xref: /openbmc/qemu/hw/ppc/spapr_hcall.c (revision 406d2aa2)
1 #include "qemu/osdep.h"
2 #include "qapi/error.h"
3 #include "sysemu/hw_accel.h"
4 #include "sysemu/sysemu.h"
5 #include "qemu/log.h"
6 #include "qemu/error-report.h"
7 #include "cpu.h"
8 #include "exec/exec-all.h"
9 #include "helper_regs.h"
10 #include "hw/ppc/spapr.h"
11 #include "mmu-hash64.h"
12 #include "cpu-models.h"
13 #include "trace.h"
14 #include "kvm_ppc.h"
15 #include "hw/ppc/spapr_ovec.h"
16 #include "mmu-book3s-v3.h"
17 
18 struct SPRSyncState {
19     int spr;
20     target_ulong value;
21     target_ulong mask;
22 };
23 
24 static void do_spr_sync(CPUState *cs, run_on_cpu_data arg)
25 {
26     struct SPRSyncState *s = arg.host_ptr;
27     PowerPCCPU *cpu = POWERPC_CPU(cs);
28     CPUPPCState *env = &cpu->env;
29 
30     cpu_synchronize_state(cs);
31     env->spr[s->spr] &= ~s->mask;
32     env->spr[s->spr] |= s->value;
33 }
34 
35 static void set_spr(CPUState *cs, int spr, target_ulong value,
36                     target_ulong mask)
37 {
38     struct SPRSyncState s = {
39         .spr = spr,
40         .value = value,
41         .mask = mask
42     };
43     run_on_cpu(cs, do_spr_sync, RUN_ON_CPU_HOST_PTR(&s));
44 }
45 
46 static bool has_spr(PowerPCCPU *cpu, int spr)
47 {
48     /* We can test whether the SPR is defined by checking for a valid name */
49     return cpu->env.spr_cb[spr].name != NULL;
50 }
51 
52 static inline bool valid_ptex(PowerPCCPU *cpu, target_ulong ptex)
53 {
54     /*
55      * hash value/pteg group index is normalized by HPT mask
56      */
57     if (((ptex & ~7ULL) / HPTES_PER_GROUP) & ~ppc_hash64_hpt_mask(cpu)) {
58         return false;
59     }
60     return true;
61 }
62 
63 static bool is_ram_address(sPAPRMachineState *spapr, hwaddr addr)
64 {
65     MachineState *machine = MACHINE(spapr);
66     MemoryHotplugState *hpms = &spapr->hotplug_memory;
67 
68     if (addr < machine->ram_size) {
69         return true;
70     }
71     if ((addr >= hpms->base)
72         && ((addr - hpms->base) < memory_region_size(&hpms->mr))) {
73         return true;
74     }
75 
76     return false;
77 }
78 
79 static target_ulong h_enter(PowerPCCPU *cpu, sPAPRMachineState *spapr,
80                             target_ulong opcode, target_ulong *args)
81 {
82     target_ulong flags = args[0];
83     target_ulong ptex = args[1];
84     target_ulong pteh = args[2];
85     target_ulong ptel = args[3];
86     unsigned apshift;
87     target_ulong raddr;
88     target_ulong slot;
89     const ppc_hash_pte64_t *hptes;
90 
91     apshift = ppc_hash64_hpte_page_shift_noslb(cpu, pteh, ptel);
92     if (!apshift) {
93         /* Bad page size encoding */
94         return H_PARAMETER;
95     }
96 
97     raddr = (ptel & HPTE64_R_RPN) & ~((1ULL << apshift) - 1);
98 
99     if (is_ram_address(spapr, raddr)) {
100         /* Regular RAM - should have WIMG=0010 */
101         if ((ptel & HPTE64_R_WIMG) != HPTE64_R_M) {
102             return H_PARAMETER;
103         }
104     } else {
105         target_ulong wimg_flags;
106         /* Looks like an IO address */
107         /* FIXME: What WIMG combinations could be sensible for IO?
108          * For now we allow WIMG=010x, but are there others? */
109         /* FIXME: Should we check against registered IO addresses? */
110         wimg_flags = (ptel & (HPTE64_R_W | HPTE64_R_I | HPTE64_R_M));
111 
112         if (wimg_flags != HPTE64_R_I &&
113             wimg_flags != (HPTE64_R_I | HPTE64_R_M)) {
114             return H_PARAMETER;
115         }
116     }
117 
118     pteh &= ~0x60ULL;
119 
120     if (!valid_ptex(cpu, ptex)) {
121         return H_PARAMETER;
122     }
123 
124     slot = ptex & 7ULL;
125     ptex = ptex & ~7ULL;
126 
127     if (likely((flags & H_EXACT) == 0)) {
128         hptes = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP);
129         for (slot = 0; slot < 8; slot++) {
130             if (!(ppc_hash64_hpte0(cpu, hptes, slot) & HPTE64_V_VALID)) {
131                 break;
132             }
133         }
134         ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP);
135         if (slot == 8) {
136             return H_PTEG_FULL;
137         }
138     } else {
139         hptes = ppc_hash64_map_hptes(cpu, ptex + slot, 1);
140         if (ppc_hash64_hpte0(cpu, hptes, 0) & HPTE64_V_VALID) {
141             ppc_hash64_unmap_hptes(cpu, hptes, ptex + slot, 1);
142             return H_PTEG_FULL;
143         }
144         ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1);
145     }
146 
147     ppc_hash64_store_hpte(cpu, ptex + slot, pteh | HPTE64_V_HPTE_DIRTY, ptel);
148 
149     args[0] = ptex + slot;
150     return H_SUCCESS;
151 }
152 
153 typedef enum {
154     REMOVE_SUCCESS = 0,
155     REMOVE_NOT_FOUND = 1,
156     REMOVE_PARM = 2,
157     REMOVE_HW = 3,
158 } RemoveResult;
159 
160 static RemoveResult remove_hpte(PowerPCCPU *cpu, target_ulong ptex,
161                                 target_ulong avpn,
162                                 target_ulong flags,
163                                 target_ulong *vp, target_ulong *rp)
164 {
165     const ppc_hash_pte64_t *hptes;
166     target_ulong v, r;
167 
168     if (!valid_ptex(cpu, ptex)) {
169         return REMOVE_PARM;
170     }
171 
172     hptes = ppc_hash64_map_hptes(cpu, ptex, 1);
173     v = ppc_hash64_hpte0(cpu, hptes, 0);
174     r = ppc_hash64_hpte1(cpu, hptes, 0);
175     ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1);
176 
177     if ((v & HPTE64_V_VALID) == 0 ||
178         ((flags & H_AVPN) && (v & ~0x7fULL) != avpn) ||
179         ((flags & H_ANDCOND) && (v & avpn) != 0)) {
180         return REMOVE_NOT_FOUND;
181     }
182     *vp = v;
183     *rp = r;
184     ppc_hash64_store_hpte(cpu, ptex, HPTE64_V_HPTE_DIRTY, 0);
185     ppc_hash64_tlb_flush_hpte(cpu, ptex, v, r);
186     return REMOVE_SUCCESS;
187 }
188 
189 static target_ulong h_remove(PowerPCCPU *cpu, sPAPRMachineState *spapr,
190                              target_ulong opcode, target_ulong *args)
191 {
192     CPUPPCState *env = &cpu->env;
193     target_ulong flags = args[0];
194     target_ulong ptex = args[1];
195     target_ulong avpn = args[2];
196     RemoveResult ret;
197 
198     ret = remove_hpte(cpu, ptex, avpn, flags,
199                       &args[0], &args[1]);
200 
201     switch (ret) {
202     case REMOVE_SUCCESS:
203         check_tlb_flush(env, true);
204         return H_SUCCESS;
205 
206     case REMOVE_NOT_FOUND:
207         return H_NOT_FOUND;
208 
209     case REMOVE_PARM:
210         return H_PARAMETER;
211 
212     case REMOVE_HW:
213         return H_HARDWARE;
214     }
215 
216     g_assert_not_reached();
217 }
218 
219 #define H_BULK_REMOVE_TYPE             0xc000000000000000ULL
220 #define   H_BULK_REMOVE_REQUEST        0x4000000000000000ULL
221 #define   H_BULK_REMOVE_RESPONSE       0x8000000000000000ULL
222 #define   H_BULK_REMOVE_END            0xc000000000000000ULL
223 #define H_BULK_REMOVE_CODE             0x3000000000000000ULL
224 #define   H_BULK_REMOVE_SUCCESS        0x0000000000000000ULL
225 #define   H_BULK_REMOVE_NOT_FOUND      0x1000000000000000ULL
226 #define   H_BULK_REMOVE_PARM           0x2000000000000000ULL
227 #define   H_BULK_REMOVE_HW             0x3000000000000000ULL
228 #define H_BULK_REMOVE_RC               0x0c00000000000000ULL
229 #define H_BULK_REMOVE_FLAGS            0x0300000000000000ULL
230 #define   H_BULK_REMOVE_ABSOLUTE       0x0000000000000000ULL
231 #define   H_BULK_REMOVE_ANDCOND        0x0100000000000000ULL
232 #define   H_BULK_REMOVE_AVPN           0x0200000000000000ULL
233 #define H_BULK_REMOVE_PTEX             0x00ffffffffffffffULL
234 
235 #define H_BULK_REMOVE_MAX_BATCH        4
236 
237 static target_ulong h_bulk_remove(PowerPCCPU *cpu, sPAPRMachineState *spapr,
238                                   target_ulong opcode, target_ulong *args)
239 {
240     CPUPPCState *env = &cpu->env;
241     int i;
242     target_ulong rc = H_SUCCESS;
243 
244     for (i = 0; i < H_BULK_REMOVE_MAX_BATCH; i++) {
245         target_ulong *tsh = &args[i*2];
246         target_ulong tsl = args[i*2 + 1];
247         target_ulong v, r, ret;
248 
249         if ((*tsh & H_BULK_REMOVE_TYPE) == H_BULK_REMOVE_END) {
250             break;
251         } else if ((*tsh & H_BULK_REMOVE_TYPE) != H_BULK_REMOVE_REQUEST) {
252             return H_PARAMETER;
253         }
254 
255         *tsh &= H_BULK_REMOVE_PTEX | H_BULK_REMOVE_FLAGS;
256         *tsh |= H_BULK_REMOVE_RESPONSE;
257 
258         if ((*tsh & H_BULK_REMOVE_ANDCOND) && (*tsh & H_BULK_REMOVE_AVPN)) {
259             *tsh |= H_BULK_REMOVE_PARM;
260             return H_PARAMETER;
261         }
262 
263         ret = remove_hpte(cpu, *tsh & H_BULK_REMOVE_PTEX, tsl,
264                           (*tsh & H_BULK_REMOVE_FLAGS) >> 26,
265                           &v, &r);
266 
267         *tsh |= ret << 60;
268 
269         switch (ret) {
270         case REMOVE_SUCCESS:
271             *tsh |= (r & (HPTE64_R_C | HPTE64_R_R)) << 43;
272             break;
273 
274         case REMOVE_PARM:
275             rc = H_PARAMETER;
276             goto exit;
277 
278         case REMOVE_HW:
279             rc = H_HARDWARE;
280             goto exit;
281         }
282     }
283  exit:
284     check_tlb_flush(env, true);
285 
286     return rc;
287 }
288 
289 static target_ulong h_protect(PowerPCCPU *cpu, sPAPRMachineState *spapr,
290                               target_ulong opcode, target_ulong *args)
291 {
292     CPUPPCState *env = &cpu->env;
293     target_ulong flags = args[0];
294     target_ulong ptex = args[1];
295     target_ulong avpn = args[2];
296     const ppc_hash_pte64_t *hptes;
297     target_ulong v, r;
298 
299     if (!valid_ptex(cpu, ptex)) {
300         return H_PARAMETER;
301     }
302 
303     hptes = ppc_hash64_map_hptes(cpu, ptex, 1);
304     v = ppc_hash64_hpte0(cpu, hptes, 0);
305     r = ppc_hash64_hpte1(cpu, hptes, 0);
306     ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1);
307 
308     if ((v & HPTE64_V_VALID) == 0 ||
309         ((flags & H_AVPN) && (v & ~0x7fULL) != avpn)) {
310         return H_NOT_FOUND;
311     }
312 
313     r &= ~(HPTE64_R_PP0 | HPTE64_R_PP | HPTE64_R_N |
314            HPTE64_R_KEY_HI | HPTE64_R_KEY_LO);
315     r |= (flags << 55) & HPTE64_R_PP0;
316     r |= (flags << 48) & HPTE64_R_KEY_HI;
317     r |= flags & (HPTE64_R_PP | HPTE64_R_N | HPTE64_R_KEY_LO);
318     ppc_hash64_store_hpte(cpu, ptex,
319                           (v & ~HPTE64_V_VALID) | HPTE64_V_HPTE_DIRTY, 0);
320     ppc_hash64_tlb_flush_hpte(cpu, ptex, v, r);
321     /* Flush the tlb */
322     check_tlb_flush(env, true);
323     /* Don't need a memory barrier, due to qemu's global lock */
324     ppc_hash64_store_hpte(cpu, ptex, v | HPTE64_V_HPTE_DIRTY, r);
325     return H_SUCCESS;
326 }
327 
328 static target_ulong h_read(PowerPCCPU *cpu, sPAPRMachineState *spapr,
329                            target_ulong opcode, target_ulong *args)
330 {
331     target_ulong flags = args[0];
332     target_ulong ptex = args[1];
333     uint8_t *hpte;
334     int i, ridx, n_entries = 1;
335 
336     if (!valid_ptex(cpu, ptex)) {
337         return H_PARAMETER;
338     }
339 
340     if (flags & H_READ_4) {
341         /* Clear the two low order bits */
342         ptex &= ~(3ULL);
343         n_entries = 4;
344     }
345 
346     hpte = spapr->htab + (ptex * HASH_PTE_SIZE_64);
347 
348     for (i = 0, ridx = 0; i < n_entries; i++) {
349         args[ridx++] = ldq_p(hpte);
350         args[ridx++] = ldq_p(hpte + (HASH_PTE_SIZE_64/2));
351         hpte += HASH_PTE_SIZE_64;
352     }
353 
354     return H_SUCCESS;
355 }
356 
357 struct sPAPRPendingHPT {
358     /* These fields are read-only after initialization */
359     int shift;
360     QemuThread thread;
361 
362     /* These fields are protected by the BQL */
363     bool complete;
364 
365     /* These fields are private to the preparation thread if
366      * !complete, otherwise protected by the BQL */
367     int ret;
368     void *hpt;
369 };
370 
371 static void free_pending_hpt(sPAPRPendingHPT *pending)
372 {
373     if (pending->hpt) {
374         qemu_vfree(pending->hpt);
375     }
376 
377     g_free(pending);
378 }
379 
380 static void *hpt_prepare_thread(void *opaque)
381 {
382     sPAPRPendingHPT *pending = opaque;
383     size_t size = 1ULL << pending->shift;
384 
385     pending->hpt = qemu_memalign(size, size);
386     if (pending->hpt) {
387         memset(pending->hpt, 0, size);
388         pending->ret = H_SUCCESS;
389     } else {
390         pending->ret = H_NO_MEM;
391     }
392 
393     qemu_mutex_lock_iothread();
394 
395     if (SPAPR_MACHINE(qdev_get_machine())->pending_hpt == pending) {
396         /* Ready to go */
397         pending->complete = true;
398     } else {
399         /* We've been cancelled, clean ourselves up */
400         free_pending_hpt(pending);
401     }
402 
403     qemu_mutex_unlock_iothread();
404     return NULL;
405 }
406 
407 /* Must be called with BQL held */
408 static void cancel_hpt_prepare(sPAPRMachineState *spapr)
409 {
410     sPAPRPendingHPT *pending = spapr->pending_hpt;
411 
412     /* Let the thread know it's cancelled */
413     spapr->pending_hpt = NULL;
414 
415     if (!pending) {
416         /* Nothing to do */
417         return;
418     }
419 
420     if (!pending->complete) {
421         /* thread will clean itself up */
422         return;
423     }
424 
425     free_pending_hpt(pending);
426 }
427 
428 /* Convert a return code from the KVM ioctl()s implementing resize HPT
429  * into a PAPR hypercall return code */
430 static target_ulong resize_hpt_convert_rc(int ret)
431 {
432     if (ret >= 100000) {
433         return H_LONG_BUSY_ORDER_100_SEC;
434     } else if (ret >= 10000) {
435         return H_LONG_BUSY_ORDER_10_SEC;
436     } else if (ret >= 1000) {
437         return H_LONG_BUSY_ORDER_1_SEC;
438     } else if (ret >= 100) {
439         return H_LONG_BUSY_ORDER_100_MSEC;
440     } else if (ret >= 10) {
441         return H_LONG_BUSY_ORDER_10_MSEC;
442     } else if (ret > 0) {
443         return H_LONG_BUSY_ORDER_1_MSEC;
444     }
445 
446     switch (ret) {
447     case 0:
448         return H_SUCCESS;
449     case -EPERM:
450         return H_AUTHORITY;
451     case -EINVAL:
452         return H_PARAMETER;
453     case -ENXIO:
454         return H_CLOSED;
455     case -ENOSPC:
456         return H_PTEG_FULL;
457     case -EBUSY:
458         return H_BUSY;
459     case -ENOMEM:
460         return H_NO_MEM;
461     default:
462         return H_HARDWARE;
463     }
464 }
465 
466 static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
467                                          sPAPRMachineState *spapr,
468                                          target_ulong opcode,
469                                          target_ulong *args)
470 {
471     target_ulong flags = args[0];
472     int shift = args[1];
473     sPAPRPendingHPT *pending = spapr->pending_hpt;
474     uint64_t current_ram_size;
475     int rc;
476 
477     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
478         return H_AUTHORITY;
479     }
480 
481     if (!spapr->htab_shift) {
482         /* Radix guest, no HPT */
483         return H_NOT_AVAILABLE;
484     }
485 
486     trace_spapr_h_resize_hpt_prepare(flags, shift);
487 
488     if (flags != 0) {
489         return H_PARAMETER;
490     }
491 
492     if (shift && ((shift < 18) || (shift > 46))) {
493         return H_PARAMETER;
494     }
495 
496     current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
497 
498     /* We only allow the guest to allocate an HPT one order above what
499      * we'd normally give them (to stop a small guest claiming a huge
500      * chunk of resources in the HPT */
501     if (shift > (spapr_hpt_shift_for_ramsize(current_ram_size) + 1)) {
502         return H_RESOURCE;
503     }
504 
505     rc = kvmppc_resize_hpt_prepare(cpu, flags, shift);
506     if (rc != -ENOSYS) {
507         return resize_hpt_convert_rc(rc);
508     }
509 
510     if (pending) {
511         /* something already in progress */
512         if (pending->shift == shift) {
513             /* and it's suitable */
514             if (pending->complete) {
515                 return pending->ret;
516             } else {
517                 return H_LONG_BUSY_ORDER_100_MSEC;
518             }
519         }
520 
521         /* not suitable, cancel and replace */
522         cancel_hpt_prepare(spapr);
523     }
524 
525     if (!shift) {
526         /* nothing to do */
527         return H_SUCCESS;
528     }
529 
530     /* start new prepare */
531 
532     pending = g_new0(sPAPRPendingHPT, 1);
533     pending->shift = shift;
534     pending->ret = H_HARDWARE;
535 
536     qemu_thread_create(&pending->thread, "sPAPR HPT prepare",
537                        hpt_prepare_thread, pending, QEMU_THREAD_DETACHED);
538 
539     spapr->pending_hpt = pending;
540 
541     /* In theory we could estimate the time more accurately based on
542      * the new size, but there's not much point */
543     return H_LONG_BUSY_ORDER_100_MSEC;
544 }
545 
546 static uint64_t new_hpte_load0(void *htab, uint64_t pteg, int slot)
547 {
548     uint8_t *addr = htab;
549 
550     addr += pteg * HASH_PTEG_SIZE_64;
551     addr += slot * HASH_PTE_SIZE_64;
552     return  ldq_p(addr);
553 }
554 
555 static void new_hpte_store(void *htab, uint64_t pteg, int slot,
556                            uint64_t pte0, uint64_t pte1)
557 {
558     uint8_t *addr = htab;
559 
560     addr += pteg * HASH_PTEG_SIZE_64;
561     addr += slot * HASH_PTE_SIZE_64;
562 
563     stq_p(addr, pte0);
564     stq_p(addr + HASH_PTE_SIZE_64 / 2, pte1);
565 }
566 
567 static int rehash_hpte(PowerPCCPU *cpu,
568                        const ppc_hash_pte64_t *hptes,
569                        void *old_hpt, uint64_t oldsize,
570                        void *new_hpt, uint64_t newsize,
571                        uint64_t pteg, int slot)
572 {
573     uint64_t old_hash_mask = (oldsize >> 7) - 1;
574     uint64_t new_hash_mask = (newsize >> 7) - 1;
575     target_ulong pte0 = ppc_hash64_hpte0(cpu, hptes, slot);
576     target_ulong pte1;
577     uint64_t avpn;
578     unsigned base_pg_shift;
579     uint64_t hash, new_pteg, replace_pte0;
580 
581     if (!(pte0 & HPTE64_V_VALID) || !(pte0 & HPTE64_V_BOLTED)) {
582         return H_SUCCESS;
583     }
584 
585     pte1 = ppc_hash64_hpte1(cpu, hptes, slot);
586 
587     base_pg_shift = ppc_hash64_hpte_page_shift_noslb(cpu, pte0, pte1);
588     assert(base_pg_shift); /* H_ENTER shouldn't allow a bad encoding */
589     avpn = HPTE64_V_AVPN_VAL(pte0) & ~(((1ULL << base_pg_shift) - 1) >> 23);
590 
591     if (pte0 & HPTE64_V_SECONDARY) {
592         pteg = ~pteg;
593     }
594 
595     if ((pte0 & HPTE64_V_SSIZE) == HPTE64_V_SSIZE_256M) {
596         uint64_t offset, vsid;
597 
598         /* We only have 28 - 23 bits of offset in avpn */
599         offset = (avpn & 0x1f) << 23;
600         vsid = avpn >> 5;
601         /* We can find more bits from the pteg value */
602         if (base_pg_shift < 23) {
603             offset |= ((vsid ^ pteg) & old_hash_mask) << base_pg_shift;
604         }
605 
606         hash = vsid ^ (offset >> base_pg_shift);
607     } else if ((pte0 & HPTE64_V_SSIZE) == HPTE64_V_SSIZE_1T) {
608         uint64_t offset, vsid;
609 
610         /* We only have 40 - 23 bits of seg_off in avpn */
611         offset = (avpn & 0x1ffff) << 23;
612         vsid = avpn >> 17;
613         if (base_pg_shift < 23) {
614             offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask)
615                 << base_pg_shift;
616         }
617 
618         hash = vsid ^ (vsid << 25) ^ (offset >> base_pg_shift);
619     } else {
620         error_report("rehash_pte: Bad segment size in HPTE");
621         return H_HARDWARE;
622     }
623 
624     new_pteg = hash & new_hash_mask;
625     if (pte0 & HPTE64_V_SECONDARY) {
626         assert(~pteg == (hash & old_hash_mask));
627         new_pteg = ~new_pteg;
628     } else {
629         assert(pteg == (hash & old_hash_mask));
630     }
631     assert((oldsize != newsize) || (pteg == new_pteg));
632     replace_pte0 = new_hpte_load0(new_hpt, new_pteg, slot);
633     /*
634      * Strictly speaking, we don't need all these tests, since we only
635      * ever rehash bolted HPTEs.  We might in future handle non-bolted
636      * HPTEs, though so make the logic correct for those cases as
637      * well.
638      */
639     if (replace_pte0 & HPTE64_V_VALID) {
640         assert(newsize < oldsize);
641         if (replace_pte0 & HPTE64_V_BOLTED) {
642             if (pte0 & HPTE64_V_BOLTED) {
643                 /* Bolted collision, nothing we can do */
644                 return H_PTEG_FULL;
645             } else {
646                 /* Discard this hpte */
647                 return H_SUCCESS;
648             }
649         }
650     }
651 
652     new_hpte_store(new_hpt, new_pteg, slot, pte0, pte1);
653     return H_SUCCESS;
654 }
655 
656 static int rehash_hpt(PowerPCCPU *cpu,
657                       void *old_hpt, uint64_t oldsize,
658                       void *new_hpt, uint64_t newsize)
659 {
660     uint64_t n_ptegs = oldsize >> 7;
661     uint64_t pteg;
662     int slot;
663     int rc;
664 
665     for (pteg = 0; pteg < n_ptegs; pteg++) {
666         hwaddr ptex = pteg * HPTES_PER_GROUP;
667         const ppc_hash_pte64_t *hptes
668             = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP);
669 
670         if (!hptes) {
671             return H_HARDWARE;
672         }
673 
674         for (slot = 0; slot < HPTES_PER_GROUP; slot++) {
675             rc = rehash_hpte(cpu, hptes, old_hpt, oldsize, new_hpt, newsize,
676                              pteg, slot);
677             if (rc != H_SUCCESS) {
678                 ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP);
679                 return rc;
680             }
681         }
682         ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP);
683     }
684 
685     return H_SUCCESS;
686 }
687 
688 static void do_push_sregs_to_kvm_pr(CPUState *cs, run_on_cpu_data data)
689 {
690     int ret;
691 
692     cpu_synchronize_state(cs);
693 
694     ret = kvmppc_put_books_sregs(POWERPC_CPU(cs));
695     if (ret < 0) {
696         error_report("failed to push sregs to KVM: %s", strerror(-ret));
697         exit(1);
698     }
699 }
700 
701 static void push_sregs_to_kvm_pr(sPAPRMachineState *spapr)
702 {
703     CPUState *cs;
704 
705     /*
706      * This is a hack for the benefit of KVM PR - it abuses the SDR1
707      * slot in kvm_sregs to communicate the userspace address of the
708      * HPT
709      */
710     if (!kvm_enabled() || !spapr->htab) {
711         return;
712     }
713 
714     CPU_FOREACH(cs) {
715         run_on_cpu(cs, do_push_sregs_to_kvm_pr, RUN_ON_CPU_NULL);
716     }
717 }
718 
719 static target_ulong h_resize_hpt_commit(PowerPCCPU *cpu,
720                                         sPAPRMachineState *spapr,
721                                         target_ulong opcode,
722                                         target_ulong *args)
723 {
724     target_ulong flags = args[0];
725     target_ulong shift = args[1];
726     sPAPRPendingHPT *pending = spapr->pending_hpt;
727     int rc;
728     size_t newsize;
729 
730     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
731         return H_AUTHORITY;
732     }
733 
734     trace_spapr_h_resize_hpt_commit(flags, shift);
735 
736     rc = kvmppc_resize_hpt_commit(cpu, flags, shift);
737     if (rc != -ENOSYS) {
738         return resize_hpt_convert_rc(rc);
739     }
740 
741     if (flags != 0) {
742         return H_PARAMETER;
743     }
744 
745     if (!pending || (pending->shift != shift)) {
746         /* no matching prepare */
747         return H_CLOSED;
748     }
749 
750     if (!pending->complete) {
751         /* prepare has not completed */
752         return H_BUSY;
753     }
754 
755     /* Shouldn't have got past PREPARE without an HPT */
756     g_assert(spapr->htab_shift);
757 
758     newsize = 1ULL << pending->shift;
759     rc = rehash_hpt(cpu, spapr->htab, HTAB_SIZE(spapr),
760                     pending->hpt, newsize);
761     if (rc == H_SUCCESS) {
762         qemu_vfree(spapr->htab);
763         spapr->htab = pending->hpt;
764         spapr->htab_shift = pending->shift;
765 
766         push_sregs_to_kvm_pr(spapr);
767 
768         pending->hpt = NULL; /* so it's not free()d */
769     }
770 
771     /* Clean up */
772     spapr->pending_hpt = NULL;
773     free_pending_hpt(pending);
774 
775     return rc;
776 }
777 
778 static target_ulong h_set_sprg0(PowerPCCPU *cpu, sPAPRMachineState *spapr,
779                                 target_ulong opcode, target_ulong *args)
780 {
781     cpu_synchronize_state(CPU(cpu));
782     cpu->env.spr[SPR_SPRG0] = args[0];
783 
784     return H_SUCCESS;
785 }
786 
787 static target_ulong h_set_dabr(PowerPCCPU *cpu, sPAPRMachineState *spapr,
788                                target_ulong opcode, target_ulong *args)
789 {
790     if (!has_spr(cpu, SPR_DABR)) {
791         return H_HARDWARE;              /* DABR register not available */
792     }
793     cpu_synchronize_state(CPU(cpu));
794 
795     if (has_spr(cpu, SPR_DABRX)) {
796         cpu->env.spr[SPR_DABRX] = 0x3;  /* Use Problem and Privileged state */
797     } else if (!(args[0] & 0x4)) {      /* Breakpoint Translation set? */
798         return H_RESERVED_DABR;
799     }
800 
801     cpu->env.spr[SPR_DABR] = args[0];
802     return H_SUCCESS;
803 }
804 
805 static target_ulong h_set_xdabr(PowerPCCPU *cpu, sPAPRMachineState *spapr,
806                                 target_ulong opcode, target_ulong *args)
807 {
808     target_ulong dabrx = args[1];
809 
810     if (!has_spr(cpu, SPR_DABR) || !has_spr(cpu, SPR_DABRX)) {
811         return H_HARDWARE;
812     }
813 
814     if ((dabrx & ~0xfULL) != 0 || (dabrx & H_DABRX_HYPERVISOR) != 0
815         || (dabrx & (H_DABRX_KERNEL | H_DABRX_USER)) == 0) {
816         return H_PARAMETER;
817     }
818 
819     cpu_synchronize_state(CPU(cpu));
820     cpu->env.spr[SPR_DABRX] = dabrx;
821     cpu->env.spr[SPR_DABR] = args[0];
822 
823     return H_SUCCESS;
824 }
825 
826 static target_ulong h_page_init(PowerPCCPU *cpu, sPAPRMachineState *spapr,
827                                 target_ulong opcode, target_ulong *args)
828 {
829     target_ulong flags = args[0];
830     hwaddr dst = args[1];
831     hwaddr src = args[2];
832     hwaddr len = TARGET_PAGE_SIZE;
833     uint8_t *pdst, *psrc;
834     target_long ret = H_SUCCESS;
835 
836     if (flags & ~(H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE
837                   | H_COPY_PAGE | H_ZERO_PAGE)) {
838         qemu_log_mask(LOG_UNIMP, "h_page_init: Bad flags (" TARGET_FMT_lx "\n",
839                       flags);
840         return H_PARAMETER;
841     }
842 
843     /* Map-in destination */
844     if (!is_ram_address(spapr, dst) || (dst & ~TARGET_PAGE_MASK) != 0) {
845         return H_PARAMETER;
846     }
847     pdst = cpu_physical_memory_map(dst, &len, 1);
848     if (!pdst || len != TARGET_PAGE_SIZE) {
849         return H_PARAMETER;
850     }
851 
852     if (flags & H_COPY_PAGE) {
853         /* Map-in source, copy to destination, and unmap source again */
854         if (!is_ram_address(spapr, src) || (src & ~TARGET_PAGE_MASK) != 0) {
855             ret = H_PARAMETER;
856             goto unmap_out;
857         }
858         psrc = cpu_physical_memory_map(src, &len, 0);
859         if (!psrc || len != TARGET_PAGE_SIZE) {
860             ret = H_PARAMETER;
861             goto unmap_out;
862         }
863         memcpy(pdst, psrc, len);
864         cpu_physical_memory_unmap(psrc, len, 0, len);
865     } else if (flags & H_ZERO_PAGE) {
866         memset(pdst, 0, len);          /* Just clear the destination page */
867     }
868 
869     if (kvm_enabled() && (flags & H_ICACHE_SYNCHRONIZE) != 0) {
870         kvmppc_dcbst_range(cpu, pdst, len);
871     }
872     if (flags & (H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE)) {
873         if (kvm_enabled()) {
874             kvmppc_icbi_range(cpu, pdst, len);
875         } else {
876             tb_flush(CPU(cpu));
877         }
878     }
879 
880 unmap_out:
881     cpu_physical_memory_unmap(pdst, TARGET_PAGE_SIZE, 1, len);
882     return ret;
883 }
884 
885 #define FLAGS_REGISTER_VPA         0x0000200000000000ULL
886 #define FLAGS_REGISTER_DTL         0x0000400000000000ULL
887 #define FLAGS_REGISTER_SLBSHADOW   0x0000600000000000ULL
888 #define FLAGS_DEREGISTER_VPA       0x0000a00000000000ULL
889 #define FLAGS_DEREGISTER_DTL       0x0000c00000000000ULL
890 #define FLAGS_DEREGISTER_SLBSHADOW 0x0000e00000000000ULL
891 
892 #define VPA_MIN_SIZE           640
893 #define VPA_SIZE_OFFSET        0x4
894 #define VPA_SHARED_PROC_OFFSET 0x9
895 #define VPA_SHARED_PROC_VAL    0x2
896 
897 static target_ulong register_vpa(CPUPPCState *env, target_ulong vpa)
898 {
899     CPUState *cs = CPU(ppc_env_get_cpu(env));
900     uint16_t size;
901     uint8_t tmp;
902 
903     if (vpa == 0) {
904         hcall_dprintf("Can't cope with registering a VPA at logical 0\n");
905         return H_HARDWARE;
906     }
907 
908     if (vpa % env->dcache_line_size) {
909         return H_PARAMETER;
910     }
911     /* FIXME: bounds check the address */
912 
913     size = lduw_be_phys(cs->as, vpa + 0x4);
914 
915     if (size < VPA_MIN_SIZE) {
916         return H_PARAMETER;
917     }
918 
919     /* VPA is not allowed to cross a page boundary */
920     if ((vpa / 4096) != ((vpa + size - 1) / 4096)) {
921         return H_PARAMETER;
922     }
923 
924     env->vpa_addr = vpa;
925 
926     tmp = ldub_phys(cs->as, env->vpa_addr + VPA_SHARED_PROC_OFFSET);
927     tmp |= VPA_SHARED_PROC_VAL;
928     stb_phys(cs->as, env->vpa_addr + VPA_SHARED_PROC_OFFSET, tmp);
929 
930     return H_SUCCESS;
931 }
932 
933 static target_ulong deregister_vpa(CPUPPCState *env, target_ulong vpa)
934 {
935     if (env->slb_shadow_addr) {
936         return H_RESOURCE;
937     }
938 
939     if (env->dtl_addr) {
940         return H_RESOURCE;
941     }
942 
943     env->vpa_addr = 0;
944     return H_SUCCESS;
945 }
946 
947 static target_ulong register_slb_shadow(CPUPPCState *env, target_ulong addr)
948 {
949     CPUState *cs = CPU(ppc_env_get_cpu(env));
950     uint32_t size;
951 
952     if (addr == 0) {
953         hcall_dprintf("Can't cope with SLB shadow at logical 0\n");
954         return H_HARDWARE;
955     }
956 
957     size = ldl_be_phys(cs->as, addr + 0x4);
958     if (size < 0x8) {
959         return H_PARAMETER;
960     }
961 
962     if ((addr / 4096) != ((addr + size - 1) / 4096)) {
963         return H_PARAMETER;
964     }
965 
966     if (!env->vpa_addr) {
967         return H_RESOURCE;
968     }
969 
970     env->slb_shadow_addr = addr;
971     env->slb_shadow_size = size;
972 
973     return H_SUCCESS;
974 }
975 
976 static target_ulong deregister_slb_shadow(CPUPPCState *env, target_ulong addr)
977 {
978     env->slb_shadow_addr = 0;
979     env->slb_shadow_size = 0;
980     return H_SUCCESS;
981 }
982 
983 static target_ulong register_dtl(CPUPPCState *env, target_ulong addr)
984 {
985     CPUState *cs = CPU(ppc_env_get_cpu(env));
986     uint32_t size;
987 
988     if (addr == 0) {
989         hcall_dprintf("Can't cope with DTL at logical 0\n");
990         return H_HARDWARE;
991     }
992 
993     size = ldl_be_phys(cs->as, addr + 0x4);
994 
995     if (size < 48) {
996         return H_PARAMETER;
997     }
998 
999     if (!env->vpa_addr) {
1000         return H_RESOURCE;
1001     }
1002 
1003     env->dtl_addr = addr;
1004     env->dtl_size = size;
1005 
1006     return H_SUCCESS;
1007 }
1008 
1009 static target_ulong deregister_dtl(CPUPPCState *env, target_ulong addr)
1010 {
1011     env->dtl_addr = 0;
1012     env->dtl_size = 0;
1013 
1014     return H_SUCCESS;
1015 }
1016 
1017 static target_ulong h_register_vpa(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1018                                    target_ulong opcode, target_ulong *args)
1019 {
1020     target_ulong flags = args[0];
1021     target_ulong procno = args[1];
1022     target_ulong vpa = args[2];
1023     target_ulong ret = H_PARAMETER;
1024     CPUPPCState *tenv;
1025     PowerPCCPU *tcpu;
1026 
1027     tcpu = spapr_find_cpu(procno);
1028     if (!tcpu) {
1029         return H_PARAMETER;
1030     }
1031     tenv = &tcpu->env;
1032 
1033     switch (flags) {
1034     case FLAGS_REGISTER_VPA:
1035         ret = register_vpa(tenv, vpa);
1036         break;
1037 
1038     case FLAGS_DEREGISTER_VPA:
1039         ret = deregister_vpa(tenv, vpa);
1040         break;
1041 
1042     case FLAGS_REGISTER_SLBSHADOW:
1043         ret = register_slb_shadow(tenv, vpa);
1044         break;
1045 
1046     case FLAGS_DEREGISTER_SLBSHADOW:
1047         ret = deregister_slb_shadow(tenv, vpa);
1048         break;
1049 
1050     case FLAGS_REGISTER_DTL:
1051         ret = register_dtl(tenv, vpa);
1052         break;
1053 
1054     case FLAGS_DEREGISTER_DTL:
1055         ret = deregister_dtl(tenv, vpa);
1056         break;
1057     }
1058 
1059     return ret;
1060 }
1061 
1062 static target_ulong h_cede(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1063                            target_ulong opcode, target_ulong *args)
1064 {
1065     CPUPPCState *env = &cpu->env;
1066     CPUState *cs = CPU(cpu);
1067 
1068     env->msr |= (1ULL << MSR_EE);
1069     hreg_compute_hflags(env);
1070     if (!cpu_has_work(cs)) {
1071         cs->halted = 1;
1072         cs->exception_index = EXCP_HLT;
1073         cs->exit_request = 1;
1074     }
1075     return H_SUCCESS;
1076 }
1077 
1078 static target_ulong h_rtas(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1079                            target_ulong opcode, target_ulong *args)
1080 {
1081     target_ulong rtas_r3 = args[0];
1082     uint32_t token = rtas_ld(rtas_r3, 0);
1083     uint32_t nargs = rtas_ld(rtas_r3, 1);
1084     uint32_t nret = rtas_ld(rtas_r3, 2);
1085 
1086     return spapr_rtas_call(cpu, spapr, token, nargs, rtas_r3 + 12,
1087                            nret, rtas_r3 + 12 + 4*nargs);
1088 }
1089 
1090 static target_ulong h_logical_load(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1091                                    target_ulong opcode, target_ulong *args)
1092 {
1093     CPUState *cs = CPU(cpu);
1094     target_ulong size = args[0];
1095     target_ulong addr = args[1];
1096 
1097     switch (size) {
1098     case 1:
1099         args[0] = ldub_phys(cs->as, addr);
1100         return H_SUCCESS;
1101     case 2:
1102         args[0] = lduw_phys(cs->as, addr);
1103         return H_SUCCESS;
1104     case 4:
1105         args[0] = ldl_phys(cs->as, addr);
1106         return H_SUCCESS;
1107     case 8:
1108         args[0] = ldq_phys(cs->as, addr);
1109         return H_SUCCESS;
1110     }
1111     return H_PARAMETER;
1112 }
1113 
1114 static target_ulong h_logical_store(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1115                                     target_ulong opcode, target_ulong *args)
1116 {
1117     CPUState *cs = CPU(cpu);
1118 
1119     target_ulong size = args[0];
1120     target_ulong addr = args[1];
1121     target_ulong val  = args[2];
1122 
1123     switch (size) {
1124     case 1:
1125         stb_phys(cs->as, addr, val);
1126         return H_SUCCESS;
1127     case 2:
1128         stw_phys(cs->as, addr, val);
1129         return H_SUCCESS;
1130     case 4:
1131         stl_phys(cs->as, addr, val);
1132         return H_SUCCESS;
1133     case 8:
1134         stq_phys(cs->as, addr, val);
1135         return H_SUCCESS;
1136     }
1137     return H_PARAMETER;
1138 }
1139 
1140 static target_ulong h_logical_memop(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1141                                     target_ulong opcode, target_ulong *args)
1142 {
1143     CPUState *cs = CPU(cpu);
1144 
1145     target_ulong dst   = args[0]; /* Destination address */
1146     target_ulong src   = args[1]; /* Source address */
1147     target_ulong esize = args[2]; /* Element size (0=1,1=2,2=4,3=8) */
1148     target_ulong count = args[3]; /* Element count */
1149     target_ulong op    = args[4]; /* 0 = copy, 1 = invert */
1150     uint64_t tmp;
1151     unsigned int mask = (1 << esize) - 1;
1152     int step = 1 << esize;
1153 
1154     if (count > 0x80000000) {
1155         return H_PARAMETER;
1156     }
1157 
1158     if ((dst & mask) || (src & mask) || (op > 1)) {
1159         return H_PARAMETER;
1160     }
1161 
1162     if (dst >= src && dst < (src + (count << esize))) {
1163             dst = dst + ((count - 1) << esize);
1164             src = src + ((count - 1) << esize);
1165             step = -step;
1166     }
1167 
1168     while (count--) {
1169         switch (esize) {
1170         case 0:
1171             tmp = ldub_phys(cs->as, src);
1172             break;
1173         case 1:
1174             tmp = lduw_phys(cs->as, src);
1175             break;
1176         case 2:
1177             tmp = ldl_phys(cs->as, src);
1178             break;
1179         case 3:
1180             tmp = ldq_phys(cs->as, src);
1181             break;
1182         default:
1183             return H_PARAMETER;
1184         }
1185         if (op == 1) {
1186             tmp = ~tmp;
1187         }
1188         switch (esize) {
1189         case 0:
1190             stb_phys(cs->as, dst, tmp);
1191             break;
1192         case 1:
1193             stw_phys(cs->as, dst, tmp);
1194             break;
1195         case 2:
1196             stl_phys(cs->as, dst, tmp);
1197             break;
1198         case 3:
1199             stq_phys(cs->as, dst, tmp);
1200             break;
1201         }
1202         dst = dst + step;
1203         src = src + step;
1204     }
1205 
1206     return H_SUCCESS;
1207 }
1208 
1209 static target_ulong h_logical_icbi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1210                                    target_ulong opcode, target_ulong *args)
1211 {
1212     /* Nothing to do on emulation, KVM will trap this in the kernel */
1213     return H_SUCCESS;
1214 }
1215 
1216 static target_ulong h_logical_dcbf(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1217                                    target_ulong opcode, target_ulong *args)
1218 {
1219     /* Nothing to do on emulation, KVM will trap this in the kernel */
1220     return H_SUCCESS;
1221 }
1222 
1223 static target_ulong h_set_mode_resource_le(PowerPCCPU *cpu,
1224                                            target_ulong mflags,
1225                                            target_ulong value1,
1226                                            target_ulong value2)
1227 {
1228     CPUState *cs;
1229 
1230     if (value1) {
1231         return H_P3;
1232     }
1233     if (value2) {
1234         return H_P4;
1235     }
1236 
1237     switch (mflags) {
1238     case H_SET_MODE_ENDIAN_BIG:
1239         CPU_FOREACH(cs) {
1240             set_spr(cs, SPR_LPCR, 0, LPCR_ILE);
1241         }
1242         spapr_pci_switch_vga(true);
1243         return H_SUCCESS;
1244 
1245     case H_SET_MODE_ENDIAN_LITTLE:
1246         CPU_FOREACH(cs) {
1247             set_spr(cs, SPR_LPCR, LPCR_ILE, LPCR_ILE);
1248         }
1249         spapr_pci_switch_vga(false);
1250         return H_SUCCESS;
1251     }
1252 
1253     return H_UNSUPPORTED_FLAG;
1254 }
1255 
1256 static target_ulong h_set_mode_resource_addr_trans_mode(PowerPCCPU *cpu,
1257                                                         target_ulong mflags,
1258                                                         target_ulong value1,
1259                                                         target_ulong value2)
1260 {
1261     CPUState *cs;
1262     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
1263 
1264     if (!(pcc->insns_flags2 & PPC2_ISA207S)) {
1265         return H_P2;
1266     }
1267     if (value1) {
1268         return H_P3;
1269     }
1270     if (value2) {
1271         return H_P4;
1272     }
1273 
1274     if (mflags == AIL_RESERVED) {
1275         return H_UNSUPPORTED_FLAG;
1276     }
1277 
1278     CPU_FOREACH(cs) {
1279         set_spr(cs, SPR_LPCR, mflags << LPCR_AIL_SHIFT, LPCR_AIL);
1280     }
1281 
1282     return H_SUCCESS;
1283 }
1284 
1285 static target_ulong h_set_mode(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1286                                target_ulong opcode, target_ulong *args)
1287 {
1288     target_ulong resource = args[1];
1289     target_ulong ret = H_P2;
1290 
1291     switch (resource) {
1292     case H_SET_MODE_RESOURCE_LE:
1293         ret = h_set_mode_resource_le(cpu, args[0], args[2], args[3]);
1294         break;
1295     case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
1296         ret = h_set_mode_resource_addr_trans_mode(cpu, args[0],
1297                                                   args[2], args[3]);
1298         break;
1299     }
1300 
1301     return ret;
1302 }
1303 
1304 static target_ulong h_clean_slb(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1305                                 target_ulong opcode, target_ulong *args)
1306 {
1307     qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n",
1308                   opcode, " (H_CLEAN_SLB)");
1309     return H_FUNCTION;
1310 }
1311 
1312 static target_ulong h_invalidate_pid(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1313                                      target_ulong opcode, target_ulong *args)
1314 {
1315     qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n",
1316                   opcode, " (H_INVALIDATE_PID)");
1317     return H_FUNCTION;
1318 }
1319 
1320 static void spapr_check_setup_free_hpt(sPAPRMachineState *spapr,
1321                                        uint64_t patbe_old, uint64_t patbe_new)
1322 {
1323     /*
1324      * We have 4 Options:
1325      * HASH->HASH || RADIX->RADIX || NOTHING->RADIX : Do Nothing
1326      * HASH->RADIX                                  : Free HPT
1327      * RADIX->HASH                                  : Allocate HPT
1328      * NOTHING->HASH                                : Allocate HPT
1329      * Note: NOTHING implies the case where we said the guest could choose
1330      *       later and so assumed radix and now it's called H_REG_PROC_TBL
1331      */
1332 
1333     if ((patbe_old & PATBE1_GR) == (patbe_new & PATBE1_GR)) {
1334         /* We assume RADIX, so this catches all the "Do Nothing" cases */
1335     } else if (!(patbe_old & PATBE1_GR)) {
1336         /* HASH->RADIX : Free HPT */
1337         spapr_free_hpt(spapr);
1338     } else if (!(patbe_new & PATBE1_GR)) {
1339         /* RADIX->HASH || NOTHING->HASH : Allocate HPT */
1340         spapr_setup_hpt_and_vrma(spapr);
1341     }
1342     return;
1343 }
1344 
1345 #define FLAGS_MASK              0x01FULL
1346 #define FLAG_MODIFY             0x10
1347 #define FLAG_REGISTER           0x08
1348 #define FLAG_RADIX              0x04
1349 #define FLAG_HASH_PROC_TBL      0x02
1350 #define FLAG_GTSE               0x01
1351 
1352 static target_ulong h_register_process_table(PowerPCCPU *cpu,
1353                                              sPAPRMachineState *spapr,
1354                                              target_ulong opcode,
1355                                              target_ulong *args)
1356 {
1357     CPUState *cs;
1358     target_ulong flags = args[0];
1359     target_ulong proc_tbl = args[1];
1360     target_ulong page_size = args[2];
1361     target_ulong table_size = args[3];
1362     uint64_t cproc;
1363 
1364     if (flags & ~FLAGS_MASK) { /* Check no reserved bits are set */
1365         return H_PARAMETER;
1366     }
1367     if (flags & FLAG_MODIFY) {
1368         if (flags & FLAG_REGISTER) {
1369             if (flags & FLAG_RADIX) { /* Register new RADIX process table */
1370                 if (proc_tbl & 0xfff || proc_tbl >> 60) {
1371                     return H_P2;
1372                 } else if (page_size) {
1373                     return H_P3;
1374                 } else if (table_size > 24) {
1375                     return H_P4;
1376                 }
1377                 cproc = PATBE1_GR | proc_tbl | table_size;
1378             } else { /* Register new HPT process table */
1379                 if (flags & FLAG_HASH_PROC_TBL) { /* Hash with Segment Tables */
1380                     /* TODO - Not Supported */
1381                     /* Technically caused by flag bits => H_PARAMETER */
1382                     return H_PARAMETER;
1383                 } else { /* Hash with SLB */
1384                     if (proc_tbl >> 38) {
1385                         return H_P2;
1386                     } else if (page_size & ~0x7) {
1387                         return H_P3;
1388                     } else if (table_size > 24) {
1389                         return H_P4;
1390                     }
1391                 }
1392                 cproc = (proc_tbl << 25) | page_size << 5 | table_size;
1393             }
1394 
1395         } else { /* Deregister current process table */
1396             /* Set to benign value: (current GR) | 0. This allows
1397              * deregistration in KVM to succeed even if the radix bit in flags
1398              * doesn't match the radix bit in the old PATB. */
1399             cproc = spapr->patb_entry & PATBE1_GR;
1400         }
1401     } else { /* Maintain current registration */
1402         if (!(flags & FLAG_RADIX) != !(spapr->patb_entry & PATBE1_GR)) {
1403             /* Technically caused by flag bits => H_PARAMETER */
1404             return H_PARAMETER; /* Existing Process Table Mismatch */
1405         }
1406         cproc = spapr->patb_entry;
1407     }
1408 
1409     /* Check if we need to setup OR free the hpt */
1410     spapr_check_setup_free_hpt(spapr, spapr->patb_entry, cproc);
1411 
1412     spapr->patb_entry = cproc; /* Save new process table */
1413 
1414     /* Update the UPRT and GTSE bits in the LPCR for all cpus */
1415     CPU_FOREACH(cs) {
1416         set_spr(cs, SPR_LPCR,
1417                 ((flags & (FLAG_RADIX | FLAG_HASH_PROC_TBL)) ? LPCR_UPRT : 0) |
1418                 ((flags & FLAG_GTSE) ? LPCR_GTSE : 0),
1419                 LPCR_UPRT | LPCR_GTSE);
1420     }
1421 
1422     if (kvm_enabled()) {
1423         return kvmppc_configure_v3_mmu(cpu, flags & FLAG_RADIX,
1424                                        flags & FLAG_GTSE, cproc);
1425     }
1426     return H_SUCCESS;
1427 }
1428 
1429 #define H_SIGNAL_SYS_RESET_ALL         -1
1430 #define H_SIGNAL_SYS_RESET_ALLBUTSELF  -2
1431 
1432 static target_ulong h_signal_sys_reset(PowerPCCPU *cpu,
1433                                        sPAPRMachineState *spapr,
1434                                        target_ulong opcode, target_ulong *args)
1435 {
1436     target_long target = args[0];
1437     CPUState *cs;
1438 
1439     if (target < 0) {
1440         /* Broadcast */
1441         if (target < H_SIGNAL_SYS_RESET_ALLBUTSELF) {
1442             return H_PARAMETER;
1443         }
1444 
1445         CPU_FOREACH(cs) {
1446             PowerPCCPU *c = POWERPC_CPU(cs);
1447 
1448             if (target == H_SIGNAL_SYS_RESET_ALLBUTSELF) {
1449                 if (c == cpu) {
1450                     continue;
1451                 }
1452             }
1453             run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
1454         }
1455         return H_SUCCESS;
1456 
1457     } else {
1458         /* Unicast */
1459         cs = CPU(spapr_find_cpu(target));
1460         if (cs) {
1461             run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
1462             return H_SUCCESS;
1463         }
1464         return H_PARAMETER;
1465     }
1466 }
1467 
1468 static uint32_t cas_check_pvr(sPAPRMachineState *spapr, PowerPCCPU *cpu,
1469                               target_ulong *addr, bool *raw_mode_supported,
1470                               Error **errp)
1471 {
1472     bool explicit_match = false; /* Matched the CPU's real PVR */
1473     uint32_t max_compat = spapr->max_compat_pvr;
1474     uint32_t best_compat = 0;
1475     int i;
1476 
1477     /*
1478      * We scan the supplied table of PVRs looking for two things
1479      *   1. Is our real CPU PVR in the list?
1480      *   2. What's the "best" listed logical PVR
1481      */
1482     for (i = 0; i < 512; ++i) {
1483         uint32_t pvr, pvr_mask;
1484 
1485         pvr_mask = ldl_be_phys(&address_space_memory, *addr);
1486         pvr = ldl_be_phys(&address_space_memory, *addr + 4);
1487         *addr += 8;
1488 
1489         if (~pvr_mask & pvr) {
1490             break; /* Terminator record */
1491         }
1492 
1493         if ((cpu->env.spr[SPR_PVR] & pvr_mask) == (pvr & pvr_mask)) {
1494             explicit_match = true;
1495         } else {
1496             if (ppc_check_compat(cpu, pvr, best_compat, max_compat)) {
1497                 best_compat = pvr;
1498             }
1499         }
1500     }
1501 
1502     if ((best_compat == 0) && (!explicit_match || max_compat)) {
1503         /* We couldn't find a suitable compatibility mode, and either
1504          * the guest doesn't support "raw" mode for this CPU, or raw
1505          * mode is disabled because a maximum compat mode is set */
1506         error_setg(errp, "Couldn't negotiate a suitable PVR during CAS");
1507         return 0;
1508     }
1509 
1510     *raw_mode_supported = explicit_match;
1511 
1512     /* Parsing finished */
1513     trace_spapr_cas_pvr(cpu->compat_pvr, explicit_match, best_compat);
1514 
1515     return best_compat;
1516 }
1517 
1518 static target_ulong h_client_architecture_support(PowerPCCPU *cpu,
1519                                                   sPAPRMachineState *spapr,
1520                                                   target_ulong opcode,
1521                                                   target_ulong *args)
1522 {
1523     /* Working address in data buffer */
1524     target_ulong addr = ppc64_phys_to_real(args[0]);
1525     target_ulong ov_table;
1526     uint32_t cas_pvr;
1527     sPAPROptionVector *ov1_guest, *ov5_guest, *ov5_cas_old, *ov5_updates;
1528     bool guest_radix;
1529     Error *local_err = NULL;
1530     bool raw_mode_supported = false;
1531 
1532     cas_pvr = cas_check_pvr(spapr, cpu, &addr, &raw_mode_supported, &local_err);
1533     if (local_err) {
1534         error_report_err(local_err);
1535         return H_HARDWARE;
1536     }
1537 
1538     /* Update CPUs */
1539     if (cpu->compat_pvr != cas_pvr) {
1540         ppc_set_compat_all(cas_pvr, &local_err);
1541         if (local_err) {
1542             /* We fail to set compat mode (likely because running with KVM PR),
1543              * but maybe we can fallback to raw mode if the guest supports it.
1544              */
1545             if (!raw_mode_supported) {
1546                 error_report_err(local_err);
1547                 return H_HARDWARE;
1548             }
1549             local_err = NULL;
1550         }
1551     }
1552 
1553     /* For the future use: here @ov_table points to the first option vector */
1554     ov_table = addr;
1555 
1556     ov1_guest = spapr_ovec_parse_vector(ov_table, 1);
1557     ov5_guest = spapr_ovec_parse_vector(ov_table, 5);
1558     if (spapr_ovec_test(ov5_guest, OV5_MMU_BOTH)) {
1559         error_report("guest requested hash and radix MMU, which is invalid.");
1560         exit(EXIT_FAILURE);
1561     }
1562     /* The radix/hash bit in byte 24 requires special handling: */
1563     guest_radix = spapr_ovec_test(ov5_guest, OV5_MMU_RADIX_300);
1564     spapr_ovec_clear(ov5_guest, OV5_MMU_RADIX_300);
1565 
1566     /*
1567      * HPT resizing is a bit of a special case, because when enabled
1568      * we assume an HPT guest will support it until it says it
1569      * doesn't, instead of assuming it won't support it until it says
1570      * it does.  Strictly speaking that approach could break for
1571      * guests which don't make a CAS call, but those are so old we
1572      * don't care about them.  Without that assumption we'd have to
1573      * make at least a temporary allocation of an HPT sized for max
1574      * memory, which could be impossibly difficult under KVM HV if
1575      * maxram is large.
1576      */
1577     if (!guest_radix && !spapr_ovec_test(ov5_guest, OV5_HPT_RESIZE)) {
1578         int maxshift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1579 
1580         if (spapr->resize_hpt == SPAPR_RESIZE_HPT_REQUIRED) {
1581             error_report(
1582                 "h_client_architecture_support: Guest doesn't support HPT resizing, but resize-hpt=required");
1583             exit(1);
1584         }
1585 
1586         if (spapr->htab_shift < maxshift) {
1587             /* Guest doesn't know about HPT resizing, so we
1588              * pre-emptively resize for the maximum permitted RAM.  At
1589              * the point this is called, nothing should have been
1590              * entered into the existing HPT */
1591             spapr_reallocate_hpt(spapr, maxshift, &error_fatal);
1592             push_sregs_to_kvm_pr(spapr);
1593         }
1594     }
1595 
1596     /* NOTE: there are actually a number of ov5 bits where input from the
1597      * guest is always zero, and the platform/QEMU enables them independently
1598      * of guest input. To model these properly we'd want some sort of mask,
1599      * but since they only currently apply to memory migration as defined
1600      * by LoPAPR 1.1, 14.5.4.8, which QEMU doesn't implement, we don't need
1601      * to worry about this for now.
1602      */
1603     ov5_cas_old = spapr_ovec_clone(spapr->ov5_cas);
1604 
1605     /* also clear the radix/hash bit from the current ov5_cas bits to
1606      * be in sync with the newly ov5 bits. Else the radix bit will be
1607      * seen as being removed and this will generate a reset loop
1608      */
1609     spapr_ovec_clear(ov5_cas_old, OV5_MMU_RADIX_300);
1610 
1611     /* full range of negotiated ov5 capabilities */
1612     spapr_ovec_intersect(spapr->ov5_cas, spapr->ov5, ov5_guest);
1613     spapr_ovec_cleanup(ov5_guest);
1614     /* capabilities that have been added since CAS-generated guest reset.
1615      * if capabilities have since been removed, generate another reset
1616      */
1617     ov5_updates = spapr_ovec_new();
1618     spapr->cas_reboot = spapr_ovec_diff(ov5_updates,
1619                                         ov5_cas_old, spapr->ov5_cas);
1620     /* Now that processing is finished, set the radix/hash bit for the
1621      * guest if it requested a valid mode; otherwise terminate the boot. */
1622     if (guest_radix) {
1623         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1624             error_report("Guest requested unavailable MMU mode (radix).");
1625             exit(EXIT_FAILURE);
1626         }
1627         spapr_ovec_set(spapr->ov5_cas, OV5_MMU_RADIX_300);
1628     } else {
1629         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1630             && !kvmppc_has_cap_mmu_hash_v3()) {
1631             error_report("Guest requested unavailable MMU mode (hash).");
1632             exit(EXIT_FAILURE);
1633         }
1634     }
1635     spapr->cas_legacy_guest_workaround = !spapr_ovec_test(ov1_guest,
1636                                                           OV1_PPC_3_00);
1637     if (!spapr->cas_reboot) {
1638         /* If ppc_spapr_reset() did not set up a HPT but one is necessary
1639          * (because the guest isn't going to use radix) then set it up here. */
1640         if ((spapr->patb_entry & PATBE1_GR) && !guest_radix) {
1641             /* legacy hash or new hash: */
1642             spapr_setup_hpt_and_vrma(spapr);
1643         }
1644         spapr->cas_reboot =
1645             (spapr_h_cas_compose_response(spapr, args[1], args[2],
1646                                           ov5_updates) != 0);
1647     }
1648     spapr_ovec_cleanup(ov5_updates);
1649 
1650     if (spapr->cas_reboot) {
1651         qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
1652     }
1653 
1654     return H_SUCCESS;
1655 }
1656 
1657 static spapr_hcall_fn papr_hypercall_table[(MAX_HCALL_OPCODE / 4) + 1];
1658 static spapr_hcall_fn kvmppc_hypercall_table[KVMPPC_HCALL_MAX - KVMPPC_HCALL_BASE + 1];
1659 
1660 void spapr_register_hypercall(target_ulong opcode, spapr_hcall_fn fn)
1661 {
1662     spapr_hcall_fn *slot;
1663 
1664     if (opcode <= MAX_HCALL_OPCODE) {
1665         assert((opcode & 0x3) == 0);
1666 
1667         slot = &papr_hypercall_table[opcode / 4];
1668     } else {
1669         assert((opcode >= KVMPPC_HCALL_BASE) && (opcode <= KVMPPC_HCALL_MAX));
1670 
1671         slot = &kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE];
1672     }
1673 
1674     assert(!(*slot));
1675     *slot = fn;
1676 }
1677 
1678 target_ulong spapr_hypercall(PowerPCCPU *cpu, target_ulong opcode,
1679                              target_ulong *args)
1680 {
1681     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1682 
1683     if ((opcode <= MAX_HCALL_OPCODE)
1684         && ((opcode & 0x3) == 0)) {
1685         spapr_hcall_fn fn = papr_hypercall_table[opcode / 4];
1686 
1687         if (fn) {
1688             return fn(cpu, spapr, opcode, args);
1689         }
1690     } else if ((opcode >= KVMPPC_HCALL_BASE) &&
1691                (opcode <= KVMPPC_HCALL_MAX)) {
1692         spapr_hcall_fn fn = kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE];
1693 
1694         if (fn) {
1695             return fn(cpu, spapr, opcode, args);
1696         }
1697     }
1698 
1699     qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x" TARGET_FMT_lx "\n",
1700                   opcode);
1701     return H_FUNCTION;
1702 }
1703 
1704 static void hypercall_register_types(void)
1705 {
1706     /* hcall-pft */
1707     spapr_register_hypercall(H_ENTER, h_enter);
1708     spapr_register_hypercall(H_REMOVE, h_remove);
1709     spapr_register_hypercall(H_PROTECT, h_protect);
1710     spapr_register_hypercall(H_READ, h_read);
1711 
1712     /* hcall-bulk */
1713     spapr_register_hypercall(H_BULK_REMOVE, h_bulk_remove);
1714 
1715     /* hcall-hpt-resize */
1716     spapr_register_hypercall(H_RESIZE_HPT_PREPARE, h_resize_hpt_prepare);
1717     spapr_register_hypercall(H_RESIZE_HPT_COMMIT, h_resize_hpt_commit);
1718 
1719     /* hcall-splpar */
1720     spapr_register_hypercall(H_REGISTER_VPA, h_register_vpa);
1721     spapr_register_hypercall(H_CEDE, h_cede);
1722     spapr_register_hypercall(H_SIGNAL_SYS_RESET, h_signal_sys_reset);
1723 
1724     /* processor register resource access h-calls */
1725     spapr_register_hypercall(H_SET_SPRG0, h_set_sprg0);
1726     spapr_register_hypercall(H_SET_DABR, h_set_dabr);
1727     spapr_register_hypercall(H_SET_XDABR, h_set_xdabr);
1728     spapr_register_hypercall(H_PAGE_INIT, h_page_init);
1729     spapr_register_hypercall(H_SET_MODE, h_set_mode);
1730 
1731     /* In Memory Table MMU h-calls */
1732     spapr_register_hypercall(H_CLEAN_SLB, h_clean_slb);
1733     spapr_register_hypercall(H_INVALIDATE_PID, h_invalidate_pid);
1734     spapr_register_hypercall(H_REGISTER_PROC_TBL, h_register_process_table);
1735 
1736     /* "debugger" hcalls (also used by SLOF). Note: We do -not- differenciate
1737      * here between the "CI" and the "CACHE" variants, they will use whatever
1738      * mapping attributes qemu is using. When using KVM, the kernel will
1739      * enforce the attributes more strongly
1740      */
1741     spapr_register_hypercall(H_LOGICAL_CI_LOAD, h_logical_load);
1742     spapr_register_hypercall(H_LOGICAL_CI_STORE, h_logical_store);
1743     spapr_register_hypercall(H_LOGICAL_CACHE_LOAD, h_logical_load);
1744     spapr_register_hypercall(H_LOGICAL_CACHE_STORE, h_logical_store);
1745     spapr_register_hypercall(H_LOGICAL_ICBI, h_logical_icbi);
1746     spapr_register_hypercall(H_LOGICAL_DCBF, h_logical_dcbf);
1747     spapr_register_hypercall(KVMPPC_H_LOGICAL_MEMOP, h_logical_memop);
1748 
1749     /* qemu/KVM-PPC specific hcalls */
1750     spapr_register_hypercall(KVMPPC_H_RTAS, h_rtas);
1751 
1752     /* ibm,client-architecture-support support */
1753     spapr_register_hypercall(KVMPPC_H_CAS, h_client_architecture_support);
1754 }
1755 
1756 type_init(hypercall_register_types)
1757