1/*P:500
2 * Just as userspace programs request kernel operations through a system
3 * 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
6 * code is basically a one big switch statement.
7:*/
8
9/*  Copyright (C) 2006 Rusty Russell IBM Corporation
10
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>
28#include <linux/ktime.h>
29#include <asm/page.h>
30#include <asm/pgtable.h>
31#include "lg.h"
32
33/*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 */
37static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
38{
39	switch (args->arg0) {
40	case LHCALL_FLUSH_ASYNC:
41		/*
42		 * This call does nothing, except by breaking out of the Guest
43		 * it makes us process all the asynchronous hypercalls.
44		 */
45		break;
46	case LHCALL_SEND_INTERRUPTS:
47		/*
48		 * This call does nothing too, but by breaking out of the Guest
49		 * it makes us process any pending interrupts.
50		 */
51		break;
52	case LHCALL_LGUEST_INIT:
53		/*
54		 * You can't get here unless you're already initialized.  Don't
55		 * do that.
56		 */
57		kill_guest(cpu, "already have lguest_data");
58		break;
59	case LHCALL_SHUTDOWN: {
60		char msg[128];
61		/*
62		 * Shutdown is such a trivial hypercall that we do it in five
63		 * 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		 */
68		__lgread(cpu, msg, args->arg1, sizeof(msg));
69		msg[sizeof(msg)-1] = '\0';
70		kill_guest(cpu, "CRASH: %s", msg);
71		if (args->arg2 == LGUEST_SHUTDOWN_RESTART)
72			cpu->lg->dead = ERR_PTR(-ERESTART);
73		break;
74	}
75	case LHCALL_FLUSH_TLB:
76		/* FLUSH_TLB comes in two flavors, depending on the argument: */
77		if (args->arg1)
78			guest_pagetable_clear_all(cpu);
79		else
80			guest_pagetable_flush_user(cpu);
81		break;
82
83	/*
84	 * All these calls simply pass the arguments through to the right
85	 * routines.
86	 */
87	case LHCALL_NEW_PGTABLE:
88		guest_new_pagetable(cpu, args->arg1);
89		break;
90	case LHCALL_SET_STACK:
91		guest_set_stack(cpu, args->arg1, args->arg2, args->arg3);
92		break;
93	case LHCALL_SET_PTE:
94#ifdef CONFIG_X86_PAE
95		guest_set_pte(cpu, args->arg1, args->arg2,
96				__pte(args->arg3 | (u64)args->arg4 << 32));
97#else
98		guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3));
99#endif
100		break;
101	case LHCALL_SET_PGD:
102		guest_set_pgd(cpu->lg, args->arg1, args->arg2);
103		break;
104#ifdef CONFIG_X86_PAE
105	case LHCALL_SET_PMD:
106		guest_set_pmd(cpu->lg, args->arg1, args->arg2);
107		break;
108#endif
109	case LHCALL_SET_CLOCKEVENT:
110		guest_set_clockevent(cpu, args->arg1);
111		break;
112	case LHCALL_TS:
113		/* This sets the TS flag, as we saw used in run_guest(). */
114		cpu->ts = args->arg1;
115		break;
116	case LHCALL_HALT:
117		/* Similarly, this sets the halted flag for run_guest(). */
118		cpu->halted = 1;
119		break;
120	default:
121		/* It should be an architecture-specific hypercall. */
122		if (lguest_arch_do_hcall(cpu, args))
123			kill_guest(cpu, "Bad hypercall %li\n", args->arg0);
124	}
125}
126
127/*H:124
128 * Asynchronous hypercalls are easy: we just look in the array in the
129 * Guest's "struct lguest_data" to see if any new ones are marked "ready".
130 *
131 * We are careful to do these in order: obviously we respect the order the
132 * Guest put them in the ring, but we also promise the Guest that they will
133 * happen before any normal hypercall (which is why we check this before
134 * checking for a normal hcall).
135 */
136static void do_async_hcalls(struct lg_cpu *cpu)
137{
138	unsigned int i;
139	u8 st[LHCALL_RING_SIZE];
140
141	/* For simplicity, we copy the entire call status array in at once. */
142	if (copy_from_user(&st, &cpu->lg->lguest_data->hcall_status, sizeof(st)))
143		return;
144
145	/* We process "struct lguest_data"s hcalls[] ring once. */
146	for (i = 0; i < ARRAY_SIZE(st); i++) {
147		struct hcall_args args;
148		/*
149		 * We remember where we were up to from last time.  This makes
150		 * sure that the hypercalls are done in the order the Guest
151		 * places them in the ring.
152		 */
153		unsigned int n = cpu->next_hcall;
154
155		/* 0xFF means there's no call here (yet). */
156		if (st[n] == 0xFF)
157			break;
158
159		/*
160		 * OK, we have hypercall.  Increment the "next_hcall" cursor,
161		 * and wrap back to 0 if we reach the end.
162		 */
163		if (++cpu->next_hcall == LHCALL_RING_SIZE)
164			cpu->next_hcall = 0;
165
166		/*
167		 * Copy the hypercall arguments into a local copy of the
168		 * hcall_args struct.
169		 */
170		if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n],
171				   sizeof(struct hcall_args))) {
172			kill_guest(cpu, "Fetching async hypercalls");
173			break;
174		}
175
176		/* Do the hypercall, same as a normal one. */
177		do_hcall(cpu, &args);
178
179		/* Mark the hypercall done. */
180		if (put_user(0xFF, &cpu->lg->lguest_data->hcall_status[n])) {
181			kill_guest(cpu, "Writing result for async hypercall");
182			break;
183		}
184
185		/*
186		 * Stop doing hypercalls if they want to notify the Launcher:
187		 * it needs to service this first.
188		 */
189		if (cpu->pending.trap)
190			break;
191	}
192}
193
194/*
195 * Last of all, we look at what happens first of all.  The very first time the
196 * Guest makes a hypercall, we end up here to set things up:
197 */
198static void initialize(struct lg_cpu *cpu)
199{
200	/*
201	 * You can't do anything until you're initialized.  The Guest knows the
202	 * rules, so we're unforgiving here.
203	 */
204	if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) {
205		kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0);
206		return;
207	}
208
209	if (lguest_arch_init_hypercalls(cpu))
210		kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
211
212	/*
213	 * The Guest tells us where we're not to deliver interrupts by putting
214	 * the instruction address into "struct lguest_data".
215	 */
216	if (get_user(cpu->lg->noirq_iret, &cpu->lg->lguest_data->noirq_iret))
217		kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
218
219	/*
220	 * We write the current time into the Guest's data page once so it can
221	 * set its clock.
222	 */
223	write_timestamp(cpu);
224
225	/* page_tables.c will also do some setup. */
226	page_table_guest_data_init(cpu);
227
228	/*
229	 * This is the one case where the above accesses might have been the
230	 * first write to a Guest page.  This may have caused a copy-on-write
231	 * fault, but the old page might be (read-only) in the Guest
232	 * pagetable.
233	 */
234	guest_pagetable_clear_all(cpu);
235}
236/*:*/
237
238/*M:013
239 * If a Guest reads from a page (so creates a mapping) that it has never
240 * written to, and then the Launcher writes to it (ie. the output of a virtual
241 * device), the Guest will still see the old page.  In practice, this never
242 * happens: why would the Guest read a page which it has never written to?  But
243 * a similar scenario might one day bite us, so it's worth mentioning.
244 *
245 * Note that if we used a shared anonymous mapping in the Launcher instead of
246 * mapping /dev/zero private, we wouldn't worry about cop-on-write.  And we
247 * need that to switch the Launcher to processes (away from threads) anyway.
248:*/
249
250/*H:100
251 * Hypercalls
252 *
253 * Remember from the Guest, hypercalls come in two flavors: normal and
254 * asynchronous.  This file handles both of types.
255 */
256void do_hypercalls(struct lg_cpu *cpu)
257{
258	/* Not initialized yet?  This hypercall must do it. */
259	if (unlikely(!cpu->lg->lguest_data)) {
260		/* Set up the "struct lguest_data" */
261		initialize(cpu);
262		/* Hcall is done. */
263		cpu->hcall = NULL;
264		return;
265	}
266
267	/*
268	 * The Guest has initialized.
269	 *
270	 * Look in the hypercall ring for the async hypercalls:
271	 */
272	do_async_hcalls(cpu);
273
274	/*
275	 * If we stopped reading the hypercall ring because the Guest did a
276	 * NOTIFY to the Launcher, we want to return now.  Otherwise we do
277	 * the hypercall.
278	 */
279	if (!cpu->pending.trap) {
280		do_hcall(cpu, cpu->hcall);
281		/*
282		 * Tricky point: we reset the hcall pointer to mark the
283		 * hypercall as "done".  We use the hcall pointer rather than
284		 * the trap number to indicate a hypercall is pending.
285		 * Normally it doesn't matter: the Guest will run again and
286		 * update the trap number before we come back here.
287		 *
288		 * However, if we are signalled or the Guest sends I/O to the
289		 * Launcher, the run_guest() loop will exit without running the
290		 * Guest.  When it comes back it would try to re-run the
291		 * hypercall.  Finding that bug sucked.
292		 */
293		cpu->hcall = NULL;
294	}
295}
296
297/*
298 * This routine supplies the Guest with time: it's used for wallclock time at
299 * initial boot and as a rough time source if the TSC isn't available.
300 */
301void write_timestamp(struct lg_cpu *cpu)
302{
303	struct timespec now;
304	ktime_get_real_ts(&now);
305	if (copy_to_user(&cpu->lg->lguest_data->time,
306			 &now, sizeof(struct timespec)))
307		kill_guest(cpu, "Writing timestamp");
308}
309