blob: 90663e01a56e0d4544b944da2fd81e5052a05a2f [file] [log] [blame]
Rusty Russellf938d2c2007-07-26 10:41:02 -07001/*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
Rusty Russella6bd8e12008-03-28 11:05:53 -05003 * Host to do something. This file also contains useful helper routines. :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -07004#include <linux/module.h>
5#include <linux/stringify.h>
6#include <linux/stddef.h>
7#include <linux/io.h>
8#include <linux/mm.h>
9#include <linux/vmalloc.h>
10#include <linux/cpu.h>
11#include <linux/freezer.h>
Jes Sorensen625efab2007-10-22 11:03:28 +100012#include <linux/highmem.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070013#include <asm/paravirt.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070014#include <asm/pgtable.h>
15#include <asm/uaccess.h>
16#include <asm/poll.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070017#include <asm/asm-offsets.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070018#include "lg.h"
19
Rusty Russelld7e28ff2007-07-19 01:49:23 -070020
21static struct vm_struct *switcher_vma;
22static struct page **switcher_page;
23
Rusty Russelld7e28ff2007-07-19 01:49:23 -070024/* This One Big lock protects all inter-guest data structures. */
25DEFINE_MUTEX(lguest_lock);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070026
Rusty Russellbff672e2007-07-26 10:41:04 -070027/*H:010 We need to set up the Switcher at a high virtual address. Remember the
28 * Switcher is a few hundred bytes of assembler code which actually changes the
29 * CPU to run the Guest, and then changes back to the Host when a trap or
30 * interrupt happens.
31 *
32 * The Switcher code must be at the same virtual address in the Guest as the
33 * Host since it will be running as the switchover occurs.
34 *
35 * Trying to map memory at a particular address is an unusual thing to do, so
Jes Sorensen625efab2007-10-22 11:03:28 +100036 * it's not a simple one-liner. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070037static __init int map_switcher(void)
38{
39 int i, err;
40 struct page **pagep;
41
Rusty Russellbff672e2007-07-26 10:41:04 -070042 /*
43 * Map the Switcher in to high memory.
44 *
45 * It turns out that if we choose the address 0xFFC00000 (4MB under the
46 * top virtual address), it makes setting up the page tables really
47 * easy.
48 */
49
Rusty Russella6bd8e12008-03-28 11:05:53 -050050 /* We allocate an array of struct page pointers. map_vm_area() wants
51 * this, rather than just an array of pages. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070052 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
53 GFP_KERNEL);
54 if (!switcher_page) {
55 err = -ENOMEM;
56 goto out;
57 }
58
Rusty Russellbff672e2007-07-26 10:41:04 -070059 /* Now we actually allocate the pages. The Guest will see these pages,
60 * so we make sure they're zeroed. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070061 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
62 unsigned long addr = get_zeroed_page(GFP_KERNEL);
63 if (!addr) {
64 err = -ENOMEM;
65 goto free_some_pages;
66 }
67 switcher_page[i] = virt_to_page(addr);
68 }
69
Rusty Russellf14ae652008-03-11 09:35:56 -050070 /* First we check that the Switcher won't overlap the fixmap area at
71 * the top of memory. It's currently nowhere near, but it could have
72 * very strange effects if it ever happened. */
73 if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
74 err = -ENOMEM;
75 printk("lguest: mapping switcher would thwack fixmap\n");
76 goto free_pages;
77 }
78
Rusty Russellbff672e2007-07-26 10:41:04 -070079 /* Now we reserve the "virtual memory area" we want: 0xFFC00000
80 * (SWITCHER_ADDR). We might not get it in theory, but in practice
Rusty Russellf14ae652008-03-11 09:35:56 -050081 * it's worked so far. The end address needs +1 because __get_vm_area
82 * allocates an extra guard page, so we need space for that. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070083 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
Rusty Russellf14ae652008-03-11 09:35:56 -050084 VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
85 + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070086 if (!switcher_vma) {
87 err = -ENOMEM;
88 printk("lguest: could not map switcher pages high\n");
89 goto free_pages;
90 }
91
Rusty Russellbff672e2007-07-26 10:41:04 -070092 /* This code actually sets up the pages we've allocated to appear at
93 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
94 * kind of pages we're mapping (kernel pages), and a pointer to our
95 * array of struct pages. It increments that pointer, but we don't
96 * care. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070097 pagep = switcher_page;
98 err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
99 if (err) {
100 printk("lguest: map_vm_area failed: %i\n", err);
101 goto free_vma;
102 }
Rusty Russellbff672e2007-07-26 10:41:04 -0700103
Jes Sorensen625efab2007-10-22 11:03:28 +1000104 /* Now the Switcher is mapped at the right address, we can't fail!
105 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700106 memcpy(switcher_vma->addr, start_switcher_text,
107 end_switcher_text - start_switcher_text);
108
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700109 printk(KERN_INFO "lguest: mapped switcher at %p\n",
110 switcher_vma->addr);
Rusty Russellbff672e2007-07-26 10:41:04 -0700111 /* And we succeeded... */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700112 return 0;
113
114free_vma:
115 vunmap(switcher_vma->addr);
116free_pages:
117 i = TOTAL_SWITCHER_PAGES;
118free_some_pages:
119 for (--i; i >= 0; i--)
120 __free_pages(switcher_page[i], 0);
121 kfree(switcher_page);
122out:
123 return err;
124}
Rusty Russellbff672e2007-07-26 10:41:04 -0700125/*:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700126
Rusty Russellbff672e2007-07-26 10:41:04 -0700127/* Cleaning up the mapping when the module is unloaded is almost...
128 * too easy. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700129static void unmap_switcher(void)
130{
131 unsigned int i;
132
Rusty Russellbff672e2007-07-26 10:41:04 -0700133 /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700134 vunmap(switcher_vma->addr);
Rusty Russellbff672e2007-07-26 10:41:04 -0700135 /* Now we just need to free the pages we copied the switcher into */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700136 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
137 __free_pages(switcher_page[i], 0);
Johannes Weiner0a707212008-07-08 10:29:42 +0200138 kfree(switcher_page);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700139}
140
Rusty Russelle1e72962007-10-25 15:02:50 +1000141/*H:032
Rusty Russelldde79782007-07-26 10:41:03 -0700142 * Dealing With Guest Memory.
143 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000144 * Before we go too much further into the Host, we need to grok the routines
145 * we use to deal with Guest memory.
146 *
Rusty Russelldde79782007-07-26 10:41:03 -0700147 * When the Guest gives us (what it thinks is) a physical address, we can use
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000148 * the normal copy_from_user() & copy_to_user() on the corresponding place in
149 * the memory region allocated by the Launcher.
Rusty Russelldde79782007-07-26 10:41:03 -0700150 *
151 * But we can't trust the Guest: it might be trying to access the Launcher
152 * code. We have to check that the range is below the pfn_limit the Launcher
153 * gave us. We have to make sure that addr + len doesn't give us a false
154 * positive by overflowing, too. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700155int lguest_address_ok(const struct lguest *lg,
156 unsigned long addr, unsigned long len)
157{
158 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
159}
160
Rusty Russell2d37f942007-10-22 11:24:24 +1000161/* This routine copies memory from the Guest. Here we can see how useful the
162 * kill_lguest() routine we met in the Launcher can be: we return a random
163 * value (all zeroes) instead of needing to return an error. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200164void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700165{
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200166 if (!lguest_address_ok(cpu->lg, addr, bytes)
167 || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700168 /* copy_from_user should do this, but as we rely on it... */
169 memset(b, 0, bytes);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200170 kill_guest(cpu, "bad read address %#lx len %u", addr, bytes);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700171 }
172}
173
Rusty Russella6bd8e12008-03-28 11:05:53 -0500174/* This is the write (copy into Guest) version. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200175void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
Rusty Russell2d37f942007-10-22 11:24:24 +1000176 unsigned bytes)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700177{
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200178 if (!lguest_address_ok(cpu->lg, addr, bytes)
179 || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0)
180 kill_guest(cpu, "bad write address %#lx len %u", addr, bytes);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700181}
Rusty Russell2d37f942007-10-22 11:24:24 +1000182/*:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700183
Rusty Russellbff672e2007-07-26 10:41:04 -0700184/*H:030 Let's jump straight to the the main loop which runs the Guest.
185 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
186 * going around and around until something interesting happens. */
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200187int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700188{
Rusty Russellbff672e2007-07-26 10:41:04 -0700189 /* We stop running once the Guest is dead. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200190 while (!cpu->lg->dead) {
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000191 /* First we run any hypercalls the Guest wants done. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200192 if (cpu->hcall)
193 do_hypercalls(cpu);
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000194
Rusty Russell15045272007-10-22 11:24:10 +1000195 /* It's possible the Guest did a NOTIFY hypercall to the
Rusty Russellbff672e2007-07-26 10:41:04 -0700196 * Launcher, in which case we return from the read() now. */
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200197 if (cpu->pending_notify) {
198 if (put_user(cpu->pending_notify, user))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700199 return -EFAULT;
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200200 return sizeof(cpu->pending_notify);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700201 }
202
Rusty Russellbff672e2007-07-26 10:41:04 -0700203 /* Check for signals */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700204 if (signal_pending(current))
205 return -ERESTARTSYS;
206
207 /* If Waker set break_out, return to Launcher. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200208 if (cpu->break_out)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700209 return -EAGAIN;
210
Rusty Russella6bd8e12008-03-28 11:05:53 -0500211 /* Check if there are any interrupts which can be delivered now:
212 * if so, this sets up the hander to be executed when we next
213 * run the Guest. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200214 maybe_do_interrupt(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700215
Rusty Russellbff672e2007-07-26 10:41:04 -0700216 /* All long-lived kernel loops need to check with this horrible
217 * thing called the freezer. If the Host is trying to suspend,
218 * it stops us. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700219 try_to_freeze();
220
Rusty Russellbff672e2007-07-26 10:41:04 -0700221 /* Just make absolutely sure the Guest is still alive. One of
222 * those hypercalls could have been fatal, for example. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200223 if (cpu->lg->dead)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700224 break;
225
Rusty Russellbff672e2007-07-26 10:41:04 -0700226 /* If the Guest asked to be stopped, we sleep. The Guest's
227 * clock timer or LHCALL_BREAK from the Waker will wake us. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200228 if (cpu->halted) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700229 set_current_state(TASK_INTERRUPTIBLE);
230 schedule();
231 continue;
232 }
233
Rusty Russellbff672e2007-07-26 10:41:04 -0700234 /* OK, now we're ready to jump into the Guest. First we put up
235 * the "Do Not Disturb" sign: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700236 local_irq_disable();
237
Jes Sorensen625efab2007-10-22 11:03:28 +1000238 /* Actually run the Guest until something happens. */
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200239 lguest_arch_run_guest(cpu);
Rusty Russellbff672e2007-07-26 10:41:04 -0700240
241 /* Now we're ready to be interrupted or moved to other CPUs */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700242 local_irq_enable();
243
Jes Sorensen625efab2007-10-22 11:03:28 +1000244 /* Now we deal with whatever happened to the Guest. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200245 lguest_arch_handle_trap(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700246 }
Jes Sorensen625efab2007-10-22 11:03:28 +1000247
Rusty Russella6bd8e12008-03-28 11:05:53 -0500248 /* Special case: Guest is 'dead' but wants a reboot. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200249 if (cpu->lg->dead == ERR_PTR(-ERESTART))
Balaji Raoec04b132007-12-28 14:26:24 +0530250 return -ERESTART;
Rusty Russella6bd8e12008-03-28 11:05:53 -0500251
Rusty Russellbff672e2007-07-26 10:41:04 -0700252 /* The Guest is dead => "No such file or directory" */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700253 return -ENOENT;
254}
255
Rusty Russellbff672e2007-07-26 10:41:04 -0700256/*H:000
257 * Welcome to the Host!
258 *
259 * By this point your brain has been tickled by the Guest code and numbed by
260 * the Launcher code; prepare for it to be stretched by the Host code. This is
261 * the heart. Let's begin at the initialization routine for the Host's lg
262 * module.
263 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700264static int __init init(void)
265{
266 int err;
267
Rusty Russellbff672e2007-07-26 10:41:04 -0700268 /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700269 if (paravirt_enabled()) {
Glauber de Oliveira Costa5c558412008-01-17 22:32:50 -0200270 printk("lguest is afraid of being a guest\n");
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700271 return -EPERM;
272 }
273
Rusty Russellbff672e2007-07-26 10:41:04 -0700274 /* First we put the Switcher up in very high virtual memory. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700275 err = map_switcher();
276 if (err)
Rusty Russellc18acd72007-10-22 11:03:35 +1000277 goto out;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700278
Rusty Russellbff672e2007-07-26 10:41:04 -0700279 /* Now we set up the pagetable implementation for the Guests. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700280 err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
Rusty Russellc18acd72007-10-22 11:03:35 +1000281 if (err)
282 goto unmap;
Rusty Russellbff672e2007-07-26 10:41:04 -0700283
Rusty Russellc18acd72007-10-22 11:03:35 +1000284 /* We might need to reserve an interrupt vector. */
285 err = init_interrupts();
286 if (err)
287 goto free_pgtables;
288
Rusty Russellbff672e2007-07-26 10:41:04 -0700289 /* /dev/lguest needs to be registered. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700290 err = lguest_device_init();
Rusty Russellc18acd72007-10-22 11:03:35 +1000291 if (err)
292 goto free_interrupts;
Rusty Russellbff672e2007-07-26 10:41:04 -0700293
Jes Sorensen625efab2007-10-22 11:03:28 +1000294 /* Finally we do some architecture-specific setup. */
295 lguest_arch_host_init();
Rusty Russellbff672e2007-07-26 10:41:04 -0700296
297 /* All good! */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700298 return 0;
Rusty Russellc18acd72007-10-22 11:03:35 +1000299
300free_interrupts:
301 free_interrupts();
302free_pgtables:
303 free_pagetables();
304unmap:
305 unmap_switcher();
306out:
307 return err;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700308}
309
Rusty Russellbff672e2007-07-26 10:41:04 -0700310/* Cleaning up is just the same code, backwards. With a little French. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700311static void __exit fini(void)
312{
313 lguest_device_remove();
Rusty Russellc18acd72007-10-22 11:03:35 +1000314 free_interrupts();
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700315 free_pagetables();
316 unmap_switcher();
Rusty Russellbff672e2007-07-26 10:41:04 -0700317
Jes Sorensen625efab2007-10-22 11:03:28 +1000318 lguest_arch_host_fini();
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700319}
Jes Sorensen625efab2007-10-22 11:03:28 +1000320/*:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700321
Rusty Russellbff672e2007-07-26 10:41:04 -0700322/* The Host side of lguest can be a module. This is a nice way for people to
323 * play with it. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700324module_init(init);
325module_exit(fini);
326MODULE_LICENSE("GPL");
327MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");