]>
Commit | Line | Data |
---|---|---|
2e04ef76 RR |
1 | /*P:500 |
2 | * Just as userspace programs request kernel operations through a system | |
f938d2c8 RR |
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 | |
2e04ef76 RR |
6 | * code is basically a one big switch statement. |
7 | :*/ | |
f938d2c8 RR |
8 | |
9 | /* Copyright (C) 2006 Rusty Russell IBM Corporation | |
d7e28ffe RR |
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> | |
ca94f2bd | 28 | #include <linux/ktime.h> |
d7e28ffe RR |
29 | #include <asm/page.h> |
30 | #include <asm/pgtable.h> | |
d7e28ffe RR |
31 | #include "lg.h" |
32 | ||
2e04ef76 RR |
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 | */ | |
73044f05 | 37 | static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) |
d7e28ffe | 38 | { |
b410e7b1 | 39 | switch (args->arg0) { |
d7e28ffe | 40 | case LHCALL_FLUSH_ASYNC: |
2e04ef76 RR |
41 | /* |
42 | * This call does nothing, except by breaking out of the Guest | |
43 | * it makes us process all the asynchronous hypercalls. | |
44 | */ | |
d7e28ffe | 45 | break; |
a32a8813 | 46 | case LHCALL_SEND_INTERRUPTS: |
2e04ef76 RR |
47 | /* |
48 | * This call does nothing too, but by breaking out of the Guest | |
49 | * it makes us process any pending interrupts. | |
50 | */ | |
a32a8813 | 51 | break; |
d7e28ffe | 52 | case LHCALL_LGUEST_INIT: |
2e04ef76 RR |
53 | /* |
54 | * You can't get here unless you're already initialized. Don't | |
55 | * do that. | |
56 | */ | |
382ac6b3 | 57 | kill_guest(cpu, "already have lguest_data"); |
d7e28ffe | 58 | break; |
ec04b13f | 59 | case LHCALL_SHUTDOWN: { |
d7e28ffe | 60 | char msg[128]; |
2e04ef76 | 61 | /* |
a91d74a3 | 62 | * Shutdown is such a trivial hypercall that we do it in five |
2e04ef76 RR |
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 | */ | |
382ac6b3 | 68 | __lgread(cpu, msg, args->arg1, sizeof(msg)); |
d7e28ffe | 69 | msg[sizeof(msg)-1] = '\0'; |
382ac6b3 | 70 | kill_guest(cpu, "CRASH: %s", msg); |
ec04b13f | 71 | if (args->arg2 == LGUEST_SHUTDOWN_RESTART) |
382ac6b3 | 72 | cpu->lg->dead = ERR_PTR(-ERESTART); |
d7e28ffe RR |
73 | break; |
74 | } | |
75 | case LHCALL_FLUSH_TLB: | |
2e04ef76 | 76 | /* FLUSH_TLB comes in two flavors, depending on the argument: */ |
b410e7b1 | 77 | if (args->arg1) |
4665ac8e | 78 | guest_pagetable_clear_all(cpu); |
d7e28ffe | 79 | else |
1713608f | 80 | guest_pagetable_flush_user(cpu); |
d7e28ffe | 81 | break; |
bff672e6 | 82 | |
2e04ef76 RR |
83 | /* |
84 | * All these calls simply pass the arguments through to the right | |
85 | * routines. | |
86 | */ | |
d7e28ffe | 87 | case LHCALL_NEW_PGTABLE: |
4665ac8e | 88 | guest_new_pagetable(cpu, args->arg1); |
d7e28ffe RR |
89 | break; |
90 | case LHCALL_SET_STACK: | |
4665ac8e | 91 | guest_set_stack(cpu, args->arg1, args->arg2, args->arg3); |
d7e28ffe RR |
92 | break; |
93 | case LHCALL_SET_PTE: | |
acdd0b62 MZ |
94 | #ifdef CONFIG_X86_PAE |
95 | guest_set_pte(cpu, args->arg1, args->arg2, | |
96 | __pte(args->arg3 | (u64)args->arg4 << 32)); | |
97 | #else | |
382ac6b3 | 98 | guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3)); |
acdd0b62 | 99 | #endif |
d7e28ffe | 100 | break; |
ebe0ba84 MZ |
101 | case LHCALL_SET_PGD: |
102 | guest_set_pgd(cpu->lg, args->arg1, args->arg2); | |
d7e28ffe | 103 | break; |
acdd0b62 MZ |
104 | #ifdef CONFIG_X86_PAE |
105 | case LHCALL_SET_PMD: | |
106 | guest_set_pmd(cpu->lg, args->arg1, args->arg2); | |
107 | break; | |
108 | #endif | |
d7e28ffe | 109 | case LHCALL_SET_CLOCKEVENT: |
ad8d8f3b | 110 | guest_set_clockevent(cpu, args->arg1); |
d7e28ffe | 111 | break; |
d7e28ffe | 112 | case LHCALL_HALT: |
bff672e6 | 113 | /* Similarly, this sets the halted flag for run_guest(). */ |
66686c2a | 114 | cpu->halted = 1; |
d7e28ffe RR |
115 | break; |
116 | default: | |
e1e72965 | 117 | /* It should be an architecture-specific hypercall. */ |
73044f05 | 118 | if (lguest_arch_do_hcall(cpu, args)) |
382ac6b3 | 119 | kill_guest(cpu, "Bad hypercall %li\n", args->arg0); |
d7e28ffe RR |
120 | } |
121 | } | |
122 | ||
2e04ef76 RR |
123 | /*H:124 |
124 | * Asynchronous hypercalls are easy: we just look in the array in the | |
b410e7b1 | 125 | * Guest's "struct lguest_data" to see if any new ones are marked "ready". |
bff672e6 RR |
126 | * |
127 | * We are careful to do these in order: obviously we respect the order the | |
128 | * Guest put them in the ring, but we also promise the Guest that they will | |
129 | * happen before any normal hypercall (which is why we check this before | |
2e04ef76 RR |
130 | * checking for a normal hcall). |
131 | */ | |
73044f05 | 132 | static void do_async_hcalls(struct lg_cpu *cpu) |
d7e28ffe RR |
133 | { |
134 | unsigned int i; | |
135 | u8 st[LHCALL_RING_SIZE]; | |
136 | ||
bff672e6 | 137 | /* For simplicity, we copy the entire call status array in at once. */ |
382ac6b3 | 138 | if (copy_from_user(&st, &cpu->lg->lguest_data->hcall_status, sizeof(st))) |
d7e28ffe RR |
139 | return; |
140 | ||
bff672e6 | 141 | /* We process "struct lguest_data"s hcalls[] ring once. */ |
d7e28ffe | 142 | for (i = 0; i < ARRAY_SIZE(st); i++) { |
b410e7b1 | 143 | struct hcall_args args; |
2e04ef76 RR |
144 | /* |
145 | * We remember where we were up to from last time. This makes | |
bff672e6 | 146 | * sure that the hypercalls are done in the order the Guest |
2e04ef76 RR |
147 | * places them in the ring. |
148 | */ | |
73044f05 | 149 | unsigned int n = cpu->next_hcall; |
d7e28ffe | 150 | |
bff672e6 | 151 | /* 0xFF means there's no call here (yet). */ |
d7e28ffe RR |
152 | if (st[n] == 0xFF) |
153 | break; | |
154 | ||
2e04ef76 RR |
155 | /* |
156 | * OK, we have hypercall. Increment the "next_hcall" cursor, | |
157 | * and wrap back to 0 if we reach the end. | |
158 | */ | |
73044f05 GOC |
159 | if (++cpu->next_hcall == LHCALL_RING_SIZE) |
160 | cpu->next_hcall = 0; | |
d7e28ffe | 161 | |
2e04ef76 RR |
162 | /* |
163 | * Copy the hypercall arguments into a local copy of the | |
164 | * hcall_args struct. | |
165 | */ | |
382ac6b3 | 166 | if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], |
b410e7b1 | 167 | sizeof(struct hcall_args))) { |
382ac6b3 | 168 | kill_guest(cpu, "Fetching async hypercalls"); |
d7e28ffe RR |
169 | break; |
170 | } | |
171 | ||
bff672e6 | 172 | /* Do the hypercall, same as a normal one. */ |
73044f05 | 173 | do_hcall(cpu, &args); |
bff672e6 RR |
174 | |
175 | /* Mark the hypercall done. */ | |
382ac6b3 GOC |
176 | if (put_user(0xFF, &cpu->lg->lguest_data->hcall_status[n])) { |
177 | kill_guest(cpu, "Writing result for async hypercall"); | |
d7e28ffe RR |
178 | break; |
179 | } | |
180 | ||
2e04ef76 RR |
181 | /* |
182 | * Stop doing hypercalls if they want to notify the Launcher: | |
183 | * it needs to service this first. | |
184 | */ | |
69a09dc1 | 185 | if (cpu->pending.trap) |
d7e28ffe RR |
186 | break; |
187 | } | |
188 | } | |
189 | ||
2e04ef76 RR |
190 | /* |
191 | * Last of all, we look at what happens first of all. The very first time the | |
192 | * Guest makes a hypercall, we end up here to set things up: | |
193 | */ | |
73044f05 | 194 | static void initialize(struct lg_cpu *cpu) |
d7e28ffe | 195 | { |
2e04ef76 RR |
196 | /* |
197 | * You can't do anything until you're initialized. The Guest knows the | |
198 | * rules, so we're unforgiving here. | |
199 | */ | |
73044f05 | 200 | if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { |
382ac6b3 | 201 | kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); |
d7e28ffe RR |
202 | return; |
203 | } | |
204 | ||
73044f05 | 205 | if (lguest_arch_init_hypercalls(cpu)) |
382ac6b3 | 206 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); |
3c6b5bfa | 207 | |
2e04ef76 RR |
208 | /* |
209 | * The Guest tells us where we're not to deliver interrupts by putting | |
2f921b5b | 210 | * the instruction address into "struct lguest_data". |
2e04ef76 | 211 | */ |
2f921b5b | 212 | if (get_user(cpu->lg->noirq_iret, &cpu->lg->lguest_data->noirq_iret)) |
382ac6b3 | 213 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); |
d7e28ffe | 214 | |
2e04ef76 RR |
215 | /* |
216 | * We write the current time into the Guest's data page once so it can | |
217 | * set its clock. | |
218 | */ | |
382ac6b3 | 219 | write_timestamp(cpu); |
6c8dca5d | 220 | |
47436aa4 | 221 | /* page_tables.c will also do some setup. */ |
382ac6b3 | 222 | page_table_guest_data_init(cpu); |
47436aa4 | 223 | |
2e04ef76 RR |
224 | /* |
225 | * This is the one case where the above accesses might have been the | |
bff672e6 | 226 | * first write to a Guest page. This may have caused a copy-on-write |
e1e72965 | 227 | * fault, but the old page might be (read-only) in the Guest |
2e04ef76 RR |
228 | * pagetable. |
229 | */ | |
4665ac8e | 230 | guest_pagetable_clear_all(cpu); |
d7e28ffe | 231 | } |
a6bd8e13 RR |
232 | /*:*/ |
233 | ||
2e04ef76 RR |
234 | /*M:013 |
235 | * If a Guest reads from a page (so creates a mapping) that it has never | |
a6bd8e13 RR |
236 | * written to, and then the Launcher writes to it (ie. the output of a virtual |
237 | * device), the Guest will still see the old page. In practice, this never | |
238 | * happens: why would the Guest read a page which it has never written to? But | |
2e04ef76 | 239 | * a similar scenario might one day bite us, so it's worth mentioning. |
a91d74a3 RR |
240 | * |
241 | * Note that if we used a shared anonymous mapping in the Launcher instead of | |
242 | * mapping /dev/zero private, we wouldn't worry about cop-on-write. And we | |
243 | * need that to switch the Launcher to processes (away from threads) anyway. | |
2e04ef76 | 244 | :*/ |
d7e28ffe | 245 | |
bff672e6 RR |
246 | /*H:100 |
247 | * Hypercalls | |
248 | * | |
249 | * Remember from the Guest, hypercalls come in two flavors: normal and | |
250 | * asynchronous. This file handles both of types. | |
251 | */ | |
73044f05 | 252 | void do_hypercalls(struct lg_cpu *cpu) |
d7e28ffe | 253 | { |
cc6d4fbc | 254 | /* Not initialized yet? This hypercall must do it. */ |
73044f05 | 255 | if (unlikely(!cpu->lg->lguest_data)) { |
cc6d4fbc | 256 | /* Set up the "struct lguest_data" */ |
73044f05 | 257 | initialize(cpu); |
cc6d4fbc | 258 | /* Hcall is done. */ |
73044f05 | 259 | cpu->hcall = NULL; |
d7e28ffe RR |
260 | return; |
261 | } | |
262 | ||
2e04ef76 RR |
263 | /* |
264 | * The Guest has initialized. | |
bff672e6 | 265 | * |
2e04ef76 RR |
266 | * Look in the hypercall ring for the async hypercalls: |
267 | */ | |
73044f05 | 268 | do_async_hcalls(cpu); |
bff672e6 | 269 | |
2e04ef76 RR |
270 | /* |
271 | * If we stopped reading the hypercall ring because the Guest did a | |
15045275 | 272 | * NOTIFY to the Launcher, we want to return now. Otherwise we do |
2e04ef76 RR |
273 | * the hypercall. |
274 | */ | |
69a09dc1 | 275 | if (!cpu->pending.trap) { |
73044f05 | 276 | do_hcall(cpu, cpu->hcall); |
2e04ef76 RR |
277 | /* |
278 | * Tricky point: we reset the hcall pointer to mark the | |
cc6d4fbc RR |
279 | * hypercall as "done". We use the hcall pointer rather than |
280 | * the trap number to indicate a hypercall is pending. | |
281 | * Normally it doesn't matter: the Guest will run again and | |
282 | * update the trap number before we come back here. | |
283 | * | |
e1e72965 | 284 | * However, if we are signalled or the Guest sends I/O to the |
cc6d4fbc RR |
285 | * Launcher, the run_guest() loop will exit without running the |
286 | * Guest. When it comes back it would try to re-run the | |
2e04ef76 RR |
287 | * hypercall. Finding that bug sucked. |
288 | */ | |
73044f05 | 289 | cpu->hcall = NULL; |
d7e28ffe RR |
290 | } |
291 | } | |
6c8dca5d | 292 | |
2e04ef76 RR |
293 | /* |
294 | * This routine supplies the Guest with time: it's used for wallclock time at | |
295 | * initial boot and as a rough time source if the TSC isn't available. | |
296 | */ | |
382ac6b3 | 297 | void write_timestamp(struct lg_cpu *cpu) |
6c8dca5d RR |
298 | { |
299 | struct timespec now; | |
300 | ktime_get_real_ts(&now); | |
382ac6b3 GOC |
301 | if (copy_to_user(&cpu->lg->lguest_data->time, |
302 | &now, sizeof(struct timespec))) | |
303 | kill_guest(cpu, "Writing timestamp"); | |
6c8dca5d | 304 | } |