Commit | Line | Data |
---|---|---|
f938d2c8 RR |
1 | /*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 | |
d7e28ffe RR |
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> | |
ca94f2bd | 26 | #include <linux/ktime.h> |
d7e28ffe RR |
27 | #include <asm/page.h> |
28 | #include <asm/pgtable.h> | |
d7e28ffe RR |
29 | #include "lg.h" |
30 | ||
b410e7b1 | 31 | /*H:120 This is the core hypercall routine: where the Guest gets what it wants. |
a6bd8e13 | 32 | * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. */ |
73044f05 | 33 | static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) |
d7e28ffe | 34 | { |
b410e7b1 | 35 | switch (args->arg0) { |
d7e28ffe | 36 | case LHCALL_FLUSH_ASYNC: |
bff672e6 RR |
37 | /* This call does nothing, except by breaking out of the Guest |
38 | * it makes us process all the asynchronous hypercalls. */ | |
d7e28ffe | 39 | break; |
a32a8813 RR |
40 | case LHCALL_SEND_INTERRUPTS: |
41 | /* This call does nothing too, but by breaking out of the Guest | |
42 | * it makes us process any pending interrupts. */ | |
43 | break; | |
d7e28ffe | 44 | case LHCALL_LGUEST_INIT: |
bff672e6 RR |
45 | /* You can't get here unless you're already initialized. Don't |
46 | * do that. */ | |
382ac6b3 | 47 | kill_guest(cpu, "already have lguest_data"); |
d7e28ffe | 48 | break; |
ec04b13f BR |
49 | case LHCALL_SHUTDOWN: { |
50 | /* Shutdown is such a trivial hypercall that we do it in four | |
bff672e6 | 51 | * lines right here. */ |
d7e28ffe | 52 | char msg[128]; |
bff672e6 RR |
53 | /* If the lgread fails, it will call kill_guest() itself; the |
54 | * kill_guest() with the message will be ignored. */ | |
382ac6b3 | 55 | __lgread(cpu, msg, args->arg1, sizeof(msg)); |
d7e28ffe | 56 | msg[sizeof(msg)-1] = '\0'; |
382ac6b3 | 57 | kill_guest(cpu, "CRASH: %s", msg); |
ec04b13f | 58 | if (args->arg2 == LGUEST_SHUTDOWN_RESTART) |
382ac6b3 | 59 | cpu->lg->dead = ERR_PTR(-ERESTART); |
d7e28ffe RR |
60 | break; |
61 | } | |
62 | case LHCALL_FLUSH_TLB: | |
bff672e6 RR |
63 | /* FLUSH_TLB comes in two flavors, depending on the |
64 | * argument: */ | |
b410e7b1 | 65 | if (args->arg1) |
4665ac8e | 66 | guest_pagetable_clear_all(cpu); |
d7e28ffe | 67 | else |
1713608f | 68 | guest_pagetable_flush_user(cpu); |
d7e28ffe | 69 | break; |
bff672e6 RR |
70 | |
71 | /* All these calls simply pass the arguments through to the right | |
72 | * routines. */ | |
d7e28ffe | 73 | case LHCALL_NEW_PGTABLE: |
4665ac8e | 74 | guest_new_pagetable(cpu, args->arg1); |
d7e28ffe RR |
75 | break; |
76 | case LHCALL_SET_STACK: | |
4665ac8e | 77 | guest_set_stack(cpu, args->arg1, args->arg2, args->arg3); |
d7e28ffe RR |
78 | break; |
79 | case LHCALL_SET_PTE: | |
acdd0b62 MZ |
80 | #ifdef CONFIG_X86_PAE |
81 | guest_set_pte(cpu, args->arg1, args->arg2, | |
82 | __pte(args->arg3 | (u64)args->arg4 << 32)); | |
83 | #else | |
382ac6b3 | 84 | guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3)); |
acdd0b62 | 85 | #endif |
d7e28ffe | 86 | break; |
ebe0ba84 MZ |
87 | case LHCALL_SET_PGD: |
88 | guest_set_pgd(cpu->lg, args->arg1, args->arg2); | |
d7e28ffe | 89 | break; |
acdd0b62 MZ |
90 | #ifdef CONFIG_X86_PAE |
91 | case LHCALL_SET_PMD: | |
92 | guest_set_pmd(cpu->lg, args->arg1, args->arg2); | |
93 | break; | |
94 | #endif | |
d7e28ffe | 95 | case LHCALL_SET_CLOCKEVENT: |
ad8d8f3b | 96 | guest_set_clockevent(cpu, args->arg1); |
d7e28ffe RR |
97 | break; |
98 | case LHCALL_TS: | |
bff672e6 | 99 | /* This sets the TS flag, as we saw used in run_guest(). */ |
4665ac8e | 100 | cpu->ts = args->arg1; |
d7e28ffe RR |
101 | break; |
102 | case LHCALL_HALT: | |
bff672e6 | 103 | /* Similarly, this sets the halted flag for run_guest(). */ |
66686c2a | 104 | cpu->halted = 1; |
d7e28ffe | 105 | break; |
15045275 | 106 | case LHCALL_NOTIFY: |
5e232f4f | 107 | cpu->pending_notify = args->arg1; |
15045275 | 108 | break; |
d7e28ffe | 109 | default: |
e1e72965 | 110 | /* It should be an architecture-specific hypercall. */ |
73044f05 | 111 | if (lguest_arch_do_hcall(cpu, args)) |
382ac6b3 | 112 | kill_guest(cpu, "Bad hypercall %li\n", args->arg0); |
d7e28ffe RR |
113 | } |
114 | } | |
b410e7b1 | 115 | /*:*/ |
d7e28ffe | 116 | |
b410e7b1 JS |
117 | /*H:124 Asynchronous hypercalls are easy: we just look in the array in the |
118 | * Guest's "struct lguest_data" to see if any new ones are marked "ready". | |
bff672e6 RR |
119 | * |
120 | * We are careful to do these in order: obviously we respect the order the | |
121 | * Guest put them in the ring, but we also promise the Guest that they will | |
122 | * happen before any normal hypercall (which is why we check this before | |
123 | * checking for a normal hcall). */ | |
73044f05 | 124 | static void do_async_hcalls(struct lg_cpu *cpu) |
d7e28ffe RR |
125 | { |
126 | unsigned int i; | |
127 | u8 st[LHCALL_RING_SIZE]; | |
128 | ||
bff672e6 | 129 | /* For simplicity, we copy the entire call status array in at once. */ |
382ac6b3 | 130 | if (copy_from_user(&st, &cpu->lg->lguest_data->hcall_status, sizeof(st))) |
d7e28ffe RR |
131 | return; |
132 | ||
bff672e6 | 133 | /* We process "struct lguest_data"s hcalls[] ring once. */ |
d7e28ffe | 134 | for (i = 0; i < ARRAY_SIZE(st); i++) { |
b410e7b1 | 135 | struct hcall_args args; |
bff672e6 RR |
136 | /* We remember where we were up to from last time. This makes |
137 | * sure that the hypercalls are done in the order the Guest | |
138 | * places them in the ring. */ | |
73044f05 | 139 | unsigned int n = cpu->next_hcall; |
d7e28ffe | 140 | |
bff672e6 | 141 | /* 0xFF means there's no call here (yet). */ |
d7e28ffe RR |
142 | if (st[n] == 0xFF) |
143 | break; | |
144 | ||
bff672e6 RR |
145 | /* OK, we have hypercall. Increment the "next_hcall" cursor, |
146 | * and wrap back to 0 if we reach the end. */ | |
73044f05 GOC |
147 | if (++cpu->next_hcall == LHCALL_RING_SIZE) |
148 | cpu->next_hcall = 0; | |
d7e28ffe | 149 | |
b410e7b1 JS |
150 | /* Copy the hypercall arguments into a local copy of |
151 | * the hcall_args struct. */ | |
382ac6b3 | 152 | if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], |
b410e7b1 | 153 | sizeof(struct hcall_args))) { |
382ac6b3 | 154 | kill_guest(cpu, "Fetching async hypercalls"); |
d7e28ffe RR |
155 | break; |
156 | } | |
157 | ||
bff672e6 | 158 | /* Do the hypercall, same as a normal one. */ |
73044f05 | 159 | do_hcall(cpu, &args); |
bff672e6 RR |
160 | |
161 | /* Mark the hypercall done. */ | |
382ac6b3 GOC |
162 | if (put_user(0xFF, &cpu->lg->lguest_data->hcall_status[n])) { |
163 | kill_guest(cpu, "Writing result for async hypercall"); | |
d7e28ffe RR |
164 | break; |
165 | } | |
166 | ||
15045275 RR |
167 | /* Stop doing hypercalls if they want to notify the Launcher: |
168 | * it needs to service this first. */ | |
5e232f4f | 169 | if (cpu->pending_notify) |
d7e28ffe RR |
170 | break; |
171 | } | |
172 | } | |
173 | ||
bff672e6 RR |
174 | /* Last of all, we look at what happens first of all. The very first time the |
175 | * Guest makes a hypercall, we end up here to set things up: */ | |
73044f05 | 176 | static void initialize(struct lg_cpu *cpu) |
d7e28ffe | 177 | { |
bff672e6 RR |
178 | /* You can't do anything until you're initialized. The Guest knows the |
179 | * rules, so we're unforgiving here. */ | |
73044f05 | 180 | if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { |
382ac6b3 | 181 | kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); |
d7e28ffe RR |
182 | return; |
183 | } | |
184 | ||
73044f05 | 185 | if (lguest_arch_init_hypercalls(cpu)) |
382ac6b3 | 186 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); |
3c6b5bfa | 187 | |
bff672e6 RR |
188 | /* The Guest tells us where we're not to deliver interrupts by putting |
189 | * the range of addresses into "struct lguest_data". */ | |
382ac6b3 GOC |
190 | if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start) |
191 | || get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end)) | |
192 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); | |
d7e28ffe | 193 | |
e1e72965 RR |
194 | /* We write the current time into the Guest's data page once so it can |
195 | * set its clock. */ | |
382ac6b3 | 196 | write_timestamp(cpu); |
6c8dca5d | 197 | |
47436aa4 | 198 | /* page_tables.c will also do some setup. */ |
382ac6b3 | 199 | page_table_guest_data_init(cpu); |
47436aa4 | 200 | |
bff672e6 RR |
201 | /* This is the one case where the above accesses might have been the |
202 | * first write to a Guest page. This may have caused a copy-on-write | |
e1e72965 RR |
203 | * fault, but the old page might be (read-only) in the Guest |
204 | * pagetable. */ | |
4665ac8e | 205 | guest_pagetable_clear_all(cpu); |
d7e28ffe | 206 | } |
a6bd8e13 RR |
207 | /*:*/ |
208 | ||
209 | /*M:013 If a Guest reads from a page (so creates a mapping) that it has never | |
210 | * written to, and then the Launcher writes to it (ie. the output of a virtual | |
211 | * device), the Guest will still see the old page. In practice, this never | |
212 | * happens: why would the Guest read a page which it has never written to? But | |
213 | * a similar scenario might one day bite us, so it's worth mentioning. :*/ | |
d7e28ffe | 214 | |
bff672e6 RR |
215 | /*H:100 |
216 | * Hypercalls | |
217 | * | |
218 | * Remember from the Guest, hypercalls come in two flavors: normal and | |
219 | * asynchronous. This file handles both of types. | |
220 | */ | |
73044f05 | 221 | void do_hypercalls(struct lg_cpu *cpu) |
d7e28ffe | 222 | { |
cc6d4fbc | 223 | /* Not initialized yet? This hypercall must do it. */ |
73044f05 | 224 | if (unlikely(!cpu->lg->lguest_data)) { |
cc6d4fbc | 225 | /* Set up the "struct lguest_data" */ |
73044f05 | 226 | initialize(cpu); |
cc6d4fbc | 227 | /* Hcall is done. */ |
73044f05 | 228 | cpu->hcall = NULL; |
d7e28ffe RR |
229 | return; |
230 | } | |
231 | ||
bff672e6 RR |
232 | /* The Guest has initialized. |
233 | * | |
234 | * Look in the hypercall ring for the async hypercalls: */ | |
73044f05 | 235 | do_async_hcalls(cpu); |
bff672e6 RR |
236 | |
237 | /* If we stopped reading the hypercall ring because the Guest did a | |
15045275 | 238 | * NOTIFY to the Launcher, we want to return now. Otherwise we do |
cc6d4fbc | 239 | * the hypercall. */ |
5e232f4f | 240 | if (!cpu->pending_notify) { |
73044f05 | 241 | do_hcall(cpu, cpu->hcall); |
cc6d4fbc RR |
242 | /* Tricky point: we reset the hcall pointer to mark the |
243 | * hypercall as "done". We use the hcall pointer rather than | |
244 | * the trap number to indicate a hypercall is pending. | |
245 | * Normally it doesn't matter: the Guest will run again and | |
246 | * update the trap number before we come back here. | |
247 | * | |
e1e72965 | 248 | * However, if we are signalled or the Guest sends I/O to the |
cc6d4fbc RR |
249 | * Launcher, the run_guest() loop will exit without running the |
250 | * Guest. When it comes back it would try to re-run the | |
a6bd8e13 | 251 | * hypercall. Finding that bug sucked. */ |
73044f05 | 252 | cpu->hcall = NULL; |
d7e28ffe RR |
253 | } |
254 | } | |
6c8dca5d RR |
255 | |
256 | /* This routine supplies the Guest with time: it's used for wallclock time at | |
257 | * initial boot and as a rough time source if the TSC isn't available. */ | |
382ac6b3 | 258 | void write_timestamp(struct lg_cpu *cpu) |
6c8dca5d RR |
259 | { |
260 | struct timespec now; | |
261 | ktime_get_real_ts(&now); | |
382ac6b3 GOC |
262 | if (copy_to_user(&cpu->lg->lguest_data->time, |
263 | &now, sizeof(struct timespec))) | |
264 | kill_guest(cpu, "Writing timestamp"); | |
6c8dca5d | 265 | } |