xref: /openbmc/qemu/target/s390x/mmu_helper.c (revision 2df1eb27)
1 /*
2  * S390x MMU related functions
3  *
4  * Copyright (c) 2011 Alexander Graf
5  * Copyright (c) 2015 Thomas Huth, IBM Corporation
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  */
17 
18 #include "qemu/osdep.h"
19 #include "qemu/error-report.h"
20 #include "exec/address-spaces.h"
21 #include "cpu.h"
22 #include "s390x-internal.h"
23 #include "kvm/kvm_s390x.h"
24 #include "sysemu/kvm.h"
25 #include "sysemu/tcg.h"
26 #include "exec/exec-all.h"
27 #include "trace.h"
28 #include "hw/hw.h"
29 #include "hw/s390x/storage-keys.h"
30 #include "hw/boards.h"
31 
32 /* Fetch/store bits in the translation exception code: */
33 #define FS_READ  0x800
34 #define FS_WRITE 0x400
35 
36 static void trigger_access_exception(CPUS390XState *env, uint32_t type,
37                                      uint64_t tec)
38 {
39     S390CPU *cpu = env_archcpu(env);
40 
41     if (kvm_enabled()) {
42         kvm_s390_access_exception(cpu, type, tec);
43     } else {
44         CPUState *cs = env_cpu(env);
45         if (type != PGM_ADDRESSING) {
46             stq_phys(cs->as, env->psa + offsetof(LowCore, trans_exc_code), tec);
47         }
48         trigger_pgm_exception(env, type);
49     }
50 }
51 
52 /* check whether the address would be proteted by Low-Address Protection */
53 static bool is_low_address(uint64_t addr)
54 {
55     return addr <= 511 || (addr >= 4096 && addr <= 4607);
56 }
57 
58 /* check whether Low-Address Protection is enabled for mmu_translate() */
59 static bool lowprot_enabled(const CPUS390XState *env, uint64_t asc)
60 {
61     if (!(env->cregs[0] & CR0_LOWPROT)) {
62         return false;
63     }
64     if (!(env->psw.mask & PSW_MASK_DAT)) {
65         return true;
66     }
67 
68     /* Check the private-space control bit */
69     switch (asc) {
70     case PSW_ASC_PRIMARY:
71         return !(env->cregs[1] & ASCE_PRIVATE_SPACE);
72     case PSW_ASC_SECONDARY:
73         return !(env->cregs[7] & ASCE_PRIVATE_SPACE);
74     case PSW_ASC_HOME:
75         return !(env->cregs[13] & ASCE_PRIVATE_SPACE);
76     default:
77         /* We don't support access register mode */
78         error_report("unsupported addressing mode");
79         exit(1);
80     }
81 }
82 
83 /**
84  * Translate real address to absolute (= physical)
85  * address by taking care of the prefix mapping.
86  */
87 target_ulong mmu_real2abs(CPUS390XState *env, target_ulong raddr)
88 {
89     if (raddr < 0x2000) {
90         return raddr + env->psa;    /* Map the lowcore. */
91     } else if (raddr >= env->psa && raddr < env->psa + 0x2000) {
92         return raddr - env->psa;    /* Map the 0 page. */
93     }
94     return raddr;
95 }
96 
97 bool mmu_absolute_addr_valid(target_ulong addr, bool is_write)
98 {
99     return address_space_access_valid(&address_space_memory,
100                                       addr & TARGET_PAGE_MASK,
101                                       TARGET_PAGE_SIZE, is_write,
102                                       MEMTXATTRS_UNSPECIFIED);
103 }
104 
105 static inline bool read_table_entry(CPUS390XState *env, hwaddr gaddr,
106                                     uint64_t *entry)
107 {
108     CPUState *cs = env_cpu(env);
109 
110     /*
111      * According to the PoP, these table addresses are "unpredictably real
112      * or absolute". Also, "it is unpredictable whether the address wraps
113      * or an addressing exception is recognized".
114      *
115      * We treat them as absolute addresses and don't wrap them.
116      */
117     if (unlikely(address_space_read(cs->as, gaddr, MEMTXATTRS_UNSPECIFIED,
118                                     entry, sizeof(*entry)) !=
119                  MEMTX_OK)) {
120         return false;
121     }
122     *entry = be64_to_cpu(*entry);
123     return true;
124 }
125 
126 static int mmu_translate_asce(CPUS390XState *env, target_ulong vaddr,
127                               uint64_t asc, uint64_t asce, target_ulong *raddr,
128                               int *flags)
129 {
130     const bool edat1 = (env->cregs[0] & CR0_EDAT) &&
131                        s390_has_feat(S390_FEAT_EDAT);
132     const bool edat2 = edat1 && s390_has_feat(S390_FEAT_EDAT_2);
133     const bool iep = (env->cregs[0] & CR0_IEP) &&
134                      s390_has_feat(S390_FEAT_INSTRUCTION_EXEC_PROT);
135     const int asce_tl = asce & ASCE_TABLE_LENGTH;
136     const int asce_p = asce & ASCE_PRIVATE_SPACE;
137     hwaddr gaddr = asce & ASCE_ORIGIN;
138     uint64_t entry;
139 
140     if (asce & ASCE_REAL_SPACE) {
141         /* direct mapping */
142         *raddr = vaddr;
143         return 0;
144     }
145 
146     switch (asce & ASCE_TYPE_MASK) {
147     case ASCE_TYPE_REGION1:
148         if (VADDR_REGION1_TL(vaddr) > asce_tl) {
149             return PGM_REG_FIRST_TRANS;
150         }
151         gaddr += VADDR_REGION1_TX(vaddr) * 8;
152         break;
153     case ASCE_TYPE_REGION2:
154         if (VADDR_REGION1_TX(vaddr)) {
155             return PGM_ASCE_TYPE;
156         }
157         if (VADDR_REGION2_TL(vaddr) > asce_tl) {
158             return PGM_REG_SEC_TRANS;
159         }
160         gaddr += VADDR_REGION2_TX(vaddr) * 8;
161         break;
162     case ASCE_TYPE_REGION3:
163         if (VADDR_REGION1_TX(vaddr) || VADDR_REGION2_TX(vaddr)) {
164             return PGM_ASCE_TYPE;
165         }
166         if (VADDR_REGION3_TL(vaddr) > asce_tl) {
167             return PGM_REG_THIRD_TRANS;
168         }
169         gaddr += VADDR_REGION3_TX(vaddr) * 8;
170         break;
171     case ASCE_TYPE_SEGMENT:
172         if (VADDR_REGION1_TX(vaddr) || VADDR_REGION2_TX(vaddr) ||
173             VADDR_REGION3_TX(vaddr)) {
174             return PGM_ASCE_TYPE;
175         }
176         if (VADDR_SEGMENT_TL(vaddr) > asce_tl) {
177             return PGM_SEGMENT_TRANS;
178         }
179         gaddr += VADDR_SEGMENT_TX(vaddr) * 8;
180         break;
181     }
182 
183     switch (asce & ASCE_TYPE_MASK) {
184     case ASCE_TYPE_REGION1:
185         if (!read_table_entry(env, gaddr, &entry)) {
186             return PGM_ADDRESSING;
187         }
188         if (entry & REGION_ENTRY_I) {
189             return PGM_REG_FIRST_TRANS;
190         }
191         if ((entry & REGION_ENTRY_TT) != REGION_ENTRY_TT_REGION1) {
192             return PGM_TRANS_SPEC;
193         }
194         if (VADDR_REGION2_TL(vaddr) < (entry & REGION_ENTRY_TF) >> 6 ||
195             VADDR_REGION2_TL(vaddr) > (entry & REGION_ENTRY_TL)) {
196             return PGM_REG_SEC_TRANS;
197         }
198         if (edat1 && (entry & REGION_ENTRY_P)) {
199             *flags &= ~PAGE_WRITE;
200         }
201         gaddr = (entry & REGION_ENTRY_ORIGIN) + VADDR_REGION2_TX(vaddr) * 8;
202         /* fall through */
203     case ASCE_TYPE_REGION2:
204         if (!read_table_entry(env, gaddr, &entry)) {
205             return PGM_ADDRESSING;
206         }
207         if (entry & REGION_ENTRY_I) {
208             return PGM_REG_SEC_TRANS;
209         }
210         if ((entry & REGION_ENTRY_TT) != REGION_ENTRY_TT_REGION2) {
211             return PGM_TRANS_SPEC;
212         }
213         if (VADDR_REGION3_TL(vaddr) < (entry & REGION_ENTRY_TF) >> 6 ||
214             VADDR_REGION3_TL(vaddr) > (entry & REGION_ENTRY_TL)) {
215             return PGM_REG_THIRD_TRANS;
216         }
217         if (edat1 && (entry & REGION_ENTRY_P)) {
218             *flags &= ~PAGE_WRITE;
219         }
220         gaddr = (entry & REGION_ENTRY_ORIGIN) + VADDR_REGION3_TX(vaddr) * 8;
221         /* fall through */
222     case ASCE_TYPE_REGION3:
223         if (!read_table_entry(env, gaddr, &entry)) {
224             return PGM_ADDRESSING;
225         }
226         if (entry & REGION_ENTRY_I) {
227             return PGM_REG_THIRD_TRANS;
228         }
229         if ((entry & REGION_ENTRY_TT) != REGION_ENTRY_TT_REGION3) {
230             return PGM_TRANS_SPEC;
231         }
232         if (edat2 && (entry & REGION3_ENTRY_CR) && asce_p) {
233             return PGM_TRANS_SPEC;
234         }
235         if (edat1 && (entry & REGION_ENTRY_P)) {
236             *flags &= ~PAGE_WRITE;
237         }
238         if (edat2 && (entry & REGION3_ENTRY_FC)) {
239             if (iep && (entry & REGION3_ENTRY_IEP)) {
240                 *flags &= ~PAGE_EXEC;
241             }
242             *raddr = (entry & REGION3_ENTRY_RFAA) |
243                      (vaddr & ~REGION3_ENTRY_RFAA);
244             return 0;
245         }
246         if (VADDR_SEGMENT_TL(vaddr) < (entry & REGION_ENTRY_TF) >> 6 ||
247             VADDR_SEGMENT_TL(vaddr) > (entry & REGION_ENTRY_TL)) {
248             return PGM_SEGMENT_TRANS;
249         }
250         gaddr = (entry & REGION_ENTRY_ORIGIN) + VADDR_SEGMENT_TX(vaddr) * 8;
251         /* fall through */
252     case ASCE_TYPE_SEGMENT:
253         if (!read_table_entry(env, gaddr, &entry)) {
254             return PGM_ADDRESSING;
255         }
256         if (entry & SEGMENT_ENTRY_I) {
257             return PGM_SEGMENT_TRANS;
258         }
259         if ((entry & SEGMENT_ENTRY_TT) != SEGMENT_ENTRY_TT_SEGMENT) {
260             return PGM_TRANS_SPEC;
261         }
262         if ((entry & SEGMENT_ENTRY_CS) && asce_p) {
263             return PGM_TRANS_SPEC;
264         }
265         if (entry & SEGMENT_ENTRY_P) {
266             *flags &= ~PAGE_WRITE;
267         }
268         if (edat1 && (entry & SEGMENT_ENTRY_FC)) {
269             if (iep && (entry & SEGMENT_ENTRY_IEP)) {
270                 *flags &= ~PAGE_EXEC;
271             }
272             *raddr = (entry & SEGMENT_ENTRY_SFAA) |
273                      (vaddr & ~SEGMENT_ENTRY_SFAA);
274             return 0;
275         }
276         gaddr = (entry & SEGMENT_ENTRY_ORIGIN) + VADDR_PAGE_TX(vaddr) * 8;
277         break;
278     }
279 
280     if (!read_table_entry(env, gaddr, &entry)) {
281         return PGM_ADDRESSING;
282     }
283     if (entry & PAGE_ENTRY_I) {
284         return PGM_PAGE_TRANS;
285     }
286     if (entry & PAGE_ENTRY_0) {
287         return PGM_TRANS_SPEC;
288     }
289     if (entry & PAGE_ENTRY_P) {
290         *flags &= ~PAGE_WRITE;
291     }
292     if (iep && (entry & PAGE_ENTRY_IEP)) {
293         *flags &= ~PAGE_EXEC;
294     }
295 
296     *raddr = entry & TARGET_PAGE_MASK;
297     return 0;
298 }
299 
300 static void mmu_handle_skey(target_ulong addr, int rw, int *flags)
301 {
302     static S390SKeysClass *skeyclass;
303     static S390SKeysState *ss;
304     uint8_t key, old_key;
305     int rc;
306 
307     /*
308      * We expect to be called with an absolute address that has already been
309      * validated, such that we can reliably use it to lookup the storage key.
310      */
311     if (unlikely(!ss)) {
312         ss = s390_get_skeys_device();
313         skeyclass = S390_SKEYS_GET_CLASS(ss);
314     }
315 
316     /*
317      * Don't enable storage keys if they are still disabled, i.e., no actual
318      * storage key instruction was issued yet.
319      */
320     if (!skeyclass->skeys_are_enabled(ss)) {
321         return;
322     }
323 
324     /*
325      * Whenever we create a new TLB entry, we set the storage key reference
326      * bit. In case we allow write accesses, we set the storage key change
327      * bit. Whenever the guest changes the storage key, we have to flush the
328      * TLBs of all CPUs (the whole TLB or all affected entries), so that the
329      * next reference/change will result in an MMU fault and make us properly
330      * update the storage key here.
331      *
332      * Note 1: "record of references ... is not necessarily accurate",
333      *         "change bit may be set in case no storing has occurred".
334      *         -> We can set reference/change bits even on exceptions.
335      * Note 2: certain accesses seem to ignore storage keys. For example,
336      *         DAT translation does not set reference bits for table accesses.
337      *
338      * TODO: key-controlled protection. Only CPU accesses make use of the
339      *       PSW key. CSS accesses are different - we have to pass in the key.
340      *
341      * TODO: we have races between getting and setting the key.
342      */
343     rc = skeyclass->get_skeys(ss, addr / TARGET_PAGE_SIZE, 1, &key);
344     if (rc) {
345         trace_get_skeys_nonzero(rc);
346         return;
347     }
348     old_key = key;
349 
350     switch (rw) {
351     case MMU_DATA_LOAD:
352     case MMU_INST_FETCH:
353         /*
354          * The TLB entry has to remain write-protected on read-faults if
355          * the storage key does not indicate a change already. Otherwise
356          * we might miss setting the change bit on write accesses.
357          */
358         if (!(key & SK_C)) {
359             *flags &= ~PAGE_WRITE;
360         }
361         break;
362     case MMU_DATA_STORE:
363         key |= SK_C;
364         break;
365     default:
366         g_assert_not_reached();
367     }
368 
369     /* Any store/fetch sets the reference bit */
370     key |= SK_R;
371 
372     if (key != old_key) {
373         rc = skeyclass->set_skeys(ss, addr / TARGET_PAGE_SIZE, 1, &key);
374         if (rc) {
375             trace_set_skeys_nonzero(rc);
376         }
377     }
378 }
379 
380 /**
381  * Translate a virtual (logical) address into a physical (absolute) address.
382  * @param vaddr  the virtual address
383  * @param rw     0 = read, 1 = write, 2 = code fetch, < 0 = load real address
384  * @param asc    address space control (one of the PSW_ASC_* modes)
385  * @param raddr  the translated address is stored to this pointer
386  * @param flags  the PAGE_READ/WRITE/EXEC flags are stored to this pointer
387  * @param tec    the translation exception code if stored to this pointer if
388  *               there is an exception to raise
389  * @return       0 = success, != 0, the exception to raise
390  */
391 int mmu_translate(CPUS390XState *env, target_ulong vaddr, int rw, uint64_t asc,
392                   target_ulong *raddr, int *flags, uint64_t *tec)
393 {
394     uint64_t asce;
395     int r;
396 
397     *tec = (vaddr & TARGET_PAGE_MASK) | (asc >> 46) |
398             (rw == MMU_DATA_STORE ? FS_WRITE : FS_READ);
399     *flags = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
400 
401     if (is_low_address(vaddr & TARGET_PAGE_MASK) && lowprot_enabled(env, asc)) {
402         /*
403          * If any part of this page is currently protected, make sure the
404          * TLB entry will not be reused.
405          *
406          * As the protected range is always the first 512 bytes of the
407          * two first pages, we are able to catch all writes to these areas
408          * just by looking at the start address (triggering the tlb miss).
409          */
410         *flags |= PAGE_WRITE_INV;
411         if (is_low_address(vaddr) && rw == MMU_DATA_STORE) {
412             /* LAP sets bit 56 */
413             *tec |= 0x80;
414             return PGM_PROTECTION;
415         }
416     }
417 
418     vaddr &= TARGET_PAGE_MASK;
419 
420     if (rw != MMU_S390_LRA && !(env->psw.mask & PSW_MASK_DAT)) {
421         *raddr = vaddr;
422         goto nodat;
423     }
424 
425     switch (asc) {
426     case PSW_ASC_PRIMARY:
427         asce = env->cregs[1];
428         break;
429     case PSW_ASC_HOME:
430         asce = env->cregs[13];
431         break;
432     case PSW_ASC_SECONDARY:
433         asce = env->cregs[7];
434         break;
435     case PSW_ASC_ACCREG:
436     default:
437         hw_error("guest switched to unknown asc mode\n");
438         break;
439     }
440 
441     /* perform the DAT translation */
442     r = mmu_translate_asce(env, vaddr, asc, asce, raddr, flags);
443     if (unlikely(r)) {
444         return r;
445     }
446 
447     /* check for DAT protection */
448     if (unlikely(rw == MMU_DATA_STORE && !(*flags & PAGE_WRITE))) {
449         /* DAT sets bit 61 only */
450         *tec |= 0x4;
451         return PGM_PROTECTION;
452     }
453 
454     /* check for Instruction-Execution-Protection */
455     if (unlikely(rw == MMU_INST_FETCH && !(*flags & PAGE_EXEC))) {
456         /* IEP sets bit 56 and 61 */
457         *tec |= 0x84;
458         return PGM_PROTECTION;
459     }
460 
461 nodat:
462     if (rw >= 0) {
463         /* Convert real address -> absolute address */
464         *raddr = mmu_real2abs(env, *raddr);
465 
466         if (!mmu_absolute_addr_valid(*raddr, rw == MMU_DATA_STORE)) {
467             *tec = 0; /* unused */
468             return PGM_ADDRESSING;
469         }
470 
471         mmu_handle_skey(*raddr, rw, flags);
472     }
473     return 0;
474 }
475 
476 /**
477  * translate_pages: Translate a set of consecutive logical page addresses
478  * to absolute addresses. This function is used for TCG and old KVM without
479  * the MEMOP interface.
480  */
481 static int translate_pages(S390CPU *cpu, vaddr addr, int nr_pages,
482                            target_ulong *pages, bool is_write, uint64_t *tec)
483 {
484     uint64_t asc = cpu->env.psw.mask & PSW_MASK_ASC;
485     CPUS390XState *env = &cpu->env;
486     int ret, i, pflags;
487 
488     for (i = 0; i < nr_pages; i++) {
489         ret = mmu_translate(env, addr, is_write, asc, &pages[i], &pflags, tec);
490         if (ret) {
491             return ret;
492         }
493         addr += TARGET_PAGE_SIZE;
494     }
495 
496     return 0;
497 }
498 
499 int s390_cpu_pv_mem_rw(S390CPU *cpu, unsigned int offset, void *hostbuf,
500                        int len, bool is_write)
501 {
502     int ret;
503 
504     if (kvm_enabled()) {
505         ret = kvm_s390_mem_op_pv(cpu, offset, hostbuf, len, is_write);
506     } else {
507         /* Protected Virtualization is a KVM/Hardware only feature */
508         g_assert_not_reached();
509     }
510     return ret;
511 }
512 
513 /**
514  * s390_cpu_virt_mem_rw:
515  * @laddr:     the logical start address
516  * @ar:        the access register number
517  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
518  * @len:       length that should be transferred
519  * @is_write:  true = write, false = read
520  * Returns:    0 on success, non-zero if an exception occurred
521  *
522  * Copy from/to guest memory using logical addresses. Note that we inject a
523  * program interrupt in case there is an error while accessing the memory.
524  *
525  * This function will always return (also for TCG), make sure to call
526  * s390_cpu_virt_mem_handle_exc() to properly exit the CPU loop.
527  */
528 int s390_cpu_virt_mem_rw(S390CPU *cpu, vaddr laddr, uint8_t ar, void *hostbuf,
529                          int len, bool is_write)
530 {
531     int currlen, nr_pages, i;
532     target_ulong *pages;
533     uint64_t tec;
534     int ret;
535 
536     if (kvm_enabled()) {
537         ret = kvm_s390_mem_op(cpu, laddr, ar, hostbuf, len, is_write);
538         if (ret >= 0) {
539             return ret;
540         }
541     }
542 
543     nr_pages = (((laddr & ~TARGET_PAGE_MASK) + len - 1) >> TARGET_PAGE_BITS)
544                + 1;
545     pages = g_malloc(nr_pages * sizeof(*pages));
546 
547     ret = translate_pages(cpu, laddr, nr_pages, pages, is_write, &tec);
548     if (ret) {
549         trigger_access_exception(&cpu->env, ret, tec);
550     } else if (hostbuf != NULL) {
551         /* Copy data by stepping through the area page by page */
552         for (i = 0; i < nr_pages; i++) {
553             currlen = MIN(len, TARGET_PAGE_SIZE - (laddr % TARGET_PAGE_SIZE));
554             cpu_physical_memory_rw(pages[i] | (laddr & ~TARGET_PAGE_MASK),
555                                    hostbuf, currlen, is_write);
556             laddr += currlen;
557             hostbuf += currlen;
558             len -= currlen;
559         }
560     }
561 
562     g_free(pages);
563     return ret;
564 }
565 
566 void s390_cpu_virt_mem_handle_exc(S390CPU *cpu, uintptr_t ra)
567 {
568     /* KVM will handle the interrupt automatically, TCG has to exit the TB */
569 #ifdef CONFIG_TCG
570     if (tcg_enabled()) {
571         cpu_loop_exit_restore(CPU(cpu), ra);
572     }
573 #endif
574 }
575 
576 /**
577  * Translate a real address into a physical (absolute) address.
578  * @param raddr  the real address
579  * @param rw     0 = read, 1 = write, 2 = code fetch
580  * @param addr   the translated address is stored to this pointer
581  * @param flags  the PAGE_READ/WRITE/EXEC flags are stored to this pointer
582  * @return       0 = success, != 0, the exception to raise
583  */
584 int mmu_translate_real(CPUS390XState *env, target_ulong raddr, int rw,
585                        target_ulong *addr, int *flags, uint64_t *tec)
586 {
587     const bool lowprot_enabled = env->cregs[0] & CR0_LOWPROT;
588 
589     *flags = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
590     if (is_low_address(raddr & TARGET_PAGE_MASK) && lowprot_enabled) {
591         /* see comment in mmu_translate() how this works */
592         *flags |= PAGE_WRITE_INV;
593         if (is_low_address(raddr) && rw == MMU_DATA_STORE) {
594             /* LAP sets bit 56 */
595             *tec = (raddr & TARGET_PAGE_MASK) | FS_WRITE | 0x80;
596             return PGM_PROTECTION;
597         }
598     }
599 
600     *addr = mmu_real2abs(env, raddr & TARGET_PAGE_MASK);
601 
602     if (!mmu_absolute_addr_valid(*addr, rw == MMU_DATA_STORE)) {
603         /* unused */
604         *tec = 0;
605         return PGM_ADDRESSING;
606     }
607 
608     mmu_handle_skey(*addr, rw, flags);
609     return 0;
610 }
611