Add linux-next specific files for 20110831
[linux-2.6/next.git] / arch / powerpc / kernel / kvm.c
blob35f27646c4ff5af053313054c953000d0be64031
1 /*
2 * Copyright (C) 2010 SUSE Linux Products GmbH. All rights reserved.
4 * Authors:
5 * Alexander Graf <agraf@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2, as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
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, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
21 #include <linux/kvm_host.h>
22 #include <linux/init.h>
23 #include <linux/export.h>
24 #include <linux/kvm_para.h>
25 #include <linux/slab.h>
26 #include <linux/of.h>
28 #include <asm/reg.h>
29 #include <asm/sections.h>
30 #include <asm/cacheflush.h>
31 #include <asm/disassemble.h>
33 #define KVM_MAGIC_PAGE (-4096L)
34 #define magic_var(x) KVM_MAGIC_PAGE + offsetof(struct kvm_vcpu_arch_shared, x)
36 #define KVM_INST_LWZ 0x80000000
37 #define KVM_INST_STW 0x90000000
38 #define KVM_INST_LD 0xe8000000
39 #define KVM_INST_STD 0xf8000000
40 #define KVM_INST_NOP 0x60000000
41 #define KVM_INST_B 0x48000000
42 #define KVM_INST_B_MASK 0x03ffffff
43 #define KVM_INST_B_MAX 0x01ffffff
45 #define KVM_MASK_RT 0x03e00000
46 #define KVM_RT_30 0x03c00000
47 #define KVM_MASK_RB 0x0000f800
48 #define KVM_INST_MFMSR 0x7c0000a6
49 #define KVM_INST_MFSPR_SPRG0 0x7c1042a6
50 #define KVM_INST_MFSPR_SPRG1 0x7c1142a6
51 #define KVM_INST_MFSPR_SPRG2 0x7c1242a6
52 #define KVM_INST_MFSPR_SPRG3 0x7c1342a6
53 #define KVM_INST_MFSPR_SRR0 0x7c1a02a6
54 #define KVM_INST_MFSPR_SRR1 0x7c1b02a6
55 #define KVM_INST_MFSPR_DAR 0x7c1302a6
56 #define KVM_INST_MFSPR_DSISR 0x7c1202a6
58 #define KVM_INST_MTSPR_SPRG0 0x7c1043a6
59 #define KVM_INST_MTSPR_SPRG1 0x7c1143a6
60 #define KVM_INST_MTSPR_SPRG2 0x7c1243a6
61 #define KVM_INST_MTSPR_SPRG3 0x7c1343a6
62 #define KVM_INST_MTSPR_SRR0 0x7c1a03a6
63 #define KVM_INST_MTSPR_SRR1 0x7c1b03a6
64 #define KVM_INST_MTSPR_DAR 0x7c1303a6
65 #define KVM_INST_MTSPR_DSISR 0x7c1203a6
67 #define KVM_INST_TLBSYNC 0x7c00046c
68 #define KVM_INST_MTMSRD_L0 0x7c000164
69 #define KVM_INST_MTMSRD_L1 0x7c010164
70 #define KVM_INST_MTMSR 0x7c000124
72 #define KVM_INST_WRTEEI_0 0x7c000146
73 #define KVM_INST_WRTEEI_1 0x7c008146
75 #define KVM_INST_MTSRIN 0x7c0001e4
77 static bool kvm_patching_worked = true;
78 static char kvm_tmp[1024 * 1024];
79 static int kvm_tmp_index;
81 static inline void kvm_patch_ins(u32 *inst, u32 new_inst)
83 *inst = new_inst;
84 flush_icache_range((ulong)inst, (ulong)inst + 4);
87 static void kvm_patch_ins_ll(u32 *inst, long addr, u32 rt)
89 #ifdef CONFIG_64BIT
90 kvm_patch_ins(inst, KVM_INST_LD | rt | (addr & 0x0000fffc));
91 #else
92 kvm_patch_ins(inst, KVM_INST_LWZ | rt | (addr & 0x0000fffc));
93 #endif
96 static void kvm_patch_ins_ld(u32 *inst, long addr, u32 rt)
98 #ifdef CONFIG_64BIT
99 kvm_patch_ins(inst, KVM_INST_LD | rt | (addr & 0x0000fffc));
100 #else
101 kvm_patch_ins(inst, KVM_INST_LWZ | rt | ((addr + 4) & 0x0000fffc));
102 #endif
105 static void kvm_patch_ins_lwz(u32 *inst, long addr, u32 rt)
107 kvm_patch_ins(inst, KVM_INST_LWZ | rt | (addr & 0x0000ffff));
110 static void kvm_patch_ins_std(u32 *inst, long addr, u32 rt)
112 #ifdef CONFIG_64BIT
113 kvm_patch_ins(inst, KVM_INST_STD | rt | (addr & 0x0000fffc));
114 #else
115 kvm_patch_ins(inst, KVM_INST_STW | rt | ((addr + 4) & 0x0000fffc));
116 #endif
119 static void kvm_patch_ins_stw(u32 *inst, long addr, u32 rt)
121 kvm_patch_ins(inst, KVM_INST_STW | rt | (addr & 0x0000fffc));
124 static void kvm_patch_ins_nop(u32 *inst)
126 kvm_patch_ins(inst, KVM_INST_NOP);
129 static void kvm_patch_ins_b(u32 *inst, int addr)
131 #if defined(CONFIG_RELOCATABLE) && defined(CONFIG_PPC_BOOK3S)
132 /* On relocatable kernels interrupts handlers and our code
133 can be in different regions, so we don't patch them */
135 extern u32 __end_interrupts;
136 if ((ulong)inst < (ulong)&__end_interrupts)
137 return;
138 #endif
140 kvm_patch_ins(inst, KVM_INST_B | (addr & KVM_INST_B_MASK));
143 static u32 *kvm_alloc(int len)
145 u32 *p;
147 if ((kvm_tmp_index + len) > ARRAY_SIZE(kvm_tmp)) {
148 printk(KERN_ERR "KVM: No more space (%d + %d)\n",
149 kvm_tmp_index, len);
150 kvm_patching_worked = false;
151 return NULL;
154 p = (void*)&kvm_tmp[kvm_tmp_index];
155 kvm_tmp_index += len;
157 return p;
160 extern u32 kvm_emulate_mtmsrd_branch_offs;
161 extern u32 kvm_emulate_mtmsrd_reg_offs;
162 extern u32 kvm_emulate_mtmsrd_orig_ins_offs;
163 extern u32 kvm_emulate_mtmsrd_len;
164 extern u32 kvm_emulate_mtmsrd[];
166 static void kvm_patch_ins_mtmsrd(u32 *inst, u32 rt)
168 u32 *p;
169 int distance_start;
170 int distance_end;
171 ulong next_inst;
173 p = kvm_alloc(kvm_emulate_mtmsrd_len * 4);
174 if (!p)
175 return;
177 /* Find out where we are and put everything there */
178 distance_start = (ulong)p - (ulong)inst;
179 next_inst = ((ulong)inst + 4);
180 distance_end = next_inst - (ulong)&p[kvm_emulate_mtmsrd_branch_offs];
182 /* Make sure we only write valid b instructions */
183 if (distance_start > KVM_INST_B_MAX) {
184 kvm_patching_worked = false;
185 return;
188 /* Modify the chunk to fit the invocation */
189 memcpy(p, kvm_emulate_mtmsrd, kvm_emulate_mtmsrd_len * 4);
190 p[kvm_emulate_mtmsrd_branch_offs] |= distance_end & KVM_INST_B_MASK;
191 switch (get_rt(rt)) {
192 case 30:
193 kvm_patch_ins_ll(&p[kvm_emulate_mtmsrd_reg_offs],
194 magic_var(scratch2), KVM_RT_30);
195 break;
196 case 31:
197 kvm_patch_ins_ll(&p[kvm_emulate_mtmsrd_reg_offs],
198 magic_var(scratch1), KVM_RT_30);
199 break;
200 default:
201 p[kvm_emulate_mtmsrd_reg_offs] |= rt;
202 break;
205 p[kvm_emulate_mtmsrd_orig_ins_offs] = *inst;
206 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_mtmsrd_len * 4);
208 /* Patch the invocation */
209 kvm_patch_ins_b(inst, distance_start);
212 extern u32 kvm_emulate_mtmsr_branch_offs;
213 extern u32 kvm_emulate_mtmsr_reg1_offs;
214 extern u32 kvm_emulate_mtmsr_reg2_offs;
215 extern u32 kvm_emulate_mtmsr_orig_ins_offs;
216 extern u32 kvm_emulate_mtmsr_len;
217 extern u32 kvm_emulate_mtmsr[];
219 static void kvm_patch_ins_mtmsr(u32 *inst, u32 rt)
221 u32 *p;
222 int distance_start;
223 int distance_end;
224 ulong next_inst;
226 p = kvm_alloc(kvm_emulate_mtmsr_len * 4);
227 if (!p)
228 return;
230 /* Find out where we are and put everything there */
231 distance_start = (ulong)p - (ulong)inst;
232 next_inst = ((ulong)inst + 4);
233 distance_end = next_inst - (ulong)&p[kvm_emulate_mtmsr_branch_offs];
235 /* Make sure we only write valid b instructions */
236 if (distance_start > KVM_INST_B_MAX) {
237 kvm_patching_worked = false;
238 return;
241 /* Modify the chunk to fit the invocation */
242 memcpy(p, kvm_emulate_mtmsr, kvm_emulate_mtmsr_len * 4);
243 p[kvm_emulate_mtmsr_branch_offs] |= distance_end & KVM_INST_B_MASK;
245 /* Make clobbered registers work too */
246 switch (get_rt(rt)) {
247 case 30:
248 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg1_offs],
249 magic_var(scratch2), KVM_RT_30);
250 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg2_offs],
251 magic_var(scratch2), KVM_RT_30);
252 break;
253 case 31:
254 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg1_offs],
255 magic_var(scratch1), KVM_RT_30);
256 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg2_offs],
257 magic_var(scratch1), KVM_RT_30);
258 break;
259 default:
260 p[kvm_emulate_mtmsr_reg1_offs] |= rt;
261 p[kvm_emulate_mtmsr_reg2_offs] |= rt;
262 break;
265 p[kvm_emulate_mtmsr_orig_ins_offs] = *inst;
266 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_mtmsr_len * 4);
268 /* Patch the invocation */
269 kvm_patch_ins_b(inst, distance_start);
272 #ifdef CONFIG_BOOKE
274 extern u32 kvm_emulate_wrteei_branch_offs;
275 extern u32 kvm_emulate_wrteei_ee_offs;
276 extern u32 kvm_emulate_wrteei_len;
277 extern u32 kvm_emulate_wrteei[];
279 static void kvm_patch_ins_wrteei(u32 *inst)
281 u32 *p;
282 int distance_start;
283 int distance_end;
284 ulong next_inst;
286 p = kvm_alloc(kvm_emulate_wrteei_len * 4);
287 if (!p)
288 return;
290 /* Find out where we are and put everything there */
291 distance_start = (ulong)p - (ulong)inst;
292 next_inst = ((ulong)inst + 4);
293 distance_end = next_inst - (ulong)&p[kvm_emulate_wrteei_branch_offs];
295 /* Make sure we only write valid b instructions */
296 if (distance_start > KVM_INST_B_MAX) {
297 kvm_patching_worked = false;
298 return;
301 /* Modify the chunk to fit the invocation */
302 memcpy(p, kvm_emulate_wrteei, kvm_emulate_wrteei_len * 4);
303 p[kvm_emulate_wrteei_branch_offs] |= distance_end & KVM_INST_B_MASK;
304 p[kvm_emulate_wrteei_ee_offs] |= (*inst & MSR_EE);
305 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_wrteei_len * 4);
307 /* Patch the invocation */
308 kvm_patch_ins_b(inst, distance_start);
311 #endif
313 #ifdef CONFIG_PPC_BOOK3S_32
315 extern u32 kvm_emulate_mtsrin_branch_offs;
316 extern u32 kvm_emulate_mtsrin_reg1_offs;
317 extern u32 kvm_emulate_mtsrin_reg2_offs;
318 extern u32 kvm_emulate_mtsrin_orig_ins_offs;
319 extern u32 kvm_emulate_mtsrin_len;
320 extern u32 kvm_emulate_mtsrin[];
322 static void kvm_patch_ins_mtsrin(u32 *inst, u32 rt, u32 rb)
324 u32 *p;
325 int distance_start;
326 int distance_end;
327 ulong next_inst;
329 p = kvm_alloc(kvm_emulate_mtsrin_len * 4);
330 if (!p)
331 return;
333 /* Find out where we are and put everything there */
334 distance_start = (ulong)p - (ulong)inst;
335 next_inst = ((ulong)inst + 4);
336 distance_end = next_inst - (ulong)&p[kvm_emulate_mtsrin_branch_offs];
338 /* Make sure we only write valid b instructions */
339 if (distance_start > KVM_INST_B_MAX) {
340 kvm_patching_worked = false;
341 return;
344 /* Modify the chunk to fit the invocation */
345 memcpy(p, kvm_emulate_mtsrin, kvm_emulate_mtsrin_len * 4);
346 p[kvm_emulate_mtsrin_branch_offs] |= distance_end & KVM_INST_B_MASK;
347 p[kvm_emulate_mtsrin_reg1_offs] |= (rb << 10);
348 p[kvm_emulate_mtsrin_reg2_offs] |= rt;
349 p[kvm_emulate_mtsrin_orig_ins_offs] = *inst;
350 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_mtsrin_len * 4);
352 /* Patch the invocation */
353 kvm_patch_ins_b(inst, distance_start);
356 #endif
358 static void kvm_map_magic_page(void *data)
360 u32 *features = data;
362 ulong in[8];
363 ulong out[8];
365 in[0] = KVM_MAGIC_PAGE;
366 in[1] = KVM_MAGIC_PAGE;
368 kvm_hypercall(in, out, HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE);
370 *features = out[0];
373 static void kvm_check_ins(u32 *inst, u32 features)
375 u32 _inst = *inst;
376 u32 inst_no_rt = _inst & ~KVM_MASK_RT;
377 u32 inst_rt = _inst & KVM_MASK_RT;
379 switch (inst_no_rt) {
380 /* Loads */
381 case KVM_INST_MFMSR:
382 kvm_patch_ins_ld(inst, magic_var(msr), inst_rt);
383 break;
384 case KVM_INST_MFSPR_SPRG0:
385 kvm_patch_ins_ld(inst, magic_var(sprg0), inst_rt);
386 break;
387 case KVM_INST_MFSPR_SPRG1:
388 kvm_patch_ins_ld(inst, magic_var(sprg1), inst_rt);
389 break;
390 case KVM_INST_MFSPR_SPRG2:
391 kvm_patch_ins_ld(inst, magic_var(sprg2), inst_rt);
392 break;
393 case KVM_INST_MFSPR_SPRG3:
394 kvm_patch_ins_ld(inst, magic_var(sprg3), inst_rt);
395 break;
396 case KVM_INST_MFSPR_SRR0:
397 kvm_patch_ins_ld(inst, magic_var(srr0), inst_rt);
398 break;
399 case KVM_INST_MFSPR_SRR1:
400 kvm_patch_ins_ld(inst, magic_var(srr1), inst_rt);
401 break;
402 case KVM_INST_MFSPR_DAR:
403 kvm_patch_ins_ld(inst, magic_var(dar), inst_rt);
404 break;
405 case KVM_INST_MFSPR_DSISR:
406 kvm_patch_ins_lwz(inst, magic_var(dsisr), inst_rt);
407 break;
409 /* Stores */
410 case KVM_INST_MTSPR_SPRG0:
411 kvm_patch_ins_std(inst, magic_var(sprg0), inst_rt);
412 break;
413 case KVM_INST_MTSPR_SPRG1:
414 kvm_patch_ins_std(inst, magic_var(sprg1), inst_rt);
415 break;
416 case KVM_INST_MTSPR_SPRG2:
417 kvm_patch_ins_std(inst, magic_var(sprg2), inst_rt);
418 break;
419 case KVM_INST_MTSPR_SPRG3:
420 kvm_patch_ins_std(inst, magic_var(sprg3), inst_rt);
421 break;
422 case KVM_INST_MTSPR_SRR0:
423 kvm_patch_ins_std(inst, magic_var(srr0), inst_rt);
424 break;
425 case KVM_INST_MTSPR_SRR1:
426 kvm_patch_ins_std(inst, magic_var(srr1), inst_rt);
427 break;
428 case KVM_INST_MTSPR_DAR:
429 kvm_patch_ins_std(inst, magic_var(dar), inst_rt);
430 break;
431 case KVM_INST_MTSPR_DSISR:
432 kvm_patch_ins_stw(inst, magic_var(dsisr), inst_rt);
433 break;
435 /* Nops */
436 case KVM_INST_TLBSYNC:
437 kvm_patch_ins_nop(inst);
438 break;
440 /* Rewrites */
441 case KVM_INST_MTMSRD_L1:
442 kvm_patch_ins_mtmsrd(inst, inst_rt);
443 break;
444 case KVM_INST_MTMSR:
445 case KVM_INST_MTMSRD_L0:
446 kvm_patch_ins_mtmsr(inst, inst_rt);
447 break;
450 switch (inst_no_rt & ~KVM_MASK_RB) {
451 #ifdef CONFIG_PPC_BOOK3S_32
452 case KVM_INST_MTSRIN:
453 if (features & KVM_MAGIC_FEAT_SR) {
454 u32 inst_rb = _inst & KVM_MASK_RB;
455 kvm_patch_ins_mtsrin(inst, inst_rt, inst_rb);
457 break;
458 break;
459 #endif
462 switch (_inst) {
463 #ifdef CONFIG_BOOKE
464 case KVM_INST_WRTEEI_0:
465 case KVM_INST_WRTEEI_1:
466 kvm_patch_ins_wrteei(inst);
467 break;
468 #endif
472 static void kvm_use_magic_page(void)
474 u32 *p;
475 u32 *start, *end;
476 u32 tmp;
477 u32 features;
479 /* Tell the host to map the magic page to -4096 on all CPUs */
480 on_each_cpu(kvm_map_magic_page, &features, 1);
482 /* Quick self-test to see if the mapping works */
483 if (__get_user(tmp, (u32*)KVM_MAGIC_PAGE)) {
484 kvm_patching_worked = false;
485 return;
488 /* Now loop through all code and find instructions */
489 start = (void*)_stext;
490 end = (void*)_etext;
492 for (p = start; p < end; p++)
493 kvm_check_ins(p, features);
495 printk(KERN_INFO "KVM: Live patching for a fast VM %s\n",
496 kvm_patching_worked ? "worked" : "failed");
499 unsigned long kvm_hypercall(unsigned long *in,
500 unsigned long *out,
501 unsigned long nr)
503 unsigned long register r0 asm("r0");
504 unsigned long register r3 asm("r3") = in[0];
505 unsigned long register r4 asm("r4") = in[1];
506 unsigned long register r5 asm("r5") = in[2];
507 unsigned long register r6 asm("r6") = in[3];
508 unsigned long register r7 asm("r7") = in[4];
509 unsigned long register r8 asm("r8") = in[5];
510 unsigned long register r9 asm("r9") = in[6];
511 unsigned long register r10 asm("r10") = in[7];
512 unsigned long register r11 asm("r11") = nr;
513 unsigned long register r12 asm("r12");
515 asm volatile("bl kvm_hypercall_start"
516 : "=r"(r0), "=r"(r3), "=r"(r4), "=r"(r5), "=r"(r6),
517 "=r"(r7), "=r"(r8), "=r"(r9), "=r"(r10), "=r"(r11),
518 "=r"(r12)
519 : "r"(r3), "r"(r4), "r"(r5), "r"(r6), "r"(r7), "r"(r8),
520 "r"(r9), "r"(r10), "r"(r11)
521 : "memory", "cc", "xer", "ctr", "lr");
523 out[0] = r4;
524 out[1] = r5;
525 out[2] = r6;
526 out[3] = r7;
527 out[4] = r8;
528 out[5] = r9;
529 out[6] = r10;
530 out[7] = r11;
532 return r3;
534 EXPORT_SYMBOL_GPL(kvm_hypercall);
536 static int kvm_para_setup(void)
538 extern u32 kvm_hypercall_start;
539 struct device_node *hyper_node;
540 u32 *insts;
541 int len, i;
543 hyper_node = of_find_node_by_path("/hypervisor");
544 if (!hyper_node)
545 return -1;
547 insts = (u32*)of_get_property(hyper_node, "hcall-instructions", &len);
548 if (len % 4)
549 return -1;
550 if (len > (4 * 4))
551 return -1;
553 for (i = 0; i < (len / 4); i++)
554 kvm_patch_ins(&(&kvm_hypercall_start)[i], insts[i]);
556 return 0;
559 static __init void kvm_free_tmp(void)
561 unsigned long start, end;
563 start = (ulong)&kvm_tmp[kvm_tmp_index + (PAGE_SIZE - 1)] & PAGE_MASK;
564 end = (ulong)&kvm_tmp[ARRAY_SIZE(kvm_tmp)] & PAGE_MASK;
566 /* Free the tmp space we don't need */
567 for (; start < end; start += PAGE_SIZE) {
568 ClearPageReserved(virt_to_page(start));
569 init_page_count(virt_to_page(start));
570 free_page(start);
571 totalram_pages++;
575 static int __init kvm_guest_init(void)
577 if (!kvm_para_available())
578 goto free_tmp;
580 if (kvm_para_setup())
581 goto free_tmp;
583 if (kvm_para_has_feature(KVM_FEATURE_MAGIC_PAGE))
584 kvm_use_magic_page();
586 #ifdef CONFIG_PPC_BOOK3S_64
587 /* Enable napping */
588 powersave_nap = 1;
589 #endif
591 free_tmp:
592 kvm_free_tmp();
594 return 0;
597 postcore_initcall(kvm_guest_init);