Guest will see these pages,
62 * so we make sure they're zeroed. */
d7e28ffe
RR
63 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
64 unsigned long addr = get_zeroed_page(GFP_KERNEL);
65 if (!addr) {
66 err = -ENOMEM;
67 goto free_some_pages;
68 }
69 switcher_page[i] = virt_to_page(addr);
70 }
71
f14ae652
RR
72 /* First we check that the Switcher won't overlap the fixmap area at
73 * the top of memory. It's currently nowhere near, but it could have
74 * very strange effects if it ever happened. */
75 if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
76 err = -ENOMEM;
77 printk("lguest: mapping switcher would thwack fixmap\n");
78 goto free_pages;
79 }
80
bff672e6
RR
81 /* Now we reserve the "virtual memory area" we want: 0xFFC00000
82 * (SWITCHER_ADDR). We might not get it in theory, but in practice
f14ae652
RR
83 * it's worked so far. The end address needs +1 because __get_vm_area
84 * allocates an extra guard page, so we need space for that. */
d7e28ffe 85 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
f14ae652
RR
86 VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
87 + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
d7e28ffe
RR
88 if (!switcher_vma) {
89 err = -ENOMEM;
90 printk("lguest: could not map switcher pages high\n");
91 goto free_pages;
92 }
93
bff672e6
RR
94 /* This code actually sets up the pages we've allocated to appear at
95 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
96 * kind of pages we're mapping (kernel pages), and a pointer to our
97 * array of struct pages. It increments that pointer, but we don't
98 * care. */
d7e28ffe
RR
99 pagep = switcher_page;
100 err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
101 if (err) {
102 printk("lguest: map_vm_area failed: %i\n", err);
103 goto free_vma;
104 }
bff672e6 105
625efab1
JS
106 /* Now the Switcher is mapped at the right address, we can't fail!
107 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
d7e28ffe
RR
108 memcpy(switcher_vma->addr, start_switcher_text,
109 end_switcher_text - start_switcher_text);
110
d7e28ffe
RR
111 printk(KERN_INFO "lguest: mapped switcher at %p\n",
112 switcher_vma->addr);
bff672e6 113 /* And we succeeded... */
d7e28ffe
RR
114 return 0;
115
116free_vma:
117 vunmap(switcher_vma->addr);
118free_pages:
119 i = TOTAL_SWITCHER_PAGES;
120free_some_pages:
121 for (--i; i >= 0; i--)
122 __free_pages(switcher_page[i], 0);
123 kfree(switcher_page);
124out:
125 return err;
126}
bff672e6 127/*:*/
d7e28ffe 128
bff672e6
RR
129/* Cleaning up the mapping when the module is unloaded is almost...
130 * too easy. */
d7e28ffe
RR
131static void unmap_switcher(void)
132{
133 unsigned int i;
134
bff672e6 135 /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
d7e28ffe 136 vunmap(switcher_vma->addr);
bff672e6 137 /* Now we just need to free the pages we copied the switcher into */
d7e28ffe
RR
138 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
139 __free_pages(switcher_page[i], 0);
140}
141
e1e72965 142/*H:032
dde79789
RR
143 * Dealing With Guest Memory.
144 *
e1e72965
RR
145 * Before we go too much further into the Host, we need to grok the routines
146 * we use to deal with Guest memory.
147 *
dde79789 148 * When the Guest gives us (what it thinks is) a physical address, we can use
3c6b5bfa
RR
149 * the normal copy_from_user() & copy_to_user() on the corresponding place in
150 * the memory region allocated by the Launcher.
dde79789
RR
151 *
152 * But we can't trust the Guest: it might be trying to access the Launcher
153 * code. We have to check that the range is below the pfn_limit the Launcher
154 * gave us. We have to make sure that addr + len doesn't give us a false
155 * positive by overflowing, too. */
d7e28ffe
RR
156int lguest_address_ok(const struct lguest *lg,
157 unsigned long addr, unsigned long len)
158{
159 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
160}
161
2d37f94a
RR
162/* This routine copies memory from the Guest. Here we can see how useful the
163 * kill_lguest() routine we met in the Launcher can be: we return a random
164 * value (all zeroes) instead of needing to return an error. */
382ac6b3 165void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
d7e28ffe 166{
382ac6b3
GOC
167 if (!lguest_address_ok(cpu->lg, addr, bytes)
168 || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) {
d7e28ffe
RR
169 /* copy_from_user should do this, but as we rely on it... */
170 memset(b, 0, bytes);
382ac6b3 171 kill_guest(cpu, "bad read address %#lx len %u", addr, bytes);
d7e28ffe
RR
172 }
173}
174
2d37f94a 175/* This is the write (copy into guest) version. */
382ac6b3 176void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
2d37f94a 177 unsigned bytes)
d7e28ffe 178{
382ac6b3
GOC
179 if (!lguest_address_ok(cpu->lg, addr, bytes)
180 || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0)
181 kill_guest(cpu, "bad write address %#lx len %u", addr, bytes);
d7e28ffe 182}
2d37f94a 183/*:*/
d7e28ffe 184
bff672e6
RR
185/*H:030 Let's jump straight to the the main loop which runs the Guest.
186 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
187 * going around and around until something interesting happens. */
d0953d42 188int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
d7e28ffe 189{
bff672e6 190 /* We stop running once the Guest is dead. */
382ac6b3 191 while (!cpu->lg->dead) {
cc6d4fbc 192 /* First we run any hypercalls the Guest wants done. */
73044f05
GOC
193 if (cpu->hcall)
194 do_hypercalls(cpu);
cc6d4fbc 195
15045275 196 /* It's possible the Guest did a NOTIFY hypercall to the
bff672e6 197 * Launcher, in which case we return from the read() now. */
5e232f4f
GOC
198 if (cpu->pending_notify) {
199 if (put_user(cpu->pending_notify, user))
d7e28ffe 200 return -EFAULT;
5e232f4f 201 return sizeof(cpu->pending_notify);
d7e28ffe
RR
202 }
203
bff672e6 204 /* Check for signals */
d7e28ffe
RR
205 if (signal_pending(current))
206 return -ERESTARTSYS;
207
208 /* If Waker set break_out, return to Launcher. */
66686c2a 209 if (cpu->break_out)
d7e28ffe
RR
210 return -EAGAIN;
211
bff672e6
RR
212 /* Check if there are any interrupts which can be delivered
213 * now: if so, this sets up the hander to be executed when we
214 * next run the Guest. */
177e449d 215 maybe_do_interrupt(cpu);
d7e28ffe 216
bff672e6
RR
217 /* All long-lived kernel loops need to check with this horrible
218 * thing called the freezer. If the Host is trying to suspend,
219 * it stops us. */
d7e28ffe
RR
220 try_to_freeze();
221
bff672e6
RR
222 /* Just make absolutely sure the Guest is still alive. One of
223 * those hypercalls could have been fatal, for example. */
382ac6b3 224 if (cpu->lg->dead)
d7e28ffe
RR
225 break;
226
bff672e6
RR
227 /* If the Guest asked to be stopped, we sleep. The Guest's
228 * clock timer or LHCALL_BREAK from the Waker will wake us. */
66686c2a 229 if (cpu->halted) {
d7e28ffe
RR
230 set_current_state(TASK_INTERRUPTIBLE);
231 schedule();
232 continue;
233 }
234
bff672e6
RR
235 /* OK, now we're ready to jump into the Guest. First we put up
236 * the "Do Not Disturb" sign: */
d7e28ffe
RR
237 local_irq_disable();
238
625efab1 239 /* Actually run the Guest until something happens. */
d0953d42 240 lguest_arch_run_guest(cpu);
bff672e6
RR
241
242 /* Now we're ready to be interrupted or moved to other CPUs */
d7e28ffe
RR
243 local_irq_enable();
244
625efab1 245 /* Now we deal with whatever happened to the Guest. */
73044f05 246 lguest_arch_handle_trap(cpu);
d7e28ffe 247 }
625efab1 248
382ac6b3 249 if (cpu->lg->dead == ERR_PTR(-ERESTART))
ec04b13f 250 return -ERESTART;
bff672e6 251 /* The Guest is dead => "No such file or directory" */
d7e28ffe
RR
252 return -ENOENT;
253}
254
bff672e6
RR
255/*H:000
256 * Welcome to the Host!
257 *
258 * By this point your brain has been tickled by the Guest code and numbed by
259 * the Launcher code; prepare for it to be stretched by the Host code. This is
260 * the heart. Let's begin at the initialization routine for the Host's lg
261 * module.
262 */
d7e28ffe
RR
263static int __init init(void)
264{
265 int err;
266
bff672e6 267 /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
d7e28ffe 268 if (paravirt_enabled()) {
5c55841d 269 printk("lguest is afraid of being a guest\n");
d7e28ffe
RR
270 return -EPERM;
271 }
272
bff672e6 273 /* First we put the Switcher up in very high virtual memory. */
d7e28ffe
RR
274 err = map_switcher();
275 if (err)
c18acd73 276 goto out;
d7e28ffe 277
bff672e6 278 /* Now we set up the pagetable implementation for the Guests. */
d7e28ffe 279 err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
c18acd73
RR
280 if (err)
281 goto unmap;
bff672e6 282
c18acd73
RR
283 /* We might need to reserve an interrupt vector. */
284 err = init_interrupts();
285 if (err)
286 goto free_pgtables;
287
bff672e6 288 /* /dev/lguest needs to be registered. */
d7e28ffe 289 err = lguest_device_init();
c18acd73
RR
290 if (err)
291 goto free_interrupts;
bff672e6 292
625efab1
JS
293 /* Finally we do some architecture-specific setup. */
294 lguest_arch_host_init();
bff672e6
RR
295
296 /* All good! */
d7e28ffe 297 return 0;
c18acd73
RR
298
299free_interrupts:
300 free_interrupts();
301free_pgtables:
302 free_pagetables();
303unmap:
304 unmap_switcher();
305out:
306 return err;
d7e28ffe
RR
307}
308
bff672e6 309/* Cleaning up is just the same code, backwards. With a little French. */
d7e28ffe
RR
310static void __exit fini(void)
311{
312 lguest_device_remove();
c18acd73 313 free_interrupts();
d7e28ffe
RR
314 free_pagetables();
315 unmap_switcher();
bff672e6 316
625efab1 317 lguest_arch_host_fini();
d7e28ffe 318}
625efab1 319/*:*/
d7e28ffe 320
bff672e6
RR
321/* The Host side of lguest can be a module. This is a nice way for people to
322 * play with it. */
d7e28ffe
RR
323module_init(init);
324module_exit(fini);
325MODULE_LICENSE("GPL");
326MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");