drivers/lguest/hypercalls.c

   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
   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 <irq_vectors.h>
  29 #include "lg.h"
  30
  31 /*H:120 This is the core hypercall routine: where the Guest gets what it
  32  * wants.  Or gets killed.  Or, in the case of LHCALL_CRASH, both.
  33  *
  34  * Remember from the Guest: %eax == which call to make, and the arguments are
  35  * packed into %edx, %ebx and %ecx if needed. */
  36 static void do_hcall(struct lguest *lg, struct lguest_regs *regs)
  37 {
  38         switch (regs->eax) {
  39         case LHCALL_FLUSH_ASYNC:
  40                 /* This call does nothing, except by breaking out of the Guest
  41                  * it makes us process all the asynchronous hypercalls. */
  42                 break;
  43         case LHCALL_LGUEST_INIT:
  44                 /* You can't get here unless you're already initialized.  Don't
  45                  * do that. */
  46                 kill_guest(lg, "already have lguest_data");
  47                 break;
  48         case LHCALL_CRASH: {
  49                 /* Crash is such a trivial hypercall that we do it in four
  50                  * lines right here. */
  51                 char msg[128];
  52                 /* If the lgread fails, it will call kill_guest() itself; the
  53                  * kill_guest() with the message will be ignored. */
  54                 lgread(lg, msg, regs->edx, sizeof(msg));
  55                 msg[sizeof(msg)-1] = '\0';
  56                 kill_guest(lg, "CRASH: %s", msg);
  57                 break;
  58         }
  59         case LHCALL_FLUSH_TLB:
  60                 /* FLUSH_TLB comes in two flavors, depending on the
  61                  * argument: */
  62                 if (regs->edx)
  63                         guest_pagetable_clear_all(lg);
  64                 else
  65                         guest_pagetable_flush_user(lg);
  66                 break;
  67         case LHCALL_GET_WALLCLOCK: {
  68                 /* The Guest wants to know the real time in seconds since 1970,
  69                  * in good Unix tradition. */
  70                 struct timespec ts;
  71                 ktime_get_real_ts(&ts);
  72                 regs->eax = ts.tv_sec;
  73                 break;
  74         }
  75         case LHCALL_BIND_DMA:
  76                 /* BIND_DMA really wants four arguments, but it's the only call
  77                  * which does.  So the Guest packs the number of buffers and
  78                  * the interrupt number into the final argument, and we decode
  79                  * it here.  This can legitimately fail, since we currently
  80                  * place a limit on the number of DMA pools a Guest can have.
  81                  * So we return true or false from this call. */
  82                 regs->eax = bind_dma(lg, regs->edx, regs->ebx,
  83                                      regs->ecx >> 8, regs->ecx & 0xFF);
  84                 break;
  85
  86         /* All these calls simply pass the arguments through to the right
  87          * routines. */
  88         case LHCALL_SEND_DMA:
  89                 send_dma(lg, regs->edx, regs->ebx);
  90                 break;
  91         case LHCALL_LOAD_GDT:
  92                 load_guest_gdt(lg, regs->edx, regs->ebx);
  93                 break;
  94         case LHCALL_LOAD_IDT_ENTRY:
  95                 load_guest_idt_entry(lg, regs->edx, regs->ebx, regs->ecx);
  96                 break;
  97         case LHCALL_NEW_PGTABLE:
  98                 guest_new_pagetable(lg, regs->edx);
  99                 break;
 100         case LHCALL_SET_STACK:
 101                 guest_set_stack(lg, regs->edx, regs->ebx, regs->ecx);
 102                 break;
 103         case LHCALL_SET_PTE:
 104                 guest_set_pte(lg, regs->edx, regs->ebx, mkgpte(regs->ecx));
 105                 break;
 106         case LHCALL_SET_PMD:
 107                 guest_set_pmd(lg, regs->edx, regs->ebx);
 108                 break;
 109         case LHCALL_LOAD_TLS:
 110                 guest_load_tls(lg, regs->edx);
 111                 break;
 112         case LHCALL_SET_CLOCKEVENT:
 113                 guest_set_clockevent(lg, regs->edx);
 114                 break;
 115
 116         case LHCALL_TS:
 117                 /* This sets the TS flag, as we saw used in run_guest(). */
 118                 lg->ts = regs->edx;
 119                 break;
 120         case LHCALL_HALT:
 121                 /* Similarly, this sets the halted flag for run_guest(). */
 122                 lg->halted = 1;
 123                 break;
 124         default:
 125                 kill_guest(lg, "Bad hypercall %li\n", regs->eax);
 126         }
 127 }
 128
 129 /* Asynchronous hypercalls are easy: we just look in the array in the Guest's
 130  * "struct lguest_data" and see if there are any new ones marked "ready".
 131  *
 132  * We are careful to do these in order: obviously we respect the order the
 133  * Guest put them in the ring, but we also promise the Guest that they will
 134  * happen before any normal hypercall (which is why we check this before
 135  * checking for a normal hcall). */
 136 static void do_async_hcalls(struct lguest *lg)
 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, &lg->lguest_data->hcall_status, sizeof(st)))
 143                 return;
 144
 145
 146         /* We process "struct lguest_data"s hcalls[] ring once. */
 147         for (i = 0; i < ARRAY_SIZE(st); i++) {
 148                 struct lguest_regs regs;
 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                 unsigned int n = lg->next_hcall;
 153
 154                 /* 0xFF means there's no call here (yet). */
 155                 if (st[n] == 0xFF)
 156                         break;
 157
 158                 /* OK, we have hypercall.  Increment the "next_hcall" cursor,
 159                  * and wrap back to 0 if we reach the end. */
 160                 if (++lg->next_hcall == LHCALL_RING_SIZE)
 161                         lg->next_hcall = 0;
 162
 163                 /* We copy the hypercall arguments into a fake register
 164                  * structure.  This makes life simple for do_hcall(). */
 165                 if (get_user(regs.eax, &lg->lguest_data->hcalls[n].eax)
 166                     || get_user(regs.edx, &lg->lguest_data->hcalls[n].edx)
 167                     || get_user(regs.ecx, &lg->lguest_data->hcalls[n].ecx)
 168                     || get_user(regs.ebx, &lg->lguest_data->hcalls[n].ebx)) {
 169                         kill_guest(lg, "Fetching async hypercalls");
 170                         break;
 171                 }
 172
 173                 /* Do the hypercall, same as a normal one. */
 174                 do_hcall(lg, &regs);
 175
 176                 /* Mark the hypercall done. */
 177                 if (put_user(0xFF, &lg->lguest_data->hcall_status[n])) {
 178                         kill_guest(lg, "Writing result for async hypercall");
 179                         break;
 180                 }
 181
 182                 /* Stop doing hypercalls if we've just done a DMA to the
 183                  * Launcher: it needs to service this first. */
 184                 if (lg->dma_is_pending)
 185                         break;
 186         }
 187 }
 188
 189 /* Last of all, we look at what happens first of all.  The very first time the
 190  * Guest makes a hypercall, we end up here to set things up: */
 191 static void initialize(struct lguest *lg)
 192 {
 193         u32 tsc_speed;
 194
 195         /* You can't do anything until you're initialized.  The Guest knows the
 196          * rules, so we're unforgiving here. */
 197         if (lg->regs->eax != LHCALL_LGUEST_INIT) {
 198                 kill_guest(lg, "hypercall %li before LGUEST_INIT",
 199                            lg->regs->eax);
 200                 return;
 201         }
 202
 203         /* We insist that the Time Stamp Counter exist and doesn't change with
 204          * cpu frequency.  Some devious chip manufacturers decided that TSC
 205          * changes could be handled in software.  I decided that time going
 206          * backwards might be good for benchmarks, but it's bad for users.
 207          *
 208          * We also insist that the TSC be stable: the kernel detects unreliable
 209          * TSCs for its own purposes, and we use that here. */
 210         if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
 211                 tsc_speed = tsc_khz;
 212         else
 213                 tsc_speed = 0;
 214
 215         /* The pointer to the Guest's "struct lguest_data" is the only
 216          * argument. */
 217         lg->lguest_data = (struct lguest_data __user *)lg->regs->edx;
 218         /* If we check the address they gave is OK now, we can simply
 219          * copy_to_user/from_user from now on rather than using lgread/lgwrite.
 220          * I put this in to show that I'm not immune to writing stupid
 221          * optimizations. */
 222         if (!lguest_address_ok(lg, lg->regs->edx, sizeof(*lg->lguest_data))) {
 223                 kill_guest(lg, "bad guest page %p", lg->lguest_data);
 224                 return;
 225         }
 226         /* The Guest tells us where we're not to deliver interrupts by putting
 227          * the range of addresses into "struct lguest_data". */
 228         if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start)
 229             || get_user(lg->noirq_end, &lg->lguest_data->noirq_end)
 230             /* We tell the Guest that it can't use the top 4MB of virtual
 231              * addresses used by the Switcher. */
 232             || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem)
 233             || put_user(tsc_speed, &lg->lguest_data->tsc_khz)
 234             /* We also give the Guest a unique id, as used in lguest_net.c. */
 235             || put_user(lg->guestid, &lg->lguest_data->guestid))
 236                 kill_guest(lg, "bad guest page %p", lg->lguest_data);
 237
 238         /* This is the one case where the above accesses might have been the
 239          * first write to a Guest page.  This may have caused a copy-on-write
 240          * fault, but the Guest might be referring to the old (read-only)
 241          * page. */
 242         guest_pagetable_clear_all(lg);
 243 }
 244 /* Now we've examined the hypercall code; our Guest can make requests.  There
 245  * is one other way we can do things for the Guest, as we see in
 246  * emulate_insn(). */
 247
 248 /*H:110 Tricky point: we mark the hypercall as "done" once we've done it.
 249  * Normally we don't need to do this: the Guest will run again and update the
 250  * trap number before we come back around the run_guest() loop to
 251  * do_hypercalls().
 252  *
 253  * However, if we are signalled or the Guest sends DMA to the Launcher, that
 254  * loop will exit without running the Guest.  When it comes back it would try
 255  * to re-run the hypercall. */
 256 static void clear_hcall(struct lguest *lg)
 257 {
 258         lg->regs->trapnum = 255;
 259 }
 260
 261 /*H:100
 262  * Hypercalls
 263  *
 264  * Remember from the Guest, hypercalls come in two flavors: normal and
 265  * asynchronous.  This file handles both of types.
 266  */
 267 void do_hypercalls(struct lguest *lg)
 268 {
 269         /* Not initialized yet? */
 270         if (unlikely(!lg->lguest_data)) {
 271                 /* Did the Guest make a hypercall?  We might have come back for
 272                  * some other reason (an interrupt, a different trap). */
 273                 if (lg->regs->trapnum == LGUEST_TRAP_ENTRY) {
 274                         /* Set up the "struct lguest_data" */
 275                         initialize(lg);
 276                         /* The hypercall is done. */
 277                         clear_hcall(lg);
 278                 }
 279                 return;
 280         }
 281
 282         /* The Guest has initialized.
 283          *
 284          * Look in the hypercall ring for the async hypercalls: */
 285         do_async_hcalls(lg);
 286
 287         /* If we stopped reading the hypercall ring because the Guest did a
 288          * SEND_DMA to the Launcher, we want to return now.  Otherwise if the
 289          * Guest asked us to do a hypercall, we do it. */
 290         if (!lg->dma_is_pending && lg->regs->trapnum == LGUEST_TRAP_ENTRY) {
 291                 do_hcall(lg, lg->regs);
 292                 /* The hypercall is done. */
 293                 clear_hcall(lg);
 294         }
 295 }