diff options
Diffstat (limited to 'drivers/lguest/hypercalls.c')
| -rw-r--r-- | drivers/lguest/hypercalls.c | 145 | 
1 files changed, 96 insertions, 49 deletions
| diff --git a/drivers/lguest/hypercalls.c b/drivers/lguest/hypercalls.c index c29ffa19cb74..83511eb0923d 100644 --- a/drivers/lguest/hypercalls.c +++ b/drivers/lguest/hypercalls.c | |||
| @@ -1,8 +1,10 @@ | |||
| 1 | /*P:500 Just as userspace programs request kernel operations through a system | 1 | /*P:500 | 
| 2 | * Just as userspace programs request kernel operations through a system | ||
| 2 | * call, the Guest requests Host operations through a "hypercall". You might | 3 | * 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 | * 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 | * 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 | * code is basically a one big switch statement. | 
| 7 | :*/ | ||
| 6 | 8 | ||
| 7 | /* Copyright (C) 2006 Rusty Russell IBM Corporation | 9 | /* Copyright (C) 2006 Rusty Russell IBM Corporation | 
| 8 | 10 | ||
| @@ -28,30 +30,41 @@ | |||
| 28 | #include <asm/pgtable.h> | 30 | #include <asm/pgtable.h> | 
| 29 | #include "lg.h" | 31 | #include "lg.h" | 
| 30 | 32 | ||
| 31 | /*H:120 This is the core hypercall routine: where the Guest gets what it wants. | 33 | /*H:120 | 
| 32 | * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. */ | 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 | */ | ||
| 33 | static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) | 37 | static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) | 
| 34 | { | 38 | { | 
| 35 | switch (args->arg0) { | 39 | switch (args->arg0) { | 
| 36 | case LHCALL_FLUSH_ASYNC: | 40 | case LHCALL_FLUSH_ASYNC: | 
| 37 | /* This call does nothing, except by breaking out of the Guest | 41 | /* | 
| 38 | * it makes us process all the asynchronous hypercalls. */ | 42 | * This call does nothing, except by breaking out of the Guest | 
| 43 | * it makes us process all the asynchronous hypercalls. | ||
| 44 | */ | ||
| 39 | break; | 45 | break; | 
| 40 | case LHCALL_SEND_INTERRUPTS: | 46 | case LHCALL_SEND_INTERRUPTS: | 
| 41 | /* This call does nothing too, but by breaking out of the Guest | 47 | /* | 
| 42 | * it makes us process any pending interrupts. */ | 48 | * This call does nothing too, but by breaking out of the Guest | 
| 49 | * it makes us process any pending interrupts. | ||
| 50 | */ | ||
| 43 | break; | 51 | break; | 
| 44 | case LHCALL_LGUEST_INIT: | 52 | case LHCALL_LGUEST_INIT: | 
| 45 | /* You can't get here unless you're already initialized. Don't | 53 | /* | 
| 46 | * do that. */ | 54 | * You can't get here unless you're already initialized. Don't | 
| 55 | * do that. | ||
| 56 | */ | ||
| 47 | kill_guest(cpu, "already have lguest_data"); | 57 | kill_guest(cpu, "already have lguest_data"); | 
| 48 | break; | 58 | break; | 
| 49 | case LHCALL_SHUTDOWN: { | 59 | case LHCALL_SHUTDOWN: { | 
| 50 | /* Shutdown is such a trivial hypercall that we do it in four | ||
| 51 | * lines right here. */ | ||
| 52 | char msg[128]; | 60 | char msg[128]; | 
| 53 | /* If the lgread fails, it will call kill_guest() itself; the | 61 | /* | 
| 54 | * kill_guest() with the message will be ignored. */ | 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 | */ | ||
| 55 | __lgread(cpu, msg, args->arg1, sizeof(msg)); | 68 | __lgread(cpu, msg, args->arg1, sizeof(msg)); | 
| 56 | msg[sizeof(msg)-1] = '\0'; | 69 | msg[sizeof(msg)-1] = '\0'; | 
| 57 | kill_guest(cpu, "CRASH: %s", msg); | 70 | kill_guest(cpu, "CRASH: %s", msg); | 
| @@ -60,16 +73,17 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) | |||
| 60 | break; | 73 | break; | 
| 61 | } | 74 | } | 
| 62 | case LHCALL_FLUSH_TLB: | 75 | case LHCALL_FLUSH_TLB: | 
| 63 | /* FLUSH_TLB comes in two flavors, depending on the | 76 | /* FLUSH_TLB comes in two flavors, depending on the argument: */ | 
| 64 | * argument: */ | ||
| 65 | if (args->arg1) | 77 | if (args->arg1) | 
| 66 | guest_pagetable_clear_all(cpu); | 78 | guest_pagetable_clear_all(cpu); | 
| 67 | else | 79 | else | 
| 68 | guest_pagetable_flush_user(cpu); | 80 | guest_pagetable_flush_user(cpu); | 
| 69 | break; | 81 | break; | 
| 70 | 82 | ||
| 71 | /* All these calls simply pass the arguments through to the right | 83 | /* | 
| 72 | * routines. */ | 84 | * All these calls simply pass the arguments through to the right | 
| 85 | * routines. | ||
| 86 | */ | ||
| 73 | case LHCALL_NEW_PGTABLE: | 87 | case LHCALL_NEW_PGTABLE: | 
| 74 | guest_new_pagetable(cpu, args->arg1); | 88 | guest_new_pagetable(cpu, args->arg1); | 
| 75 | break; | 89 | break; | 
| @@ -112,15 +126,16 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) | |||
| 112 | kill_guest(cpu, "Bad hypercall %li\n", args->arg0); | 126 | kill_guest(cpu, "Bad hypercall %li\n", args->arg0); | 
| 113 | } | 127 | } | 
| 114 | } | 128 | } | 
| 115 | /*:*/ | ||
| 116 | 129 | ||
| 117 | /*H:124 Asynchronous hypercalls are easy: we just look in the array in the | 130 | /*H:124 | 
| 131 | * 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". | 132 | * Guest's "struct lguest_data" to see if any new ones are marked "ready". | 
| 119 | * | 133 | * | 
| 120 | * We are careful to do these in order: obviously we respect the order the | 134 | * 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 | 135 | * 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 | 136 | * happen before any normal hypercall (which is why we check this before | 
| 123 | * checking for a normal hcall). */ | 137 | * checking for a normal hcall). | 
| 138 | */ | ||
| 124 | static void do_async_hcalls(struct lg_cpu *cpu) | 139 | static void do_async_hcalls(struct lg_cpu *cpu) | 
| 125 | { | 140 | { | 
| 126 | unsigned int i; | 141 | unsigned int i; | 
| @@ -133,22 +148,28 @@ static void do_async_hcalls(struct lg_cpu *cpu) | |||
| 133 | /* We process "struct lguest_data"s hcalls[] ring once. */ | 148 | /* We process "struct lguest_data"s hcalls[] ring once. */ | 
| 134 | for (i = 0; i < ARRAY_SIZE(st); i++) { | 149 | for (i = 0; i < ARRAY_SIZE(st); i++) { | 
| 135 | struct hcall_args args; | 150 | struct hcall_args args; | 
| 136 | /* We remember where we were up to from last time. This makes | 151 | /* | 
| 152 | * We remember where we were up to from last time. This makes | ||
| 137 | * sure that the hypercalls are done in the order the Guest | 153 | * sure that the hypercalls are done in the order the Guest | 
| 138 | * places them in the ring. */ | 154 | * places them in the ring. | 
| 155 | */ | ||
| 139 | unsigned int n = cpu->next_hcall; | 156 | unsigned int n = cpu->next_hcall; | 
| 140 | 157 | ||
| 141 | /* 0xFF means there's no call here (yet). */ | 158 | /* 0xFF means there's no call here (yet). */ | 
| 142 | if (st[n] == 0xFF) | 159 | if (st[n] == 0xFF) | 
| 143 | break; | 160 | break; | 
| 144 | 161 | ||
| 145 | /* OK, we have hypercall. Increment the "next_hcall" cursor, | 162 | /* | 
| 146 | * and wrap back to 0 if we reach the end. */ | 163 | * OK, we have hypercall. Increment the "next_hcall" cursor, | 
| 164 | * and wrap back to 0 if we reach the end. | ||
| 165 | */ | ||
| 147 | if (++cpu->next_hcall == LHCALL_RING_SIZE) | 166 | if (++cpu->next_hcall == LHCALL_RING_SIZE) | 
| 148 | cpu->next_hcall = 0; | 167 | cpu->next_hcall = 0; | 
| 149 | 168 | ||
| 150 | /* Copy the hypercall arguments into a local copy of | 169 | /* | 
| 151 | * the hcall_args struct. */ | 170 | * Copy the hypercall arguments into a local copy of the | 
| 171 | * hcall_args struct. | ||
| 172 | */ | ||
| 152 | if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], | 173 | if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], | 
| 153 | sizeof(struct hcall_args))) { | 174 | sizeof(struct hcall_args))) { | 
| 154 | kill_guest(cpu, "Fetching async hypercalls"); | 175 | kill_guest(cpu, "Fetching async hypercalls"); | 
| @@ -164,19 +185,25 @@ static void do_async_hcalls(struct lg_cpu *cpu) | |||
| 164 | break; | 185 | break; | 
| 165 | } | 186 | } | 
| 166 | 187 | ||
| 167 | /* Stop doing hypercalls if they want to notify the Launcher: | 188 | /* | 
| 168 | * it needs to service this first. */ | 189 | * Stop doing hypercalls if they want to notify the Launcher: | 
| 190 | * it needs to service this first. | ||
| 191 | */ | ||
| 169 | if (cpu->pending_notify) | 192 | if (cpu->pending_notify) | 
| 170 | break; | 193 | break; | 
| 171 | } | 194 | } | 
| 172 | } | 195 | } | 
| 173 | 196 | ||
| 174 | /* Last of all, we look at what happens first of all. The very first time the | 197 | /* | 
| 175 | * Guest makes a hypercall, we end up here to set things up: */ | 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 | */ | ||
| 176 | static void initialize(struct lg_cpu *cpu) | 201 | static void initialize(struct lg_cpu *cpu) | 
| 177 | { | 202 | { | 
| 178 | /* You can't do anything until you're initialized. The Guest knows the | 203 | /* | 
| 179 | * rules, so we're unforgiving here. */ | 204 | * You can't do anything until you're initialized. The Guest knows the | 
| 205 | * rules, so we're unforgiving here. | ||
| 206 | */ | ||
| 180 | if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { | 207 | if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { | 
| 181 | kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); | 208 | kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); | 
| 182 | return; | 209 | return; | 
| @@ -185,32 +212,44 @@ static void initialize(struct lg_cpu *cpu) | |||
| 185 | if (lguest_arch_init_hypercalls(cpu)) | 212 | if (lguest_arch_init_hypercalls(cpu)) | 
| 186 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); | 213 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); | 
| 187 | 214 | ||
| 188 | /* The Guest tells us where we're not to deliver interrupts by putting | 215 | /* | 
| 189 | * the range of addresses into "struct lguest_data". */ | 216 | * The Guest tells us where we're not to deliver interrupts by putting | 
| 217 | * the range of addresses into "struct lguest_data". | ||
| 218 | */ | ||
| 190 | if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start) | 219 | 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)) | 220 | || get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end)) | 
| 192 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); | 221 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); | 
| 193 | 222 | ||
| 194 | /* We write the current time into the Guest's data page once so it can | 223 | /* | 
| 195 | * set its clock. */ | 224 | * We write the current time into the Guest's data page once so it can | 
| 225 | * set its clock. | ||
| 226 | */ | ||
| 196 | write_timestamp(cpu); | 227 | write_timestamp(cpu); | 
| 197 | 228 | ||
| 198 | /* page_tables.c will also do some setup. */ | 229 | /* page_tables.c will also do some setup. */ | 
| 199 | page_table_guest_data_init(cpu); | 230 | page_table_guest_data_init(cpu); | 
| 200 | 231 | ||
| 201 | /* This is the one case where the above accesses might have been the | 232 | /* | 
| 233 | * 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 | 234 | * first write to a Guest page. This may have caused a copy-on-write | 
| 203 | * fault, but the old page might be (read-only) in the Guest | 235 | * fault, but the old page might be (read-only) in the Guest | 
| 204 | * pagetable. */ | 236 | * pagetable. | 
| 237 | */ | ||
| 205 | guest_pagetable_clear_all(cpu); | 238 | guest_pagetable_clear_all(cpu); | 
| 206 | } | 239 | } | 
| 207 | /*:*/ | 240 | /*:*/ | 
| 208 | 241 | ||
| 209 | /*M:013 If a Guest reads from a page (so creates a mapping) that it has never | 242 | /*M:013 | 
| 243 | * 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 | 244 | * 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 | 245 | * 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 | 246 | * 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. :*/ | 247 | * a similar scenario might one day bite us, so it's worth mentioning. | 
| 248 | * | ||
| 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. | ||
| 252 | :*/ | ||
| 214 | 253 | ||
| 215 | /*H:100 | 254 | /*H:100 | 
| 216 | * Hypercalls | 255 | * Hypercalls | 
| @@ -229,17 +268,22 @@ void do_hypercalls(struct lg_cpu *cpu) | |||
| 229 | return; | 268 | return; | 
| 230 | } | 269 | } | 
| 231 | 270 | ||
| 232 | /* The Guest has initialized. | 271 | /* | 
| 272 | * The Guest has initialized. | ||
| 233 | * | 273 | * | 
| 234 | * Look in the hypercall ring for the async hypercalls: */ | 274 | * Look in the hypercall ring for the async hypercalls: | 
| 275 | */ | ||
| 235 | do_async_hcalls(cpu); | 276 | do_async_hcalls(cpu); | 
| 236 | 277 | ||
| 237 | /* If we stopped reading the hypercall ring because the Guest did a | 278 | /* | 
| 279 | * If we stopped reading the hypercall ring because the Guest did a | ||
| 238 | * NOTIFY to the Launcher, we want to return now. Otherwise we do | 280 | * NOTIFY to the Launcher, we want to return now. Otherwise we do | 
| 239 | * the hypercall. */ | 281 | * the hypercall. | 
| 282 | */ | ||
| 240 | if (!cpu->pending_notify) { | 283 | if (!cpu->pending_notify) { | 
| 241 | do_hcall(cpu, cpu->hcall); | 284 | do_hcall(cpu, cpu->hcall); | 
| 242 | /* Tricky point: we reset the hcall pointer to mark the | 285 | /* | 
| 286 | * Tricky point: we reset the hcall pointer to mark the | ||
| 243 | * hypercall as "done". We use the hcall pointer rather than | 287 | * hypercall as "done". We use the hcall pointer rather than | 
| 244 | * the trap number to indicate a hypercall is pending. | 288 | * the trap number to indicate a hypercall is pending. | 
| 245 | * Normally it doesn't matter: the Guest will run again and | 289 | * Normally it doesn't matter: the Guest will run again and | 
| @@ -248,13 +292,16 @@ void do_hypercalls(struct lg_cpu *cpu) | |||
| 248 | * However, if we are signalled or the Guest sends I/O to the | 292 | * However, if we are signalled or the Guest sends I/O to the | 
| 249 | * Launcher, the run_guest() loop will exit without running the | 293 | * Launcher, the run_guest() loop will exit without running the | 
| 250 | * Guest. When it comes back it would try to re-run the | 294 | * Guest. When it comes back it would try to re-run the | 
| 251 | * hypercall. Finding that bug sucked. */ | 295 | * hypercall. Finding that bug sucked. | 
| 296 | */ | ||
| 252 | cpu->hcall = NULL; | 297 | cpu->hcall = NULL; | 
| 253 | } | 298 | } | 
| 254 | } | 299 | } | 
| 255 | 300 | ||
| 256 | /* This routine supplies the Guest with time: it's used for wallclock time at | 301 | /* | 
| 257 | * initial boot and as a rough time source if the TSC isn't available. */ | 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 | */ | ||
| 258 | void write_timestamp(struct lg_cpu *cpu) | 305 | void write_timestamp(struct lg_cpu *cpu) | 
| 259 | { | 306 | { | 
| 260 | struct timespec now; | 307 | struct timespec now; | 
