]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/lguest/x86/core.c
2 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
3 * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 #include <linux/kernel.h>
21 #include <linux/start_kernel.h>
22 #include <linux/string.h>
23 #include <linux/console.h>
24 #include <linux/screen_info.h>
25 #include <linux/irq.h>
26 #include <linux/interrupt.h>
27 #include <linux/clocksource.h>
28 #include <linux/clockchips.h>
29 #include <linux/cpu.h>
30 #include <linux/lguest.h>
31 #include <linux/lguest_launcher.h>
32 #include <linux/lguest_bus.h>
33 #include <asm/paravirt.h>
34 #include <asm/param.h>
36 #include <asm/pgtable.h>
38 #include <asm/setup.h>
39 #include <asm/lguest.h>
40 #include <asm/uaccess.h>
44 static int cpu_had_pge
;
48 unsigned short segment
;
51 /* Offset from where switcher.S was compiled to where we've copied it */
52 static unsigned long switcher_offset(void)
54 return SWITCHER_ADDR
- (unsigned long)start_switcher_text
;
57 /* This cpu's struct lguest_pages. */
58 static struct lguest_pages
*lguest_pages(unsigned int cpu
)
60 return &(((struct lguest_pages
*)
61 (SWITCHER_ADDR
+ SHARED_SWITCHER_PAGES
*PAGE_SIZE
))[cpu
]);
64 static DEFINE_PER_CPU(struct lguest
*, last_guest
);
67 * We are getting close to the Switcher.
69 * Remember that each CPU has two pages which are visible to the Guest when it
70 * runs on that CPU. This has to contain the state for that Guest: we copy the
71 * state in just before we run the Guest.
73 * Each Guest has "changed" flags which indicate what has changed in the Guest
74 * since it last ran. We saw this set in interrupts_and_traps.c and
77 static void copy_in_guest_info(struct lguest
*lg
, struct lguest_pages
*pages
)
79 /* Copying all this data can be quite expensive. We usually run the
80 * same Guest we ran last time (and that Guest hasn't run anywhere else
81 * meanwhile). If that's not the case, we pretend everything in the
82 * Guest has changed. */
83 if (__get_cpu_var(last_guest
) != lg
|| lg
->last_pages
!= pages
) {
84 __get_cpu_var(last_guest
) = lg
;
85 lg
->last_pages
= pages
;
86 lg
->changed
= CHANGED_ALL
;
89 /* These copies are pretty cheap, so we do them unconditionally: */
90 /* Save the current Host top-level page directory. */
91 pages
->state
.host_cr3
= __pa(current
->mm
->pgd
);
92 /* Set up the Guest's page tables to see this CPU's pages (and no
93 * other CPU's pages). */
94 map_switcher_in_guest(lg
, pages
);
95 /* Set up the two "TSS" members which tell the CPU what stack to use
96 * for traps which do directly into the Guest (ie. traps at privilege
98 pages
->state
.guest_tss
.esp1
= lg
->esp1
;
99 pages
->state
.guest_tss
.ss1
= lg
->ss1
;
101 /* Copy direct-to-Guest trap entries. */
102 if (lg
->changed
& CHANGED_IDT
)
103 copy_traps(lg
, pages
->state
.guest_idt
, default_idt_entries
);
105 /* Copy all GDT entries which the Guest can change. */
106 if (lg
->changed
& CHANGED_GDT
)
107 copy_gdt(lg
, pages
->state
.guest_gdt
);
108 /* If only the TLS entries have changed, copy them. */
109 else if (lg
->changed
& CHANGED_GDT_TLS
)
110 copy_gdt_tls(lg
, pages
->state
.guest_gdt
);
112 /* Mark the Guest as unchanged for next time. */
116 /* Finally: the code to actually call into the Switcher to run the Guest. */
117 static void run_guest_once(struct lguest
*lg
, struct lguest_pages
*pages
)
119 /* This is a dummy value we need for GCC's sake. */
120 unsigned int clobber
;
122 /* Copy the guest-specific information into this CPU's "struct
124 copy_in_guest_info(lg
, pages
);
126 /* Set the trap number to 256 (impossible value). If we fault while
127 * switching to the Guest (bad segment registers or bug), this will
128 * cause us to abort the Guest. */
129 lg
->regs
->trapnum
= 256;
131 /* Now: we push the "eflags" register on the stack, then do an "lcall".
132 * This is how we change from using the kernel code segment to using
133 * the dedicated lguest code segment, as well as jumping into the
136 * The lcall also pushes the old code segment (KERNEL_CS) onto the
137 * stack, then the address of this call. This stack layout happens to
138 * exactly match the stack of an interrupt... */
139 asm volatile("pushf; lcall *lguest_entry"
140 /* This is how we tell GCC that %eax ("a") and %ebx ("b")
141 * are changed by this routine. The "=" means output. */
142 : "=a"(clobber
), "=b"(clobber
)
143 /* %eax contains the pages pointer. ("0" refers to the
144 * 0-th argument above, ie "a"). %ebx contains the
145 * physical address of the Guest's top-level page
147 : "0"(pages
), "1"(__pa(lg
->pgdirs
[lg
->pgdidx
].pgdir
))
148 /* We tell gcc that all these registers could change,
149 * which means we don't have to save and restore them in
151 : "memory", "%edx", "%ecx", "%edi", "%esi");
155 /*H:040 This is the i386-specific code to setup and run the Guest. Interrupts
156 * are disabled: we own the CPU. */
157 void lguest_arch_run_guest(struct lguest
*lg
)
159 /* Remember the awfully-named TS bit? If the Guest has asked
160 * to set it we set it now, so we can trap and pass that trap
161 * to the Guest if it uses the FPU. */
165 /* SYSENTER is an optimized way of doing system calls. We
166 * can't allow it because it always jumps to privilege level 0.
167 * A normal Guest won't try it because we don't advertise it in
168 * CPUID, but a malicious Guest (or malicious Guest userspace
169 * program) could, so we tell the CPU to disable it before
170 * running the Guest. */
171 if (boot_cpu_has(X86_FEATURE_SEP
))
172 wrmsr(MSR_IA32_SYSENTER_CS
, 0, 0);
174 /* Now we actually run the Guest. It will pop back out when
175 * something interesting happens, and we can examine its
176 * registers to see what it was doing. */
177 run_guest_once(lg
, lguest_pages(raw_smp_processor_id()));
179 /* The "regs" pointer contains two extra entries which are not
180 * really registers: a trap number which says what interrupt or
181 * trap made the switcher code come back, and an error code
182 * which some traps set. */
184 /* If the Guest page faulted, then the cr2 register will tell
185 * us the bad virtual address. We have to grab this now,
186 * because once we re-enable interrupts an interrupt could
187 * fault and thus overwrite cr2, or we could even move off to a
189 if (lg
->regs
->trapnum
== 14)
190 lg
->arch
.last_pagefault
= read_cr2();
191 /* Similarly, if we took a trap because the Guest used the FPU,
192 * we have to restore the FPU it expects to see. */
193 else if (lg
->regs
->trapnum
== 7)
194 math_state_restore();
196 /* Restore SYSENTER if it's supposed to be on. */
197 if (boot_cpu_has(X86_FEATURE_SEP
))
198 wrmsr(MSR_IA32_SYSENTER_CS
, __KERNEL_CS
, 0);
201 /*H:130 Our Guest is usually so well behaved; it never tries to do things it
202 * isn't allowed to. Unfortunately, Linux's paravirtual infrastructure isn't
203 * quite complete, because it doesn't contain replacements for the Intel I/O
204 * instructions. As a result, the Guest sometimes fumbles across one during
205 * the boot process as it probes for various things which are usually attached
208 * When the Guest uses one of these instructions, we get trap #13 (General
209 * Protection Fault) and come here. We see if it's one of those troublesome
210 * instructions and skip over it. We return true if we did. */
211 static int emulate_insn(struct lguest
*lg
)
214 unsigned int insnlen
= 0, in
= 0, shift
= 0;
215 /* The eip contains the *virtual* address of the Guest's instruction:
216 * guest_pa just subtracts the Guest's page_offset. */
217 unsigned long physaddr
= guest_pa(lg
, lg
->regs
->eip
);
219 /* This must be the Guest kernel trying to do something, not userspace!
220 * The bottom two bits of the CS segment register are the privilege
222 if ((lg
->regs
->cs
& 3) != GUEST_PL
)
225 /* Decoding x86 instructions is icky. */
226 lgread(lg
, &insn
, physaddr
, 1);
228 /* 0x66 is an "operand prefix". It means it's using the upper 16 bits
229 of the eax register. */
232 /* The instruction is 1 byte so far, read the next byte. */
234 lgread(lg
, &insn
, physaddr
+ insnlen
, 1);
237 /* We can ignore the lower bit for the moment and decode the 4 opcodes
238 * we need to emulate. */
239 switch (insn
& 0xFE) {
240 case 0xE4: /* in <next byte>,%al */
244 case 0xEC: /* in (%dx),%al */
248 case 0xE6: /* out %al,<next byte> */
251 case 0xEE: /* out %al,(%dx) */
255 /* OK, we don't know what this is, can't emulate. */
259 /* If it was an "IN" instruction, they expect the result to be read
260 * into %eax, so we change %eax. We always return all-ones, which
261 * traditionally means "there's nothing there". */
263 /* Lower bit tells is whether it's a 16 or 32 bit access */
265 lg
->regs
->eax
= 0xFFFFFFFF;
267 lg
->regs
->eax
|= (0xFFFF << shift
);
269 /* Finally, we've "done" the instruction, so move past it. */
270 lg
->regs
->eip
+= insnlen
;
275 /*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
276 void lguest_arch_handle_trap(struct lguest
*lg
)
278 switch (lg
->regs
->trapnum
) {
279 case 13: /* We've intercepted a GPF. */
280 /* Check if this was one of those annoying IN or OUT
281 * instructions which we need to emulate. If so, we
282 * just go back into the Guest after we've done it. */
283 if (lg
->regs
->errcode
== 0) {
284 if (emulate_insn(lg
))
288 case 14: /* We've intercepted a page fault. */
289 /* The Guest accessed a virtual address that wasn't
290 * mapped. This happens a lot: we don't actually set
291 * up most of the page tables for the Guest at all when
292 * we start: as it runs it asks for more and more, and
293 * we set them up as required. In this case, we don't
294 * even tell the Guest that the fault happened.
296 * The errcode tells whether this was a read or a
297 * write, and whether kernel or userspace code. */
298 if (demand_page(lg
, lg
->arch
.last_pagefault
, lg
->regs
->errcode
))
301 /* OK, it's really not there (or not OK): the Guest
302 * needs to know. We write out the cr2 value so it
303 * knows where the fault occurred.
305 * Note that if the Guest were really messed up, this
306 * could happen before it's done the INITIALIZE
307 * hypercall, so lg->lguest_data will be NULL */
308 if (lg
->lguest_data
&&
309 put_user(lg
->arch
.last_pagefault
, &lg
->lguest_data
->cr2
))
310 kill_guest(lg
, "Writing cr2");
312 case 7: /* We've intercepted a Device Not Available fault. */
313 /* If the Guest doesn't want to know, we already
314 * restored the Floating Point Unit, so we just
315 * continue without telling it. */
320 /* These values mean a real interrupt occurred, in which case
321 * the Host handler has already been run. We just do a
322 * friendly check if another process should now be run, then
323 * return to run the Guest again */
326 case LGUEST_TRAP_ENTRY
:
327 /* Our 'struct hcall_args' maps directly over our regs: we set
328 * up the pointer now to indicate a hypercall is pending. */
329 lg
->hcall
= (struct hcall_args
*)lg
->regs
;
333 /* We didn't handle the trap, so it needs to go to the Guest. */
334 if (!deliver_trap(lg
, lg
->regs
->trapnum
))
335 /* If the Guest doesn't have a handler (either it hasn't
336 * registered any yet, or it's one of the faults we don't let
337 * it handle), it dies with a cryptic error message. */
338 kill_guest(lg
, "unhandled trap %li at %#lx (%#lx)",
339 lg
->regs
->trapnum
, lg
->regs
->eip
,
340 lg
->regs
->trapnum
== 14 ? lg
->arch
.last_pagefault
341 : lg
->regs
->errcode
);
344 /* Now we can look at each of the routines this calls, in increasing order of
345 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
346 * deliver_trap() and demand_page(). After all those, we'll be ready to
347 * examine the Switcher, and our philosophical understanding of the Host/Guest
348 * duality will be complete. :*/
349 static void adjust_pge(void *on
)
352 write_cr4(read_cr4() | X86_CR4_PGE
);
354 write_cr4(read_cr4() & ~X86_CR4_PGE
);
357 /*H:020 Now the Switcher is mapped and every thing else is ready, we need to do
358 * some more i386-specific initialization. */
359 void __init
lguest_arch_host_init(void)
363 /* Most of the i386/switcher.S doesn't care that it's been moved; on
364 * Intel, jumps are relative, and it doesn't access any references to
365 * external code or data.
367 * The only exception is the interrupt handlers in switcher.S: their
368 * addresses are placed in a table (default_idt_entries), so we need to
369 * update the table with the new addresses. switcher_offset() is a
370 * convenience function which returns the distance between the builtin
371 * switcher code and the high-mapped copy we just made. */
372 for (i
= 0; i
< IDT_ENTRIES
; i
++)
373 default_idt_entries
[i
] += switcher_offset();
376 * Set up the Switcher's per-cpu areas.
378 * Each CPU gets two pages of its own within the high-mapped region
379 * (aka. "struct lguest_pages"). Much of this can be initialized now,
380 * but some depends on what Guest we are running (which is set up in
381 * copy_in_guest_info()).
383 for_each_possible_cpu(i
) {
384 /* lguest_pages() returns this CPU's two pages. */
385 struct lguest_pages
*pages
= lguest_pages(i
);
386 /* This is a convenience pointer to make the code fit one
387 * statement to a line. */
388 struct lguest_ro_state
*state
= &pages
->state
;
390 /* The Global Descriptor Table: the Host has a different one
391 * for each CPU. We keep a descriptor for the GDT which says
392 * where it is and how big it is (the size is actually the last
393 * byte, not the size, hence the "-1"). */
394 state
->host_gdt_desc
.size
= GDT_SIZE
-1;
395 state
->host_gdt_desc
.address
= (long)get_cpu_gdt_table(i
);
397 /* All CPUs on the Host use the same Interrupt Descriptor
398 * Table, so we just use store_idt(), which gets this CPU's IDT
400 store_idt(&state
->host_idt_desc
);
402 /* The descriptors for the Guest's GDT and IDT can be filled
403 * out now, too. We copy the GDT & IDT into ->guest_gdt and
404 * ->guest_idt before actually running the Guest. */
405 state
->guest_idt_desc
.size
= sizeof(state
->guest_idt
)-1;
406 state
->guest_idt_desc
.address
= (long)&state
->guest_idt
;
407 state
->guest_gdt_desc
.size
= sizeof(state
->guest_gdt
)-1;
408 state
->guest_gdt_desc
.address
= (long)&state
->guest_gdt
;
410 /* We know where we want the stack to be when the Guest enters
411 * the switcher: in pages->regs. The stack grows upwards, so
412 * we start it at the end of that structure. */
413 state
->guest_tss
.esp0
= (long)(&pages
->regs
+ 1);
414 /* And this is the GDT entry to use for the stack: we keep a
415 * couple of special LGUEST entries. */
416 state
->guest_tss
.ss0
= LGUEST_DS
;
418 /* x86 can have a finegrained bitmap which indicates what I/O
419 * ports the process can use. We set it to the end of our
420 * structure, meaning "none". */
421 state
->guest_tss
.io_bitmap_base
= sizeof(state
->guest_tss
);
423 /* Some GDT entries are the same across all Guests, so we can
424 * set them up now. */
425 setup_default_gdt_entries(state
);
426 /* Most IDT entries are the same for all Guests, too.*/
427 setup_default_idt_entries(state
, default_idt_entries
);
429 /* The Host needs to be able to use the LGUEST segments on this
430 * CPU, too, so put them in the Host GDT. */
431 get_cpu_gdt_table(i
)[GDT_ENTRY_LGUEST_CS
] = FULL_EXEC_SEGMENT
;
432 get_cpu_gdt_table(i
)[GDT_ENTRY_LGUEST_DS
] = FULL_SEGMENT
;
435 /* In the Switcher, we want the %cs segment register to use the
436 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
437 * it will be undisturbed when we switch. To change %cs and jump we
438 * need this structure to feed to Intel's "lcall" instruction. */
439 lguest_entry
.offset
= (long)switch_to_guest
+ switcher_offset();
440 lguest_entry
.segment
= LGUEST_CS
;
442 /* Finally, we need to turn off "Page Global Enable". PGE is an
443 * optimization where page table entries are specially marked to show
444 * they never change. The Host kernel marks all the kernel pages this
445 * way because it's always present, even when userspace is running.
447 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
448 * switch to the Guest kernel. If you don't disable this on all CPUs,
449 * you'll get really weird bugs that you'll chase for two days.
451 * I used to turn PGE off every time we switched to the Guest and back
452 * on when we return, but that slowed the Switcher down noticibly. */
454 /* We don't need the complexity of CPUs coming and going while we're
457 if (cpu_has_pge
) { /* We have a broader idea of "global". */
458 /* Remember that this was originally set (for cleanup). */
460 /* adjust_pge is a helper function which sets or unsets the PGE
461 * bit on its CPU, depending on the argument (0 == unset). */
462 on_each_cpu(adjust_pge
, (void *)0, 0, 1);
463 /* Turn off the feature in the global feature set. */
464 clear_bit(X86_FEATURE_PGE
, boot_cpu_data
.x86_capability
);
466 unlock_cpu_hotplug();
470 void __exit
lguest_arch_host_fini(void)
472 /* If we had PGE before we started, turn it back on now. */
475 set_bit(X86_FEATURE_PGE
, boot_cpu_data
.x86_capability
);
476 /* adjust_pge's argument "1" means set PGE. */
477 on_each_cpu(adjust_pge
, (void *)1, 0, 1);
479 unlock_cpu_hotplug();
483 /*H:122 The i386-specific hypercalls simply farm out to the right functions. */
484 int lguest_arch_do_hcall(struct lguest
*lg
, struct hcall_args
*args
)
486 switch (args
->arg0
) {
487 case LHCALL_LOAD_GDT
:
488 load_guest_gdt(lg
, args
->arg1
, args
->arg2
);
490 case LHCALL_LOAD_IDT_ENTRY
:
491 load_guest_idt_entry(lg
, args
->arg1
, args
->arg2
, args
->arg3
);
493 case LHCALL_LOAD_TLS
:
494 guest_load_tls(lg
, args
->arg1
);
497 /* Bad Guest. Bad! */
503 /*H:126 i386-specific hypercall initialization: */
504 int lguest_arch_init_hypercalls(struct lguest
*lg
)
508 /* The pointer to the Guest's "struct lguest_data" is the only
509 * argument. We check that address now. */
510 if (!lguest_address_ok(lg
, lg
->hcall
->arg1
, sizeof(*lg
->lguest_data
)))
513 /* Having checked it, we simply set lg->lguest_data to point straight
514 * into the Launcher's memory at the right place and then use
515 * copy_to_user/from_user from now on, instead of lgread/write. I put
516 * this in to show that I'm not immune to writing stupid
518 lg
->lguest_data
= lg
->mem_base
+ lg
->hcall
->arg1
;
520 /* We insist that the Time Stamp Counter exist and doesn't change with
521 * cpu frequency. Some devious chip manufacturers decided that TSC
522 * changes could be handled in software. I decided that time going
523 * backwards might be good for benchmarks, but it's bad for users.
525 * We also insist that the TSC be stable: the kernel detects unreliable
526 * TSCs for its own purposes, and we use that here. */
527 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC
) && !check_tsc_unstable())
531 if (put_user(tsc_speed
, &lg
->lguest_data
->tsc_khz
))
534 /* The interrupt code might not like the system call vector. */
535 if (!check_syscall_vector(lg
))
536 kill_guest(lg
, "bad syscall vector");
540 /* Now we've examined the hypercall code; our Guest can make requests. There
541 * is one other way we can do things for the Guest, as we see in
542 * emulate_insn(). :*/
544 /*L:030 lguest_arch_setup_regs()
546 * Most of the Guest's registers are left alone: we used get_zeroed_page() to
547 * allocate the structure, so they will be 0. */
548 void lguest_arch_setup_regs(struct lguest
*lg
, unsigned long start
)
550 struct lguest_regs
*regs
= lg
->regs
;
552 /* There are four "segment" registers which the Guest needs to boot:
553 * The "code segment" register (cs) refers to the kernel code segment
554 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
555 * refer to the kernel data segment __KERNEL_DS.
557 * The privilege level is packed into the lower bits. The Guest runs
558 * at privilege level 1 (GUEST_PL).*/
559 regs
->ds
= regs
->es
= regs
->ss
= __KERNEL_DS
|GUEST_PL
;
560 regs
->cs
= __KERNEL_CS
|GUEST_PL
;
562 /* The "eflags" register contains miscellaneous flags. Bit 1 (0x002)
563 * is supposed to always be "1". Bit 9 (0x200) controls whether
564 * interrupts are enabled. We always leave interrupts enabled while
565 * running the Guest. */
566 regs
->eflags
= 0x202;
568 /* The "Extended Instruction Pointer" register says where the Guest is
572 /* %esi points to our boot information, at physical address 0, so don't
574 /* There are a couple of GDT entries the Guest expects when first