| 1 |
62 |
marcus.erl |
/*P:500 Just as userspace programs request kernel operations through a system
|
| 2 |
|
|
* call, the Guest requests Host operations through a "hypercall". You might
|
| 3 |
|
|
* notice this nomenclature doesn't really follow any logic, but the name has
|
| 4 |
|
|
* been around for long enough that we're stuck with it. As you'd expect, this
|
| 5 |
|
|
* code is basically a one big switch statement. :*/
|
| 6 |
|
|
|
| 7 |
|
|
/* Copyright (C) 2006 Rusty Russell IBM Corporation
|
| 8 |
|
|
|
| 9 |
|
|
This program is free software; you can redistribute it and/or modify
|
| 10 |
|
|
it under the terms of the GNU General Public License as published by
|
| 11 |
|
|
the Free Software Foundation; either version 2 of the License, or
|
| 12 |
|
|
(at your option) any later version.
|
| 13 |
|
|
|
| 14 |
|
|
This program is distributed in the hope that it will be useful,
|
| 15 |
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 16 |
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
| 17 |
|
|
GNU General Public License for more details.
|
| 18 |
|
|
|
| 19 |
|
|
You should have received a copy of the GNU General Public License
|
| 20 |
|
|
along with this program; if not, write to the Free Software
|
| 21 |
|
|
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
|
| 22 |
|
|
*/
|
| 23 |
|
|
#include <linux/uaccess.h>
|
| 24 |
|
|
#include <linux/syscalls.h>
|
| 25 |
|
|
#include <linux/mm.h>
|
| 26 |
|
|
#include <asm/page.h>
|
| 27 |
|
|
#include <asm/pgtable.h>
|
| 28 |
|
|
#include "lg.h"
|
| 29 |
|
|
|
| 30 |
|
|
/*H:120 This is the core hypercall routine: where the Guest gets what it wants.
|
| 31 |
|
|
* Or gets killed. Or, in the case of LHCALL_CRASH, both. */
|
| 32 |
|
|
static void do_hcall(struct lguest *lg, struct hcall_args *args)
|
| 33 |
|
|
{
|
| 34 |
|
|
switch (args->arg0) {
|
| 35 |
|
|
case LHCALL_FLUSH_ASYNC:
|
| 36 |
|
|
/* This call does nothing, except by breaking out of the Guest
|
| 37 |
|
|
* it makes us process all the asynchronous hypercalls. */
|
| 38 |
|
|
break;
|
| 39 |
|
|
case LHCALL_LGUEST_INIT:
|
| 40 |
|
|
/* You can't get here unless you're already initialized. Don't
|
| 41 |
|
|
* do that. */
|
| 42 |
|
|
kill_guest(lg, "already have lguest_data");
|
| 43 |
|
|
break;
|
| 44 |
|
|
case LHCALL_CRASH: {
|
| 45 |
|
|
/* Crash is such a trivial hypercall that we do it in four
|
| 46 |
|
|
* lines right here. */
|
| 47 |
|
|
char msg[128];
|
| 48 |
|
|
/* If the lgread fails, it will call kill_guest() itself; the
|
| 49 |
|
|
* kill_guest() with the message will be ignored. */
|
| 50 |
|
|
__lgread(lg, msg, args->arg1, sizeof(msg));
|
| 51 |
|
|
msg[sizeof(msg)-1] = '\0';
|
| 52 |
|
|
kill_guest(lg, "CRASH: %s", msg);
|
| 53 |
|
|
break;
|
| 54 |
|
|
}
|
| 55 |
|
|
case LHCALL_FLUSH_TLB:
|
| 56 |
|
|
/* FLUSH_TLB comes in two flavors, depending on the
|
| 57 |
|
|
* argument: */
|
| 58 |
|
|
if (args->arg1)
|
| 59 |
|
|
guest_pagetable_clear_all(lg);
|
| 60 |
|
|
else
|
| 61 |
|
|
guest_pagetable_flush_user(lg);
|
| 62 |
|
|
break;
|
| 63 |
|
|
|
| 64 |
|
|
/* All these calls simply pass the arguments through to the right
|
| 65 |
|
|
* routines. */
|
| 66 |
|
|
case LHCALL_NEW_PGTABLE:
|
| 67 |
|
|
guest_new_pagetable(lg, args->arg1);
|
| 68 |
|
|
break;
|
| 69 |
|
|
case LHCALL_SET_STACK:
|
| 70 |
|
|
guest_set_stack(lg, args->arg1, args->arg2, args->arg3);
|
| 71 |
|
|
break;
|
| 72 |
|
|
case LHCALL_SET_PTE:
|
| 73 |
|
|
guest_set_pte(lg, args->arg1, args->arg2, __pte(args->arg3));
|
| 74 |
|
|
break;
|
| 75 |
|
|
case LHCALL_SET_PMD:
|
| 76 |
|
|
guest_set_pmd(lg, args->arg1, args->arg2);
|
| 77 |
|
|
break;
|
| 78 |
|
|
case LHCALL_SET_CLOCKEVENT:
|
| 79 |
|
|
guest_set_clockevent(lg, args->arg1);
|
| 80 |
|
|
break;
|
| 81 |
|
|
case LHCALL_TS:
|
| 82 |
|
|
/* This sets the TS flag, as we saw used in run_guest(). */
|
| 83 |
|
|
lg->ts = args->arg1;
|
| 84 |
|
|
break;
|
| 85 |
|
|
case LHCALL_HALT:
|
| 86 |
|
|
/* Similarly, this sets the halted flag for run_guest(). */
|
| 87 |
|
|
lg->halted = 1;
|
| 88 |
|
|
break;
|
| 89 |
|
|
case LHCALL_NOTIFY:
|
| 90 |
|
|
lg->pending_notify = args->arg1;
|
| 91 |
|
|
break;
|
| 92 |
|
|
default:
|
| 93 |
|
|
/* It should be an architecture-specific hypercall. */
|
| 94 |
|
|
if (lguest_arch_do_hcall(lg, args))
|
| 95 |
|
|
kill_guest(lg, "Bad hypercall %li\n", args->arg0);
|
| 96 |
|
|
}
|
| 97 |
|
|
}
|
| 98 |
|
|
/*:*/
|
| 99 |
|
|
|
| 100 |
|
|
/*H:124 Asynchronous hypercalls are easy: we just look in the array in the
|
| 101 |
|
|
* Guest's "struct lguest_data" to see if any new ones are marked "ready".
|
| 102 |
|
|
*
|
| 103 |
|
|
* We are careful to do these in order: obviously we respect the order the
|
| 104 |
|
|
* Guest put them in the ring, but we also promise the Guest that they will
|
| 105 |
|
|
* happen before any normal hypercall (which is why we check this before
|
| 106 |
|
|
* checking for a normal hcall). */
|
| 107 |
|
|
static void do_async_hcalls(struct lguest *lg)
|
| 108 |
|
|
{
|
| 109 |
|
|
unsigned int i;
|
| 110 |
|
|
u8 st[LHCALL_RING_SIZE];
|
| 111 |
|
|
|
| 112 |
|
|
/* For simplicity, we copy the entire call status array in at once. */
|
| 113 |
|
|
if (copy_from_user(&st, &lg->lguest_data->hcall_status, sizeof(st)))
|
| 114 |
|
|
return;
|
| 115 |
|
|
|
| 116 |
|
|
/* We process "struct lguest_data"s hcalls[] ring once. */
|
| 117 |
|
|
for (i = 0; i < ARRAY_SIZE(st); i++) {
|
| 118 |
|
|
struct hcall_args args;
|
| 119 |
|
|
/* We remember where we were up to from last time. This makes
|
| 120 |
|
|
* sure that the hypercalls are done in the order the Guest
|
| 121 |
|
|
* places them in the ring. */
|
| 122 |
|
|
unsigned int n = lg->next_hcall;
|
| 123 |
|
|
|
| 124 |
|
|
/* 0xFF means there's no call here (yet). */
|
| 125 |
|
|
if (st[n] == 0xFF)
|
| 126 |
|
|
break;
|
| 127 |
|
|
|
| 128 |
|
|
/* OK, we have hypercall. Increment the "next_hcall" cursor,
|
| 129 |
|
|
* and wrap back to 0 if we reach the end. */
|
| 130 |
|
|
if (++lg->next_hcall == LHCALL_RING_SIZE)
|
| 131 |
|
|
lg->next_hcall = 0;
|
| 132 |
|
|
|
| 133 |
|
|
/* Copy the hypercall arguments into a local copy of
|
| 134 |
|
|
* the hcall_args struct. */
|
| 135 |
|
|
if (copy_from_user(&args, &lg->lguest_data->hcalls[n],
|
| 136 |
|
|
sizeof(struct hcall_args))) {
|
| 137 |
|
|
kill_guest(lg, "Fetching async hypercalls");
|
| 138 |
|
|
break;
|
| 139 |
|
|
}
|
| 140 |
|
|
|
| 141 |
|
|
/* Do the hypercall, same as a normal one. */
|
| 142 |
|
|
do_hcall(lg, &args);
|
| 143 |
|
|
|
| 144 |
|
|
/* Mark the hypercall done. */
|
| 145 |
|
|
if (put_user(0xFF, &lg->lguest_data->hcall_status[n])) {
|
| 146 |
|
|
kill_guest(lg, "Writing result for async hypercall");
|
| 147 |
|
|
break;
|
| 148 |
|
|
}
|
| 149 |
|
|
|
| 150 |
|
|
/* Stop doing hypercalls if they want to notify the Launcher:
|
| 151 |
|
|
* it needs to service this first. */
|
| 152 |
|
|
if (lg->pending_notify)
|
| 153 |
|
|
break;
|
| 154 |
|
|
}
|
| 155 |
|
|
}
|
| 156 |
|
|
|
| 157 |
|
|
/* Last of all, we look at what happens first of all. The very first time the
|
| 158 |
|
|
* Guest makes a hypercall, we end up here to set things up: */
|
| 159 |
|
|
static void initialize(struct lguest *lg)
|
| 160 |
|
|
{
|
| 161 |
|
|
/* You can't do anything until you're initialized. The Guest knows the
|
| 162 |
|
|
* rules, so we're unforgiving here. */
|
| 163 |
|
|
if (lg->hcall->arg0 != LHCALL_LGUEST_INIT) {
|
| 164 |
|
|
kill_guest(lg, "hypercall %li before INIT", lg->hcall->arg0);
|
| 165 |
|
|
return;
|
| 166 |
|
|
}
|
| 167 |
|
|
|
| 168 |
|
|
if (lguest_arch_init_hypercalls(lg))
|
| 169 |
|
|
kill_guest(lg, "bad guest page %p", lg->lguest_data);
|
| 170 |
|
|
|
| 171 |
|
|
/* The Guest tells us where we're not to deliver interrupts by putting
|
| 172 |
|
|
* the range of addresses into "struct lguest_data". */
|
| 173 |
|
|
if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start)
|
| 174 |
|
|
|| get_user(lg->noirq_end, &lg->lguest_data->noirq_end))
|
| 175 |
|
|
kill_guest(lg, "bad guest page %p", lg->lguest_data);
|
| 176 |
|
|
|
| 177 |
|
|
/* We write the current time into the Guest's data page once so it can
|
| 178 |
|
|
* set its clock. */
|
| 179 |
|
|
write_timestamp(lg);
|
| 180 |
|
|
|
| 181 |
|
|
/* page_tables.c will also do some setup. */
|
| 182 |
|
|
page_table_guest_data_init(lg);
|
| 183 |
|
|
|
| 184 |
|
|
/* This is the one case where the above accesses might have been the
|
| 185 |
|
|
* first write to a Guest page. This may have caused a copy-on-write
|
| 186 |
|
|
* fault, but the old page might be (read-only) in the Guest
|
| 187 |
|
|
* pagetable. */
|
| 188 |
|
|
guest_pagetable_clear_all(lg);
|
| 189 |
|
|
}
|
| 190 |
|
|
|
| 191 |
|
|
/*H:100
|
| 192 |
|
|
* Hypercalls
|
| 193 |
|
|
*
|
| 194 |
|
|
* Remember from the Guest, hypercalls come in two flavors: normal and
|
| 195 |
|
|
* asynchronous. This file handles both of types.
|
| 196 |
|
|
*/
|
| 197 |
|
|
void do_hypercalls(struct lguest *lg)
|
| 198 |
|
|
{
|
| 199 |
|
|
/* Not initialized yet? This hypercall must do it. */
|
| 200 |
|
|
if (unlikely(!lg->lguest_data)) {
|
| 201 |
|
|
/* Set up the "struct lguest_data" */
|
| 202 |
|
|
initialize(lg);
|
| 203 |
|
|
/* Hcall is done. */
|
| 204 |
|
|
lg->hcall = NULL;
|
| 205 |
|
|
return;
|
| 206 |
|
|
}
|
| 207 |
|
|
|
| 208 |
|
|
/* The Guest has initialized.
|
| 209 |
|
|
*
|
| 210 |
|
|
* Look in the hypercall ring for the async hypercalls: */
|
| 211 |
|
|
do_async_hcalls(lg);
|
| 212 |
|
|
|
| 213 |
|
|
/* If we stopped reading the hypercall ring because the Guest did a
|
| 214 |
|
|
* NOTIFY to the Launcher, we want to return now. Otherwise we do
|
| 215 |
|
|
* the hypercall. */
|
| 216 |
|
|
if (!lg->pending_notify) {
|
| 217 |
|
|
do_hcall(lg, lg->hcall);
|
| 218 |
|
|
/* Tricky point: we reset the hcall pointer to mark the
|
| 219 |
|
|
* hypercall as "done". We use the hcall pointer rather than
|
| 220 |
|
|
* the trap number to indicate a hypercall is pending.
|
| 221 |
|
|
* Normally it doesn't matter: the Guest will run again and
|
| 222 |
|
|
* update the trap number before we come back here.
|
| 223 |
|
|
*
|
| 224 |
|
|
* However, if we are signalled or the Guest sends I/O to the
|
| 225 |
|
|
* Launcher, the run_guest() loop will exit without running the
|
| 226 |
|
|
* Guest. When it comes back it would try to re-run the
|
| 227 |
|
|
* hypercall. */
|
| 228 |
|
|
lg->hcall = NULL;
|
| 229 |
|
|
}
|
| 230 |
|
|
}
|
| 231 |
|
|
|
| 232 |
|
|
/* This routine supplies the Guest with time: it's used for wallclock time at
|
| 233 |
|
|
* initial boot and as a rough time source if the TSC isn't available. */
|
| 234 |
|
|
void write_timestamp(struct lguest *lg)
|
| 235 |
|
|
{
|
| 236 |
|
|
struct timespec now;
|
| 237 |
|
|
ktime_get_real_ts(&now);
|
| 238 |
|
|
if (copy_to_user(&lg->lguest_data->time, &now, sizeof(struct timespec)))
|
| 239 |
|
|
kill_guest(lg, "Writing timestamp");
|
| 240 |
|
|
}
|