blob: 6023872e32d0472ffb8e59660be3985b46d9da8a [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
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. :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -07006#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>
Jes Sorensen625efab2007-10-22 11:03:28 +100014#include <linux/highmem.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070015#include <asm/paravirt.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070016#include <asm/pgtable.h>
17#include <asm/uaccess.h>
18#include <asm/poll.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070019#include <asm/asm-offsets.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070020#include "lg.h"
21
Rusty Russelld7e28ff2007-07-19 01:49:23 -070022
23static struct vm_struct *switcher_vma;
24static struct page **switcher_page;
25
Rusty Russelld7e28ff2007-07-19 01:49:23 -070026/* This One Big lock protects all inter-guest data structures. */
27DEFINE_MUTEX(lguest_lock);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070028
Rusty Russellbff672e2007-07-26 10:41:04 -070029/*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
Jes Sorensen625efab2007-10-22 11:03:28 +100038 * it's not a simple one-liner. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070039static __init int map_switcher(void)
40{
41 int i, err;
42 struct page **pagep;
43
Rusty Russellbff672e2007-07-26 10:41:04 -070044 /*
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. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070054 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
Rusty Russellbff672e2007-07-26 10:41:04 -070061 /* Now we actually allocate the pages. The Guest will see these pages,
62 * so we make sure they're zeroed. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070063 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
Rusty Russellbff672e2007-07-26 10:41:04 -070072 /* Now we reserve the "virtual memory area" we want: 0xFFC00000
73 * (SWITCHER_ADDR). We might not get it in theory, but in practice
74 * it's worked so far. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070075 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
76 VM_ALLOC, SWITCHER_ADDR, VMALLOC_END);
77 if (!switcher_vma) {
78 err = -ENOMEM;
79 printk("lguest: could not map switcher pages high\n");
80 goto free_pages;
81 }
82
Rusty Russellbff672e2007-07-26 10:41:04 -070083 /* This code actually sets up the pages we've allocated to appear at
84 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
85 * kind of pages we're mapping (kernel pages), and a pointer to our
86 * array of struct pages. It increments that pointer, but we don't
87 * care. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070088 pagep = switcher_page;
89 err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
90 if (err) {
91 printk("lguest: map_vm_area failed: %i\n", err);
92 goto free_vma;
93 }
Rusty Russellbff672e2007-07-26 10:41:04 -070094
Jes Sorensen625efab2007-10-22 11:03:28 +100095 /* Now the Switcher is mapped at the right address, we can't fail!
96 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070097 memcpy(switcher_vma->addr, start_switcher_text,
98 end_switcher_text - start_switcher_text);
99
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700100 printk(KERN_INFO "lguest: mapped switcher at %p\n",
101 switcher_vma->addr);
Rusty Russellbff672e2007-07-26 10:41:04 -0700102 /* And we succeeded... */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700103 return 0;
104
105free_vma:
106 vunmap(switcher_vma->addr);
107free_pages:
108 i = TOTAL_SWITCHER_PAGES;
109free_some_pages:
110 for (--i; i >= 0; i--)
111 __free_pages(switcher_page[i], 0);
112 kfree(switcher_page);
113out:
114 return err;
115}
Rusty Russellbff672e2007-07-26 10:41:04 -0700116/*:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700117
Rusty Russellbff672e2007-07-26 10:41:04 -0700118/* Cleaning up the mapping when the module is unloaded is almost...
119 * too easy. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700120static void unmap_switcher(void)
121{
122 unsigned int i;
123
Rusty Russellbff672e2007-07-26 10:41:04 -0700124 /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700125 vunmap(switcher_vma->addr);
Rusty Russellbff672e2007-07-26 10:41:04 -0700126 /* Now we just need to free the pages we copied the switcher into */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700127 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
128 __free_pages(switcher_page[i], 0);
129}
130
Rusty Russelle1e72962007-10-25 15:02:50 +1000131/*H:032
Rusty Russelldde79782007-07-26 10:41:03 -0700132 * Dealing With Guest Memory.
133 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000134 * Before we go too much further into the Host, we need to grok the routines
135 * we use to deal with Guest memory.
136 *
Rusty Russelldde79782007-07-26 10:41:03 -0700137 * When the Guest gives us (what it thinks is) a physical address, we can use
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000138 * the normal copy_from_user() & copy_to_user() on the corresponding place in
139 * the memory region allocated by the Launcher.
Rusty Russelldde79782007-07-26 10:41:03 -0700140 *
141 * But we can't trust the Guest: it might be trying to access the Launcher
142 * code. We have to check that the range is below the pfn_limit the Launcher
143 * gave us. We have to make sure that addr + len doesn't give us a false
144 * positive by overflowing, too. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700145int lguest_address_ok(const struct lguest *lg,
146 unsigned long addr, unsigned long len)
147{
148 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
149}
150
Rusty Russell2d37f942007-10-22 11:24:24 +1000151/* This routine copies memory from the Guest. Here we can see how useful the
152 * kill_lguest() routine we met in the Launcher can be: we return a random
153 * value (all zeroes) instead of needing to return an error. */
154void __lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700155{
156 if (!lguest_address_ok(lg, addr, bytes)
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000157 || copy_from_user(b, lg->mem_base + addr, bytes) != 0) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700158 /* copy_from_user should do this, but as we rely on it... */
159 memset(b, 0, bytes);
160 kill_guest(lg, "bad read address %#lx len %u", addr, bytes);
161 }
162}
163
Rusty Russell2d37f942007-10-22 11:24:24 +1000164/* This is the write (copy into guest) version. */
165void __lgwrite(struct lguest *lg, unsigned long addr, const void *b,
166 unsigned bytes)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700167{
168 if (!lguest_address_ok(lg, addr, bytes)
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000169 || copy_to_user(lg->mem_base + addr, b, bytes) != 0)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700170 kill_guest(lg, "bad write address %#lx len %u", addr, bytes);
171}
Rusty Russell2d37f942007-10-22 11:24:24 +1000172/*:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700173
Rusty Russellbff672e2007-07-26 10:41:04 -0700174/*H:030 Let's jump straight to the the main loop which runs the Guest.
175 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
176 * going around and around until something interesting happens. */
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200177int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700178{
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200179 struct lguest *lg = cpu->lg;
180
Rusty Russellbff672e2007-07-26 10:41:04 -0700181 /* We stop running once the Guest is dead. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700182 while (!lg->dead) {
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000183 /* First we run any hypercalls the Guest wants done. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200184 if (cpu->hcall)
185 do_hypercalls(cpu);
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000186
Rusty Russell15045272007-10-22 11:24:10 +1000187 /* It's possible the Guest did a NOTIFY hypercall to the
Rusty Russellbff672e2007-07-26 10:41:04 -0700188 * Launcher, in which case we return from the read() now. */
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200189 if (cpu->pending_notify) {
190 if (put_user(cpu->pending_notify, user))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700191 return -EFAULT;
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200192 return sizeof(cpu->pending_notify);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700193 }
194
Rusty Russellbff672e2007-07-26 10:41:04 -0700195 /* Check for signals */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700196 if (signal_pending(current))
197 return -ERESTARTSYS;
198
199 /* If Waker set break_out, return to Launcher. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200200 if (cpu->break_out)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700201 return -EAGAIN;
202
Rusty Russellbff672e2007-07-26 10:41:04 -0700203 /* Check if there are any interrupts which can be delivered
204 * now: if so, this sets up the hander to be executed when we
205 * next run the Guest. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200206 maybe_do_interrupt(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700207
Rusty Russellbff672e2007-07-26 10:41:04 -0700208 /* All long-lived kernel loops need to check with this horrible
209 * thing called the freezer. If the Host is trying to suspend,
210 * it stops us. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700211 try_to_freeze();
212
Rusty Russellbff672e2007-07-26 10:41:04 -0700213 /* Just make absolutely sure the Guest is still alive. One of
214 * those hypercalls could have been fatal, for example. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700215 if (lg->dead)
216 break;
217
Rusty Russellbff672e2007-07-26 10:41:04 -0700218 /* If the Guest asked to be stopped, we sleep. The Guest's
219 * clock timer or LHCALL_BREAK from the Waker will wake us. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200220 if (cpu->halted) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700221 set_current_state(TASK_INTERRUPTIBLE);
222 schedule();
223 continue;
224 }
225
Rusty Russellbff672e2007-07-26 10:41:04 -0700226 /* OK, now we're ready to jump into the Guest. First we put up
227 * the "Do Not Disturb" sign: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700228 local_irq_disable();
229
Jes Sorensen625efab2007-10-22 11:03:28 +1000230 /* Actually run the Guest until something happens. */
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200231 lguest_arch_run_guest(cpu);
Rusty Russellbff672e2007-07-26 10:41:04 -0700232
233 /* Now we're ready to be interrupted or moved to other CPUs */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700234 local_irq_enable();
235
Jes Sorensen625efab2007-10-22 11:03:28 +1000236 /* Now we deal with whatever happened to the Guest. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200237 lguest_arch_handle_trap(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700238 }
Jes Sorensen625efab2007-10-22 11:03:28 +1000239
Balaji Raoec04b132007-12-28 14:26:24 +0530240 if (lg->dead == ERR_PTR(-ERESTART))
241 return -ERESTART;
Rusty Russellbff672e2007-07-26 10:41:04 -0700242 /* The Guest is dead => "No such file or directory" */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700243 return -ENOENT;
244}
245
Rusty Russellbff672e2007-07-26 10:41:04 -0700246/*H:000
247 * Welcome to the Host!
248 *
249 * By this point your brain has been tickled by the Guest code and numbed by
250 * the Launcher code; prepare for it to be stretched by the Host code. This is
251 * the heart. Let's begin at the initialization routine for the Host's lg
252 * module.
253 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700254static int __init init(void)
255{
256 int err;
257
Rusty Russellbff672e2007-07-26 10:41:04 -0700258 /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700259 if (paravirt_enabled()) {
Glauber de Oliveira Costa5c558412008-01-17 22:32:50 -0200260 printk("lguest is afraid of being a guest\n");
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700261 return -EPERM;
262 }
263
Rusty Russellbff672e2007-07-26 10:41:04 -0700264 /* First we put the Switcher up in very high virtual memory. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700265 err = map_switcher();
266 if (err)
Rusty Russellc18acd72007-10-22 11:03:35 +1000267 goto out;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700268
Rusty Russellbff672e2007-07-26 10:41:04 -0700269 /* Now we set up the pagetable implementation for the Guests. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700270 err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
Rusty Russellc18acd72007-10-22 11:03:35 +1000271 if (err)
272 goto unmap;
Rusty Russellbff672e2007-07-26 10:41:04 -0700273
Rusty Russellc18acd72007-10-22 11:03:35 +1000274 /* We might need to reserve an interrupt vector. */
275 err = init_interrupts();
276 if (err)
277 goto free_pgtables;
278
Rusty Russellbff672e2007-07-26 10:41:04 -0700279 /* /dev/lguest needs to be registered. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700280 err = lguest_device_init();
Rusty Russellc18acd72007-10-22 11:03:35 +1000281 if (err)
282 goto free_interrupts;
Rusty Russellbff672e2007-07-26 10:41:04 -0700283
Jes Sorensen625efab2007-10-22 11:03:28 +1000284 /* Finally we do some architecture-specific setup. */
285 lguest_arch_host_init();
Rusty Russellbff672e2007-07-26 10:41:04 -0700286
287 /* All good! */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700288 return 0;
Rusty Russellc18acd72007-10-22 11:03:35 +1000289
290free_interrupts:
291 free_interrupts();
292free_pgtables:
293 free_pagetables();
294unmap:
295 unmap_switcher();
296out:
297 return err;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700298}
299
Rusty Russellbff672e2007-07-26 10:41:04 -0700300/* Cleaning up is just the same code, backwards. With a little French. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700301static void __exit fini(void)
302{
303 lguest_device_remove();
Rusty Russellc18acd72007-10-22 11:03:35 +1000304 free_interrupts();
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700305 free_pagetables();
306 unmap_switcher();
Rusty Russellbff672e2007-07-26 10:41:04 -0700307
Jes Sorensen625efab2007-10-22 11:03:28 +1000308 lguest_arch_host_fini();
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700309}
Jes Sorensen625efab2007-10-22 11:03:28 +1000310/*:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700311
Rusty Russellbff672e2007-07-26 10:41:04 -0700312/* The Host side of lguest can be a module. This is a nice way for people to
313 * play with it. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700314module_init(init);
315module_exit(fini);
316MODULE_LICENSE("GPL");
317MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");