blob: 83511eb0923d2f908a9e884d6470d493548e5cc0 [file] [log] [blame]
Rusty Russell2e04ef72009-07-30 16:03:45 -06001/*P:500
2 * Just as userspace programs request kernel operations through a system
Rusty Russellf938d2c2007-07-26 10:41:02 -07003 * call, the Guest requests Host operations through a "hypercall". You might
4 * notice this nomenclature doesn't really follow any logic, but the name has
5 * been around for long enough that we're stuck with it. As you'd expect, this
Rusty Russell2e04ef72009-07-30 16:03:45 -06006 * code is basically a one big switch statement.
7:*/
Rusty Russellf938d2c2007-07-26 10:41:02 -07008
9/* Copyright (C) 2006 Rusty Russell IBM Corporation
Rusty Russelld7e28ff2007-07-19 01:49:23 -070010
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2 of the License, or
14 (at your option) any later version.
15
16 This program is distributed in the hope that it will be useful,
17 but WITHOUT ANY WARRANTY; without even the implied warranty of
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 GNU General Public License for more details.
20
21 You should have received a copy of the GNU General Public License
22 along with this program; if not, write to the Free Software
23 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
24*/
25#include <linux/uaccess.h>
26#include <linux/syscalls.h>
27#include <linux/mm.h>
Glauber de Oliveira Costaca94f2b2008-01-18 23:59:07 -020028#include <linux/ktime.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070029#include <asm/page.h>
30#include <asm/pgtable.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070031#include "lg.h"
32
Rusty Russell2e04ef72009-07-30 16:03:45 -060033/*H:120
34 * This is the core hypercall routine: where the Guest gets what it wants.
35 * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both.
36 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -020037static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070038{
Jes Sorensenb410e7b2007-10-22 11:03:31 +100039 switch (args->arg0) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070040 case LHCALL_FLUSH_ASYNC:
Rusty Russell2e04ef72009-07-30 16:03:45 -060041 /*
42 * This call does nothing, except by breaking out of the Guest
43 * it makes us process all the asynchronous hypercalls.
44 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070045 break;
Rusty Russella32a8812009-06-12 22:27:02 -060046 case LHCALL_SEND_INTERRUPTS:
Rusty Russell2e04ef72009-07-30 16:03:45 -060047 /*
48 * This call does nothing too, but by breaking out of the Guest
49 * it makes us process any pending interrupts.
50 */
Rusty Russella32a8812009-06-12 22:27:02 -060051 break;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070052 case LHCALL_LGUEST_INIT:
Rusty Russell2e04ef72009-07-30 16:03:45 -060053 /*
54 * You can't get here unless you're already initialized. Don't
55 * do that.
56 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020057 kill_guest(cpu, "already have lguest_data");
Rusty Russelld7e28ff2007-07-19 01:49:23 -070058 break;
Balaji Raoec04b132007-12-28 14:26:24 +053059 case LHCALL_SHUTDOWN: {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070060 char msg[128];
Rusty Russell2e04ef72009-07-30 16:03:45 -060061 /*
Rusty Russella91d74a2009-07-30 16:03:45 -060062 * Shutdown is such a trivial hypercall that we do it in five
Rusty Russell2e04ef72009-07-30 16:03:45 -060063 * lines right here.
64 *
65 * If the lgread fails, it will call kill_guest() itself; the
66 * kill_guest() with the message will be ignored.
67 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020068 __lgread(cpu, msg, args->arg1, sizeof(msg));
Rusty Russelld7e28ff2007-07-19 01:49:23 -070069 msg[sizeof(msg)-1] = '\0';
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020070 kill_guest(cpu, "CRASH: %s", msg);
Balaji Raoec04b132007-12-28 14:26:24 +053071 if (args->arg2 == LGUEST_SHUTDOWN_RESTART)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020072 cpu->lg->dead = ERR_PTR(-ERESTART);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070073 break;
74 }
75 case LHCALL_FLUSH_TLB:
Rusty Russell2e04ef72009-07-30 16:03:45 -060076 /* FLUSH_TLB comes in two flavors, depending on the argument: */
Jes Sorensenb410e7b2007-10-22 11:03:31 +100077 if (args->arg1)
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020078 guest_pagetable_clear_all(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070079 else
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -020080 guest_pagetable_flush_user(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070081 break;
Rusty Russellbff672e2007-07-26 10:41:04 -070082
Rusty Russell2e04ef72009-07-30 16:03:45 -060083 /*
84 * All these calls simply pass the arguments through to the right
85 * routines.
86 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070087 case LHCALL_NEW_PGTABLE:
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020088 guest_new_pagetable(cpu, args->arg1);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070089 break;
90 case LHCALL_SET_STACK:
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020091 guest_set_stack(cpu, args->arg1, args->arg2, args->arg3);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070092 break;
93 case LHCALL_SET_PTE:
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -060094#ifdef CONFIG_X86_PAE
95 guest_set_pte(cpu, args->arg1, args->arg2,
96 __pte(args->arg3 | (u64)args->arg4 << 32));
97#else
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020098 guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3));
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -060099#endif
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700100 break;
Matias Zabaljaureguiebe0ba82009-05-30 15:48:08 -0300101 case LHCALL_SET_PGD:
102 guest_set_pgd(cpu->lg, args->arg1, args->arg2);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700103 break;
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600104#ifdef CONFIG_X86_PAE
105 case LHCALL_SET_PMD:
106 guest_set_pmd(cpu->lg, args->arg1, args->arg2);
107 break;
108#endif
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700109 case LHCALL_SET_CLOCKEVENT:
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200110 guest_set_clockevent(cpu, args->arg1);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700111 break;
112 case LHCALL_TS:
Rusty Russellbff672e2007-07-26 10:41:04 -0700113 /* This sets the TS flag, as we saw used in run_guest(). */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200114 cpu->ts = args->arg1;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700115 break;
116 case LHCALL_HALT:
Rusty Russellbff672e2007-07-26 10:41:04 -0700117 /* Similarly, this sets the halted flag for run_guest(). */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200118 cpu->halted = 1;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700119 break;
Rusty Russell15045272007-10-22 11:24:10 +1000120 case LHCALL_NOTIFY:
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200121 cpu->pending_notify = args->arg1;
Rusty Russell15045272007-10-22 11:24:10 +1000122 break;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700123 default:
Rusty Russelle1e72962007-10-25 15:02:50 +1000124 /* It should be an architecture-specific hypercall. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200125 if (lguest_arch_do_hcall(cpu, args))
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200126 kill_guest(cpu, "Bad hypercall %li\n", args->arg0);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700127 }
128}
129
Rusty Russell2e04ef72009-07-30 16:03:45 -0600130/*H:124
131 * Asynchronous hypercalls are easy: we just look in the array in the
Jes Sorensenb410e7b2007-10-22 11:03:31 +1000132 * Guest's "struct lguest_data" to see if any new ones are marked "ready".
Rusty Russellbff672e2007-07-26 10:41:04 -0700133 *
134 * We are careful to do these in order: obviously we respect the order the
135 * Guest put them in the ring, but we also promise the Guest that they will
136 * happen before any normal hypercall (which is why we check this before
Rusty Russell2e04ef72009-07-30 16:03:45 -0600137 * checking for a normal hcall).
138 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200139static void do_async_hcalls(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700140{
141 unsigned int i;
142 u8 st[LHCALL_RING_SIZE];
143
Rusty Russellbff672e2007-07-26 10:41:04 -0700144 /* For simplicity, we copy the entire call status array in at once. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200145 if (copy_from_user(&st, &cpu->lg->lguest_data->hcall_status, sizeof(st)))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700146 return;
147
Rusty Russellbff672e2007-07-26 10:41:04 -0700148 /* We process "struct lguest_data"s hcalls[] ring once. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700149 for (i = 0; i < ARRAY_SIZE(st); i++) {
Jes Sorensenb410e7b2007-10-22 11:03:31 +1000150 struct hcall_args args;
Rusty Russell2e04ef72009-07-30 16:03:45 -0600151 /*
152 * We remember where we were up to from last time. This makes
Rusty Russellbff672e2007-07-26 10:41:04 -0700153 * sure that the hypercalls are done in the order the Guest
Rusty Russell2e04ef72009-07-30 16:03:45 -0600154 * places them in the ring.
155 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200156 unsigned int n = cpu->next_hcall;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700157
Rusty Russellbff672e2007-07-26 10:41:04 -0700158 /* 0xFF means there's no call here (yet). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700159 if (st[n] == 0xFF)
160 break;
161
Rusty Russell2e04ef72009-07-30 16:03:45 -0600162 /*
163 * OK, we have hypercall. Increment the "next_hcall" cursor,
164 * and wrap back to 0 if we reach the end.
165 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200166 if (++cpu->next_hcall == LHCALL_RING_SIZE)
167 cpu->next_hcall = 0;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700168
Rusty Russell2e04ef72009-07-30 16:03:45 -0600169 /*
170 * Copy the hypercall arguments into a local copy of the
171 * hcall_args struct.
172 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200173 if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n],
Jes Sorensenb410e7b2007-10-22 11:03:31 +1000174 sizeof(struct hcall_args))) {
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200175 kill_guest(cpu, "Fetching async hypercalls");
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700176 break;
177 }
178
Rusty Russellbff672e2007-07-26 10:41:04 -0700179 /* Do the hypercall, same as a normal one. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200180 do_hcall(cpu, &args);
Rusty Russellbff672e2007-07-26 10:41:04 -0700181
182 /* Mark the hypercall done. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200183 if (put_user(0xFF, &cpu->lg->lguest_data->hcall_status[n])) {
184 kill_guest(cpu, "Writing result for async hypercall");
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700185 break;
186 }
187
Rusty Russell2e04ef72009-07-30 16:03:45 -0600188 /*
189 * Stop doing hypercalls if they want to notify the Launcher:
190 * it needs to service this first.
191 */
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200192 if (cpu->pending_notify)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700193 break;
194 }
195}
196
Rusty Russell2e04ef72009-07-30 16:03:45 -0600197/*
198 * Last of all, we look at what happens first of all. The very first time the
199 * Guest makes a hypercall, we end up here to set things up:
200 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200201static void initialize(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700202{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600203 /*
204 * You can't do anything until you're initialized. The Guest knows the
205 * rules, so we're unforgiving here.
206 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200207 if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) {
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200208 kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700209 return;
210 }
211
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200212 if (lguest_arch_init_hypercalls(cpu))
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200213 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000214
Rusty Russell2e04ef72009-07-30 16:03:45 -0600215 /*
216 * The Guest tells us where we're not to deliver interrupts by putting
217 * the range of addresses into "struct lguest_data".
218 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200219 if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start)
220 || get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end))
221 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700222
Rusty Russell2e04ef72009-07-30 16:03:45 -0600223 /*
224 * We write the current time into the Guest's data page once so it can
225 * set its clock.
226 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200227 write_timestamp(cpu);
Rusty Russell6c8dca52007-07-27 13:42:52 +1000228
Rusty Russell47436aa2007-10-22 11:03:36 +1000229 /* page_tables.c will also do some setup. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200230 page_table_guest_data_init(cpu);
Rusty Russell47436aa2007-10-22 11:03:36 +1000231
Rusty Russell2e04ef72009-07-30 16:03:45 -0600232 /*
233 * This is the one case where the above accesses might have been the
Rusty Russellbff672e2007-07-26 10:41:04 -0700234 * first write to a Guest page. This may have caused a copy-on-write
Rusty Russelle1e72962007-10-25 15:02:50 +1000235 * fault, but the old page might be (read-only) in the Guest
Rusty Russell2e04ef72009-07-30 16:03:45 -0600236 * pagetable.
237 */
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200238 guest_pagetable_clear_all(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700239}
Rusty Russella6bd8e12008-03-28 11:05:53 -0500240/*:*/
241
Rusty Russell2e04ef72009-07-30 16:03:45 -0600242/*M:013
243 * If a Guest reads from a page (so creates a mapping) that it has never
Rusty Russella6bd8e12008-03-28 11:05:53 -0500244 * written to, and then the Launcher writes to it (ie. the output of a virtual
245 * device), the Guest will still see the old page. In practice, this never
246 * happens: why would the Guest read a page which it has never written to? But
Rusty Russell2e04ef72009-07-30 16:03:45 -0600247 * a similar scenario might one day bite us, so it's worth mentioning.
Rusty Russella91d74a2009-07-30 16:03:45 -0600248 *
249 * Note that if we used a shared anonymous mapping in the Launcher instead of
250 * mapping /dev/zero private, we wouldn't worry about cop-on-write. And we
251 * need that to switch the Launcher to processes (away from threads) anyway.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600252:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700253
Rusty Russellbff672e2007-07-26 10:41:04 -0700254/*H:100
255 * Hypercalls
256 *
257 * Remember from the Guest, hypercalls come in two flavors: normal and
258 * asynchronous. This file handles both of types.
259 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200260void do_hypercalls(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700261{
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000262 /* Not initialized yet? This hypercall must do it. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200263 if (unlikely(!cpu->lg->lguest_data)) {
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000264 /* Set up the "struct lguest_data" */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200265 initialize(cpu);
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000266 /* Hcall is done. */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200267 cpu->hcall = NULL;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700268 return;
269 }
270
Rusty Russell2e04ef72009-07-30 16:03:45 -0600271 /*
272 * The Guest has initialized.
Rusty Russellbff672e2007-07-26 10:41:04 -0700273 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600274 * Look in the hypercall ring for the async hypercalls:
275 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200276 do_async_hcalls(cpu);
Rusty Russellbff672e2007-07-26 10:41:04 -0700277
Rusty Russell2e04ef72009-07-30 16:03:45 -0600278 /*
279 * If we stopped reading the hypercall ring because the Guest did a
Rusty Russell15045272007-10-22 11:24:10 +1000280 * NOTIFY to the Launcher, we want to return now. Otherwise we do
Rusty Russell2e04ef72009-07-30 16:03:45 -0600281 * the hypercall.
282 */
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -0200283 if (!cpu->pending_notify) {
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200284 do_hcall(cpu, cpu->hcall);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600285 /*
286 * Tricky point: we reset the hcall pointer to mark the
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000287 * hypercall as "done". We use the hcall pointer rather than
288 * the trap number to indicate a hypercall is pending.
289 * Normally it doesn't matter: the Guest will run again and
290 * update the trap number before we come back here.
291 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000292 * However, if we are signalled or the Guest sends I/O to the
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000293 * Launcher, the run_guest() loop will exit without running the
294 * Guest. When it comes back it would try to re-run the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600295 * hypercall. Finding that bug sucked.
296 */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200297 cpu->hcall = NULL;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700298 }
299}
Rusty Russell6c8dca52007-07-27 13:42:52 +1000300
Rusty Russell2e04ef72009-07-30 16:03:45 -0600301/*
302 * This routine supplies the Guest with time: it's used for wallclock time at
303 * initial boot and as a rough time source if the TSC isn't available.
304 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200305void write_timestamp(struct lg_cpu *cpu)
Rusty Russell6c8dca52007-07-27 13:42:52 +1000306{
307 struct timespec now;
308 ktime_get_real_ts(&now);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200309 if (copy_to_user(&cpu->lg->lguest_data->time,
310 &now, sizeof(struct timespec)))
311 kill_guest(cpu, "Writing timestamp");
Rusty Russell6c8dca52007-07-27 13:42:52 +1000312}