xref: /openbmc/qemu/hw/ppc/spapr_hcall.c (revision 265b578c)
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 #include "hw/mem/memory-device.h"
18 
19 struct LPCRSyncState {
20     target_ulong value;
21     target_ulong mask;
22 };
23 
24 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
25 {
26     struct LPCRSyncState *s = arg.host_ptr;
27     PowerPCCPU *cpu = POWERPC_CPU(cs);
28     CPUPPCState *env = &cpu->env;
29     target_ulong lpcr;
30 
31     cpu_synchronize_state(cs);
32     lpcr = env->spr[SPR_LPCR];
33     lpcr &= ~s->mask;
34     lpcr |= s->value;
35     ppc_store_lpcr(cpu, lpcr);
36 }
37 
38 static void set_all_lpcrs(target_ulong value, target_ulong mask)
39 {
40     CPUState *cs;
41     struct LPCRSyncState s = {
42         .value = value,
43         .mask = mask
44     };
45     CPU_FOREACH(cs) {
46         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
47     }
48 }
49 
50 static bool has_spr(PowerPCCPU *cpu, int spr)
51 {
52     /* We can test whether the SPR is defined by checking for a valid name */
53     return cpu->env.spr_cb[spr].name != NULL;
54 }
55 
56 static inline bool valid_ptex(PowerPCCPU *cpu, target_ulong ptex)
57 {
58     /*
59      * hash value/pteg group index is normalized by HPT mask
60      */
61     if (((ptex & ~7ULL) / HPTES_PER_GROUP) & ~ppc_hash64_hpt_mask(cpu)) {
62         return false;
63     }
64     return true;
65 }
66 
67 static bool is_ram_address(sPAPRMachineState *spapr, hwaddr addr)
68 {
69     MachineState *machine = MACHINE(spapr);
70     DeviceMemoryState *dms = machine->device_memory;
71 
72     if (addr < machine->ram_size) {
73         return true;
74     }
75     if ((addr >= dms->base)
76         && ((addr - dms->base) < memory_region_size(&dms->mr))) {
77         return true;
78     }
79 
80     return false;
81 }
82 
83 static target_ulong h_enter(PowerPCCPU *cpu, sPAPRMachineState *spapr,
84                             target_ulong opcode, target_ulong *args)
85 {
86     target_ulong flags = args[0];
87     target_ulong ptex = args[1];
88     target_ulong pteh = args[2];
89     target_ulong ptel = args[3];
90     unsigned apshift;
91     target_ulong raddr;
92     target_ulong slot;
93     const ppc_hash_pte64_t *hptes;
94 
95     apshift = ppc_hash64_hpte_page_shift_noslb(cpu, pteh, ptel);
96     if (!apshift) {
97         /* Bad page size encoding */
98         return H_PARAMETER;
99     }
100 
101     raddr = (ptel & HPTE64_R_RPN) & ~((1ULL << apshift) - 1);
102 
103     if (is_ram_address(spapr, raddr)) {
104         /* Regular RAM - should have WIMG=0010 */
105         if ((ptel & HPTE64_R_WIMG) != HPTE64_R_M) {
106             return H_PARAMETER;
107         }
108     } else {
109         target_ulong wimg_flags;
110         /* Looks like an IO address */
111         /* FIXME: What WIMG combinations could be sensible for IO?
112          * For now we allow WIMG=010x, but are there others? */
113         /* FIXME: Should we check against registered IO addresses? */
114         wimg_flags = (ptel & (HPTE64_R_W | HPTE64_R_I | HPTE64_R_M));
115 
116         if (wimg_flags != HPTE64_R_I &&
117             wimg_flags != (HPTE64_R_I | HPTE64_R_M)) {
118             return H_PARAMETER;
119         }
120     }
121 
122     pteh &= ~0x60ULL;
123 
124     if (!valid_ptex(cpu, ptex)) {
125         return H_PARAMETER;
126     }
127 
128     slot = ptex & 7ULL;
129     ptex = ptex & ~7ULL;
130 
131     if (likely((flags & H_EXACT) == 0)) {
132         hptes = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP);
133         for (slot = 0; slot < 8; slot++) {
134             if (!(ppc_hash64_hpte0(cpu, hptes, slot) & HPTE64_V_VALID)) {
135                 break;
136             }
137         }
138         ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP);
139         if (slot == 8) {
140             return H_PTEG_FULL;
141         }
142     } else {
143         hptes = ppc_hash64_map_hptes(cpu, ptex + slot, 1);
144         if (ppc_hash64_hpte0(cpu, hptes, 0) & HPTE64_V_VALID) {
145             ppc_hash64_unmap_hptes(cpu, hptes, ptex + slot, 1);
146             return H_PTEG_FULL;
147         }
148         ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1);
149     }
150 
151     ppc_hash64_store_hpte(cpu, ptex + slot, pteh | HPTE64_V_HPTE_DIRTY, ptel);
152 
153     args[0] = ptex + slot;
154     return H_SUCCESS;
155 }
156 
157 typedef enum {
158     REMOVE_SUCCESS = 0,
159     REMOVE_NOT_FOUND = 1,
160     REMOVE_PARM = 2,
161     REMOVE_HW = 3,
162 } RemoveResult;
163 
164 static RemoveResult remove_hpte(PowerPCCPU *cpu, target_ulong ptex,
165                                 target_ulong avpn,
166                                 target_ulong flags,
167                                 target_ulong *vp, target_ulong *rp)
168 {
169     const ppc_hash_pte64_t *hptes;
170     target_ulong v, r;
171 
172     if (!valid_ptex(cpu, ptex)) {
173         return REMOVE_PARM;
174     }
175 
176     hptes = ppc_hash64_map_hptes(cpu, ptex, 1);
177     v = ppc_hash64_hpte0(cpu, hptes, 0);
178     r = ppc_hash64_hpte1(cpu, hptes, 0);
179     ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1);
180 
181     if ((v & HPTE64_V_VALID) == 0 ||
182         ((flags & H_AVPN) && (v & ~0x7fULL) != avpn) ||
183         ((flags & H_ANDCOND) && (v & avpn) != 0)) {
184         return REMOVE_NOT_FOUND;
185     }
186     *vp = v;
187     *rp = r;
188     ppc_hash64_store_hpte(cpu, ptex, HPTE64_V_HPTE_DIRTY, 0);
189     ppc_hash64_tlb_flush_hpte(cpu, ptex, v, r);
190     return REMOVE_SUCCESS;
191 }
192 
193 static target_ulong h_remove(PowerPCCPU *cpu, sPAPRMachineState *spapr,
194                              target_ulong opcode, target_ulong *args)
195 {
196     CPUPPCState *env = &cpu->env;
197     target_ulong flags = args[0];
198     target_ulong ptex = args[1];
199     target_ulong avpn = args[2];
200     RemoveResult ret;
201 
202     ret = remove_hpte(cpu, ptex, avpn, flags,
203                       &args[0], &args[1]);
204 
205     switch (ret) {
206     case REMOVE_SUCCESS:
207         check_tlb_flush(env, true);
208         return H_SUCCESS;
209 
210     case REMOVE_NOT_FOUND:
211         return H_NOT_FOUND;
212 
213     case REMOVE_PARM:
214         return H_PARAMETER;
215 
216     case REMOVE_HW:
217         return H_HARDWARE;
218     }
219 
220     g_assert_not_reached();
221 }
222 
223 #define H_BULK_REMOVE_TYPE             0xc000000000000000ULL
224 #define   H_BULK_REMOVE_REQUEST        0x4000000000000000ULL
225 #define   H_BULK_REMOVE_RESPONSE       0x8000000000000000ULL
226 #define   H_BULK_REMOVE_END            0xc000000000000000ULL
227 #define H_BULK_REMOVE_CODE             0x3000000000000000ULL
228 #define   H_BULK_REMOVE_SUCCESS        0x0000000000000000ULL
229 #define   H_BULK_REMOVE_NOT_FOUND      0x1000000000000000ULL
230 #define   H_BULK_REMOVE_PARM           0x2000000000000000ULL
231 #define   H_BULK_REMOVE_HW             0x3000000000000000ULL
232 #define H_BULK_REMOVE_RC               0x0c00000000000000ULL
233 #define H_BULK_REMOVE_FLAGS            0x0300000000000000ULL
234 #define   H_BULK_REMOVE_ABSOLUTE       0x0000000000000000ULL
235 #define   H_BULK_REMOVE_ANDCOND        0x0100000000000000ULL
236 #define   H_BULK_REMOVE_AVPN           0x0200000000000000ULL
237 #define H_BULK_REMOVE_PTEX             0x00ffffffffffffffULL
238 
239 #define H_BULK_REMOVE_MAX_BATCH        4
240 
241 static target_ulong h_bulk_remove(PowerPCCPU *cpu, sPAPRMachineState *spapr,
242                                   target_ulong opcode, target_ulong *args)
243 {
244     CPUPPCState *env = &cpu->env;
245     int i;
246     target_ulong rc = H_SUCCESS;
247 
248     for (i = 0; i < H_BULK_REMOVE_MAX_BATCH; i++) {
249         target_ulong *tsh = &args[i*2];
250         target_ulong tsl = args[i*2 + 1];
251         target_ulong v, r, ret;
252 
253         if ((*tsh & H_BULK_REMOVE_TYPE) == H_BULK_REMOVE_END) {
254             break;
255         } else if ((*tsh & H_BULK_REMOVE_TYPE) != H_BULK_REMOVE_REQUEST) {
256             return H_PARAMETER;
257         }
258 
259         *tsh &= H_BULK_REMOVE_PTEX | H_BULK_REMOVE_FLAGS;
260         *tsh |= H_BULK_REMOVE_RESPONSE;
261 
262         if ((*tsh & H_BULK_REMOVE_ANDCOND) && (*tsh & H_BULK_REMOVE_AVPN)) {
263             *tsh |= H_BULK_REMOVE_PARM;
264             return H_PARAMETER;
265         }
266 
267         ret = remove_hpte(cpu, *tsh & H_BULK_REMOVE_PTEX, tsl,
268                           (*tsh & H_BULK_REMOVE_FLAGS) >> 26,
269                           &v, &r);
270 
271         *tsh |= ret << 60;
272 
273         switch (ret) {
274         case REMOVE_SUCCESS:
275             *tsh |= (r & (HPTE64_R_C | HPTE64_R_R)) << 43;
276             break;
277 
278         case REMOVE_PARM:
279             rc = H_PARAMETER;
280             goto exit;
281 
282         case REMOVE_HW:
283             rc = H_HARDWARE;
284             goto exit;
285         }
286     }
287  exit:
288     check_tlb_flush(env, true);
289 
290     return rc;
291 }
292 
293 static target_ulong h_protect(PowerPCCPU *cpu, sPAPRMachineState *spapr,
294                               target_ulong opcode, target_ulong *args)
295 {
296     CPUPPCState *env = &cpu->env;
297     target_ulong flags = args[0];
298     target_ulong ptex = args[1];
299     target_ulong avpn = args[2];
300     const ppc_hash_pte64_t *hptes;
301     target_ulong v, r;
302 
303     if (!valid_ptex(cpu, ptex)) {
304         return H_PARAMETER;
305     }
306 
307     hptes = ppc_hash64_map_hptes(cpu, ptex, 1);
308     v = ppc_hash64_hpte0(cpu, hptes, 0);
309     r = ppc_hash64_hpte1(cpu, hptes, 0);
310     ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1);
311 
312     if ((v & HPTE64_V_VALID) == 0 ||
313         ((flags & H_AVPN) && (v & ~0x7fULL) != avpn)) {
314         return H_NOT_FOUND;
315     }
316 
317     r &= ~(HPTE64_R_PP0 | HPTE64_R_PP | HPTE64_R_N |
318            HPTE64_R_KEY_HI | HPTE64_R_KEY_LO);
319     r |= (flags << 55) & HPTE64_R_PP0;
320     r |= (flags << 48) & HPTE64_R_KEY_HI;
321     r |= flags & (HPTE64_R_PP | HPTE64_R_N | HPTE64_R_KEY_LO);
322     ppc_hash64_store_hpte(cpu, ptex,
323                           (v & ~HPTE64_V_VALID) | HPTE64_V_HPTE_DIRTY, 0);
324     ppc_hash64_tlb_flush_hpte(cpu, ptex, v, r);
325     /* Flush the tlb */
326     check_tlb_flush(env, true);
327     /* Don't need a memory barrier, due to qemu's global lock */
328     ppc_hash64_store_hpte(cpu, ptex, v | HPTE64_V_HPTE_DIRTY, r);
329     return H_SUCCESS;
330 }
331 
332 static target_ulong h_read(PowerPCCPU *cpu, sPAPRMachineState *spapr,
333                            target_ulong opcode, target_ulong *args)
334 {
335     target_ulong flags = args[0];
336     target_ulong ptex = args[1];
337     uint8_t *hpte;
338     int i, ridx, n_entries = 1;
339 
340     if (!valid_ptex(cpu, ptex)) {
341         return H_PARAMETER;
342     }
343 
344     if (flags & H_READ_4) {
345         /* Clear the two low order bits */
346         ptex &= ~(3ULL);
347         n_entries = 4;
348     }
349 
350     hpte = spapr->htab + (ptex * HASH_PTE_SIZE_64);
351 
352     for (i = 0, ridx = 0; i < n_entries; i++) {
353         args[ridx++] = ldq_p(hpte);
354         args[ridx++] = ldq_p(hpte + (HASH_PTE_SIZE_64/2));
355         hpte += HASH_PTE_SIZE_64;
356     }
357 
358     return H_SUCCESS;
359 }
360 
361 struct sPAPRPendingHPT {
362     /* These fields are read-only after initialization */
363     int shift;
364     QemuThread thread;
365 
366     /* These fields are protected by the BQL */
367     bool complete;
368 
369     /* These fields are private to the preparation thread if
370      * !complete, otherwise protected by the BQL */
371     int ret;
372     void *hpt;
373 };
374 
375 static void free_pending_hpt(sPAPRPendingHPT *pending)
376 {
377     if (pending->hpt) {
378         qemu_vfree(pending->hpt);
379     }
380 
381     g_free(pending);
382 }
383 
384 static void *hpt_prepare_thread(void *opaque)
385 {
386     sPAPRPendingHPT *pending = opaque;
387     size_t size = 1ULL << pending->shift;
388 
389     pending->hpt = qemu_memalign(size, size);
390     if (pending->hpt) {
391         memset(pending->hpt, 0, size);
392         pending->ret = H_SUCCESS;
393     } else {
394         pending->ret = H_NO_MEM;
395     }
396 
397     qemu_mutex_lock_iothread();
398 
399     if (SPAPR_MACHINE(qdev_get_machine())->pending_hpt == pending) {
400         /* Ready to go */
401         pending->complete = true;
402     } else {
403         /* We've been cancelled, clean ourselves up */
404         free_pending_hpt(pending);
405     }
406 
407     qemu_mutex_unlock_iothread();
408     return NULL;
409 }
410 
411 /* Must be called with BQL held */
412 static void cancel_hpt_prepare(sPAPRMachineState *spapr)
413 {
414     sPAPRPendingHPT *pending = spapr->pending_hpt;
415 
416     /* Let the thread know it's cancelled */
417     spapr->pending_hpt = NULL;
418 
419     if (!pending) {
420         /* Nothing to do */
421         return;
422     }
423 
424     if (!pending->complete) {
425         /* thread will clean itself up */
426         return;
427     }
428 
429     free_pending_hpt(pending);
430 }
431 
432 /* Convert a return code from the KVM ioctl()s implementing resize HPT
433  * into a PAPR hypercall return code */
434 static target_ulong resize_hpt_convert_rc(int ret)
435 {
436     if (ret >= 100000) {
437         return H_LONG_BUSY_ORDER_100_SEC;
438     } else if (ret >= 10000) {
439         return H_LONG_BUSY_ORDER_10_SEC;
440     } else if (ret >= 1000) {
441         return H_LONG_BUSY_ORDER_1_SEC;
442     } else if (ret >= 100) {
443         return H_LONG_BUSY_ORDER_100_MSEC;
444     } else if (ret >= 10) {
445         return H_LONG_BUSY_ORDER_10_MSEC;
446     } else if (ret > 0) {
447         return H_LONG_BUSY_ORDER_1_MSEC;
448     }
449 
450     switch (ret) {
451     case 0:
452         return H_SUCCESS;
453     case -EPERM:
454         return H_AUTHORITY;
455     case -EINVAL:
456         return H_PARAMETER;
457     case -ENXIO:
458         return H_CLOSED;
459     case -ENOSPC:
460         return H_PTEG_FULL;
461     case -EBUSY:
462         return H_BUSY;
463     case -ENOMEM:
464         return H_NO_MEM;
465     default:
466         return H_HARDWARE;
467     }
468 }
469 
470 static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
471                                          sPAPRMachineState *spapr,
472                                          target_ulong opcode,
473                                          target_ulong *args)
474 {
475     target_ulong flags = args[0];
476     int shift = args[1];
477     sPAPRPendingHPT *pending = spapr->pending_hpt;
478     uint64_t current_ram_size;
479     int rc;
480 
481     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
482         return H_AUTHORITY;
483     }
484 
485     if (!spapr->htab_shift) {
486         /* Radix guest, no HPT */
487         return H_NOT_AVAILABLE;
488     }
489 
490     trace_spapr_h_resize_hpt_prepare(flags, shift);
491 
492     if (flags != 0) {
493         return H_PARAMETER;
494     }
495 
496     if (shift && ((shift < 18) || (shift > 46))) {
497         return H_PARAMETER;
498     }
499 
500     current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
501 
502     /* We only allow the guest to allocate an HPT one order above what
503      * we'd normally give them (to stop a small guest claiming a huge
504      * chunk of resources in the HPT */
505     if (shift > (spapr_hpt_shift_for_ramsize(current_ram_size) + 1)) {
506         return H_RESOURCE;
507     }
508 
509     rc = kvmppc_resize_hpt_prepare(cpu, flags, shift);
510     if (rc != -ENOSYS) {
511         return resize_hpt_convert_rc(rc);
512     }
513 
514     if (pending) {
515         /* something already in progress */
516         if (pending->shift == shift) {
517             /* and it's suitable */
518             if (pending->complete) {
519                 return pending->ret;
520             } else {
521                 return H_LONG_BUSY_ORDER_100_MSEC;
522             }
523         }
524 
525         /* not suitable, cancel and replace */
526         cancel_hpt_prepare(spapr);
527     }
528 
529     if (!shift) {
530         /* nothing to do */
531         return H_SUCCESS;
532     }
533 
534     /* start new prepare */
535 
536     pending = g_new0(sPAPRPendingHPT, 1);
537     pending->shift = shift;
538     pending->ret = H_HARDWARE;
539 
540     qemu_thread_create(&pending->thread, "sPAPR HPT prepare",
541                        hpt_prepare_thread, pending, QEMU_THREAD_DETACHED);
542 
543     spapr->pending_hpt = pending;
544 
545     /* In theory we could estimate the time more accurately based on
546      * the new size, but there's not much point */
547     return H_LONG_BUSY_ORDER_100_MSEC;
548 }
549 
550 static uint64_t new_hpte_load0(void *htab, uint64_t pteg, int slot)
551 {
552     uint8_t *addr = htab;
553 
554     addr += pteg * HASH_PTEG_SIZE_64;
555     addr += slot * HASH_PTE_SIZE_64;
556     return  ldq_p(addr);
557 }
558 
559 static void new_hpte_store(void *htab, uint64_t pteg, int slot,
560                            uint64_t pte0, uint64_t pte1)
561 {
562     uint8_t *addr = htab;
563 
564     addr += pteg * HASH_PTEG_SIZE_64;
565     addr += slot * HASH_PTE_SIZE_64;
566 
567     stq_p(addr, pte0);
568     stq_p(addr + HASH_PTE_SIZE_64 / 2, pte1);
569 }
570 
571 static int rehash_hpte(PowerPCCPU *cpu,
572                        const ppc_hash_pte64_t *hptes,
573                        void *old_hpt, uint64_t oldsize,
574                        void *new_hpt, uint64_t newsize,
575                        uint64_t pteg, int slot)
576 {
577     uint64_t old_hash_mask = (oldsize >> 7) - 1;
578     uint64_t new_hash_mask = (newsize >> 7) - 1;
579     target_ulong pte0 = ppc_hash64_hpte0(cpu, hptes, slot);
580     target_ulong pte1;
581     uint64_t avpn;
582     unsigned base_pg_shift;
583     uint64_t hash, new_pteg, replace_pte0;
584 
585     if (!(pte0 & HPTE64_V_VALID) || !(pte0 & HPTE64_V_BOLTED)) {
586         return H_SUCCESS;
587     }
588 
589     pte1 = ppc_hash64_hpte1(cpu, hptes, slot);
590 
591     base_pg_shift = ppc_hash64_hpte_page_shift_noslb(cpu, pte0, pte1);
592     assert(base_pg_shift); /* H_ENTER shouldn't allow a bad encoding */
593     avpn = HPTE64_V_AVPN_VAL(pte0) & ~(((1ULL << base_pg_shift) - 1) >> 23);
594 
595     if (pte0 & HPTE64_V_SECONDARY) {
596         pteg = ~pteg;
597     }
598 
599     if ((pte0 & HPTE64_V_SSIZE) == HPTE64_V_SSIZE_256M) {
600         uint64_t offset, vsid;
601 
602         /* We only have 28 - 23 bits of offset in avpn */
603         offset = (avpn & 0x1f) << 23;
604         vsid = avpn >> 5;
605         /* We can find more bits from the pteg value */
606         if (base_pg_shift < 23) {
607             offset |= ((vsid ^ pteg) & old_hash_mask) << base_pg_shift;
608         }
609 
610         hash = vsid ^ (offset >> base_pg_shift);
611     } else if ((pte0 & HPTE64_V_SSIZE) == HPTE64_V_SSIZE_1T) {
612         uint64_t offset, vsid;
613 
614         /* We only have 40 - 23 bits of seg_off in avpn */
615         offset = (avpn & 0x1ffff) << 23;
616         vsid = avpn >> 17;
617         if (base_pg_shift < 23) {
618             offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask)
619                 << base_pg_shift;
620         }
621 
622         hash = vsid ^ (vsid << 25) ^ (offset >> base_pg_shift);
623     } else {
624         error_report("rehash_pte: Bad segment size in HPTE");
625         return H_HARDWARE;
626     }
627 
628     new_pteg = hash & new_hash_mask;
629     if (pte0 & HPTE64_V_SECONDARY) {
630         assert(~pteg == (hash & old_hash_mask));
631         new_pteg = ~new_pteg;
632     } else {
633         assert(pteg == (hash & old_hash_mask));
634     }
635     assert((oldsize != newsize) || (pteg == new_pteg));
636     replace_pte0 = new_hpte_load0(new_hpt, new_pteg, slot);
637     /*
638      * Strictly speaking, we don't need all these tests, since we only
639      * ever rehash bolted HPTEs.  We might in future handle non-bolted
640      * HPTEs, though so make the logic correct for those cases as
641      * well.
642      */
643     if (replace_pte0 & HPTE64_V_VALID) {
644         assert(newsize < oldsize);
645         if (replace_pte0 & HPTE64_V_BOLTED) {
646             if (pte0 & HPTE64_V_BOLTED) {
647                 /* Bolted collision, nothing we can do */
648                 return H_PTEG_FULL;
649             } else {
650                 /* Discard this hpte */
651                 return H_SUCCESS;
652             }
653         }
654     }
655 
656     new_hpte_store(new_hpt, new_pteg, slot, pte0, pte1);
657     return H_SUCCESS;
658 }
659 
660 static int rehash_hpt(PowerPCCPU *cpu,
661                       void *old_hpt, uint64_t oldsize,
662                       void *new_hpt, uint64_t newsize)
663 {
664     uint64_t n_ptegs = oldsize >> 7;
665     uint64_t pteg;
666     int slot;
667     int rc;
668 
669     for (pteg = 0; pteg < n_ptegs; pteg++) {
670         hwaddr ptex = pteg * HPTES_PER_GROUP;
671         const ppc_hash_pte64_t *hptes
672             = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP);
673 
674         if (!hptes) {
675             return H_HARDWARE;
676         }
677 
678         for (slot = 0; slot < HPTES_PER_GROUP; slot++) {
679             rc = rehash_hpte(cpu, hptes, old_hpt, oldsize, new_hpt, newsize,
680                              pteg, slot);
681             if (rc != H_SUCCESS) {
682                 ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP);
683                 return rc;
684             }
685         }
686         ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP);
687     }
688 
689     return H_SUCCESS;
690 }
691 
692 static void do_push_sregs_to_kvm_pr(CPUState *cs, run_on_cpu_data data)
693 {
694     int ret;
695 
696     cpu_synchronize_state(cs);
697 
698     ret = kvmppc_put_books_sregs(POWERPC_CPU(cs));
699     if (ret < 0) {
700         error_report("failed to push sregs to KVM: %s", strerror(-ret));
701         exit(1);
702     }
703 }
704 
705 static void push_sregs_to_kvm_pr(sPAPRMachineState *spapr)
706 {
707     CPUState *cs;
708 
709     /*
710      * This is a hack for the benefit of KVM PR - it abuses the SDR1
711      * slot in kvm_sregs to communicate the userspace address of the
712      * HPT
713      */
714     if (!kvm_enabled() || !spapr->htab) {
715         return;
716     }
717 
718     CPU_FOREACH(cs) {
719         run_on_cpu(cs, do_push_sregs_to_kvm_pr, RUN_ON_CPU_NULL);
720     }
721 }
722 
723 static target_ulong h_resize_hpt_commit(PowerPCCPU *cpu,
724                                         sPAPRMachineState *spapr,
725                                         target_ulong opcode,
726                                         target_ulong *args)
727 {
728     target_ulong flags = args[0];
729     target_ulong shift = args[1];
730     sPAPRPendingHPT *pending = spapr->pending_hpt;
731     int rc;
732     size_t newsize;
733 
734     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
735         return H_AUTHORITY;
736     }
737 
738     if (!spapr->htab_shift) {
739         /* Radix guest, no HPT */
740         return H_NOT_AVAILABLE;
741     }
742 
743     trace_spapr_h_resize_hpt_commit(flags, shift);
744 
745     rc = kvmppc_resize_hpt_commit(cpu, flags, shift);
746     if (rc != -ENOSYS) {
747         rc = resize_hpt_convert_rc(rc);
748         if (rc == H_SUCCESS) {
749             /* Need to set the new htab_shift in the machine state */
750             spapr->htab_shift = shift;
751         }
752         return rc;
753     }
754 
755     if (flags != 0) {
756         return H_PARAMETER;
757     }
758 
759     if (!pending || (pending->shift != shift)) {
760         /* no matching prepare */
761         return H_CLOSED;
762     }
763 
764     if (!pending->complete) {
765         /* prepare has not completed */
766         return H_BUSY;
767     }
768 
769     /* Shouldn't have got past PREPARE without an HPT */
770     g_assert(spapr->htab_shift);
771 
772     newsize = 1ULL << pending->shift;
773     rc = rehash_hpt(cpu, spapr->htab, HTAB_SIZE(spapr),
774                     pending->hpt, newsize);
775     if (rc == H_SUCCESS) {
776         qemu_vfree(spapr->htab);
777         spapr->htab = pending->hpt;
778         spapr->htab_shift = pending->shift;
779 
780         push_sregs_to_kvm_pr(spapr);
781 
782         pending->hpt = NULL; /* so it's not free()d */
783     }
784 
785     /* Clean up */
786     spapr->pending_hpt = NULL;
787     free_pending_hpt(pending);
788 
789     return rc;
790 }
791 
792 static target_ulong h_set_sprg0(PowerPCCPU *cpu, sPAPRMachineState *spapr,
793                                 target_ulong opcode, target_ulong *args)
794 {
795     cpu_synchronize_state(CPU(cpu));
796     cpu->env.spr[SPR_SPRG0] = args[0];
797 
798     return H_SUCCESS;
799 }
800 
801 static target_ulong h_set_dabr(PowerPCCPU *cpu, sPAPRMachineState *spapr,
802                                target_ulong opcode, target_ulong *args)
803 {
804     if (!has_spr(cpu, SPR_DABR)) {
805         return H_HARDWARE;              /* DABR register not available */
806     }
807     cpu_synchronize_state(CPU(cpu));
808 
809     if (has_spr(cpu, SPR_DABRX)) {
810         cpu->env.spr[SPR_DABRX] = 0x3;  /* Use Problem and Privileged state */
811     } else if (!(args[0] & 0x4)) {      /* Breakpoint Translation set? */
812         return H_RESERVED_DABR;
813     }
814 
815     cpu->env.spr[SPR_DABR] = args[0];
816     return H_SUCCESS;
817 }
818 
819 static target_ulong h_set_xdabr(PowerPCCPU *cpu, sPAPRMachineState *spapr,
820                                 target_ulong opcode, target_ulong *args)
821 {
822     target_ulong dabrx = args[1];
823 
824     if (!has_spr(cpu, SPR_DABR) || !has_spr(cpu, SPR_DABRX)) {
825         return H_HARDWARE;
826     }
827 
828     if ((dabrx & ~0xfULL) != 0 || (dabrx & H_DABRX_HYPERVISOR) != 0
829         || (dabrx & (H_DABRX_KERNEL | H_DABRX_USER)) == 0) {
830         return H_PARAMETER;
831     }
832 
833     cpu_synchronize_state(CPU(cpu));
834     cpu->env.spr[SPR_DABRX] = dabrx;
835     cpu->env.spr[SPR_DABR] = args[0];
836 
837     return H_SUCCESS;
838 }
839 
840 static target_ulong h_page_init(PowerPCCPU *cpu, sPAPRMachineState *spapr,
841                                 target_ulong opcode, target_ulong *args)
842 {
843     target_ulong flags = args[0];
844     hwaddr dst = args[1];
845     hwaddr src = args[2];
846     hwaddr len = TARGET_PAGE_SIZE;
847     uint8_t *pdst, *psrc;
848     target_long ret = H_SUCCESS;
849 
850     if (flags & ~(H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE
851                   | H_COPY_PAGE | H_ZERO_PAGE)) {
852         qemu_log_mask(LOG_UNIMP, "h_page_init: Bad flags (" TARGET_FMT_lx "\n",
853                       flags);
854         return H_PARAMETER;
855     }
856 
857     /* Map-in destination */
858     if (!is_ram_address(spapr, dst) || (dst & ~TARGET_PAGE_MASK) != 0) {
859         return H_PARAMETER;
860     }
861     pdst = cpu_physical_memory_map(dst, &len, 1);
862     if (!pdst || len != TARGET_PAGE_SIZE) {
863         return H_PARAMETER;
864     }
865 
866     if (flags & H_COPY_PAGE) {
867         /* Map-in source, copy to destination, and unmap source again */
868         if (!is_ram_address(spapr, src) || (src & ~TARGET_PAGE_MASK) != 0) {
869             ret = H_PARAMETER;
870             goto unmap_out;
871         }
872         psrc = cpu_physical_memory_map(src, &len, 0);
873         if (!psrc || len != TARGET_PAGE_SIZE) {
874             ret = H_PARAMETER;
875             goto unmap_out;
876         }
877         memcpy(pdst, psrc, len);
878         cpu_physical_memory_unmap(psrc, len, 0, len);
879     } else if (flags & H_ZERO_PAGE) {
880         memset(pdst, 0, len);          /* Just clear the destination page */
881     }
882 
883     if (kvm_enabled() && (flags & H_ICACHE_SYNCHRONIZE) != 0) {
884         kvmppc_dcbst_range(cpu, pdst, len);
885     }
886     if (flags & (H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE)) {
887         if (kvm_enabled()) {
888             kvmppc_icbi_range(cpu, pdst, len);
889         } else {
890             tb_flush(CPU(cpu));
891         }
892     }
893 
894 unmap_out:
895     cpu_physical_memory_unmap(pdst, TARGET_PAGE_SIZE, 1, len);
896     return ret;
897 }
898 
899 #define FLAGS_REGISTER_VPA         0x0000200000000000ULL
900 #define FLAGS_REGISTER_DTL         0x0000400000000000ULL
901 #define FLAGS_REGISTER_SLBSHADOW   0x0000600000000000ULL
902 #define FLAGS_DEREGISTER_VPA       0x0000a00000000000ULL
903 #define FLAGS_DEREGISTER_DTL       0x0000c00000000000ULL
904 #define FLAGS_DEREGISTER_SLBSHADOW 0x0000e00000000000ULL
905 
906 #define VPA_MIN_SIZE           640
907 #define VPA_SIZE_OFFSET        0x4
908 #define VPA_SHARED_PROC_OFFSET 0x9
909 #define VPA_SHARED_PROC_VAL    0x2
910 
911 static target_ulong register_vpa(CPUPPCState *env, target_ulong vpa)
912 {
913     CPUState *cs = CPU(ppc_env_get_cpu(env));
914     uint16_t size;
915     uint8_t tmp;
916 
917     if (vpa == 0) {
918         hcall_dprintf("Can't cope with registering a VPA at logical 0\n");
919         return H_HARDWARE;
920     }
921 
922     if (vpa % env->dcache_line_size) {
923         return H_PARAMETER;
924     }
925     /* FIXME: bounds check the address */
926 
927     size = lduw_be_phys(cs->as, vpa + 0x4);
928 
929     if (size < VPA_MIN_SIZE) {
930         return H_PARAMETER;
931     }
932 
933     /* VPA is not allowed to cross a page boundary */
934     if ((vpa / 4096) != ((vpa + size - 1) / 4096)) {
935         return H_PARAMETER;
936     }
937 
938     env->vpa_addr = vpa;
939 
940     tmp = ldub_phys(cs->as, env->vpa_addr + VPA_SHARED_PROC_OFFSET);
941     tmp |= VPA_SHARED_PROC_VAL;
942     stb_phys(cs->as, env->vpa_addr + VPA_SHARED_PROC_OFFSET, tmp);
943 
944     return H_SUCCESS;
945 }
946 
947 static target_ulong deregister_vpa(CPUPPCState *env, target_ulong vpa)
948 {
949     if (env->slb_shadow_addr) {
950         return H_RESOURCE;
951     }
952 
953     if (env->dtl_addr) {
954         return H_RESOURCE;
955     }
956 
957     env->vpa_addr = 0;
958     return H_SUCCESS;
959 }
960 
961 static target_ulong register_slb_shadow(CPUPPCState *env, target_ulong addr)
962 {
963     CPUState *cs = CPU(ppc_env_get_cpu(env));
964     uint32_t size;
965 
966     if (addr == 0) {
967         hcall_dprintf("Can't cope with SLB shadow at logical 0\n");
968         return H_HARDWARE;
969     }
970 
971     size = ldl_be_phys(cs->as, addr + 0x4);
972     if (size < 0x8) {
973         return H_PARAMETER;
974     }
975 
976     if ((addr / 4096) != ((addr + size - 1) / 4096)) {
977         return H_PARAMETER;
978     }
979 
980     if (!env->vpa_addr) {
981         return H_RESOURCE;
982     }
983 
984     env->slb_shadow_addr = addr;
985     env->slb_shadow_size = size;
986 
987     return H_SUCCESS;
988 }
989 
990 static target_ulong deregister_slb_shadow(CPUPPCState *env, target_ulong addr)
991 {
992     env->slb_shadow_addr = 0;
993     env->slb_shadow_size = 0;
994     return H_SUCCESS;
995 }
996 
997 static target_ulong register_dtl(CPUPPCState *env, target_ulong addr)
998 {
999     CPUState *cs = CPU(ppc_env_get_cpu(env));
1000     uint32_t size;
1001 
1002     if (addr == 0) {
1003         hcall_dprintf("Can't cope with DTL at logical 0\n");
1004         return H_HARDWARE;
1005     }
1006 
1007     size = ldl_be_phys(cs->as, addr + 0x4);
1008 
1009     if (size < 48) {
1010         return H_PARAMETER;
1011     }
1012 
1013     if (!env->vpa_addr) {
1014         return H_RESOURCE;
1015     }
1016 
1017     env->dtl_addr = addr;
1018     env->dtl_size = size;
1019 
1020     return H_SUCCESS;
1021 }
1022 
1023 static target_ulong deregister_dtl(CPUPPCState *env, target_ulong addr)
1024 {
1025     env->dtl_addr = 0;
1026     env->dtl_size = 0;
1027 
1028     return H_SUCCESS;
1029 }
1030 
1031 static target_ulong h_register_vpa(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1032                                    target_ulong opcode, target_ulong *args)
1033 {
1034     target_ulong flags = args[0];
1035     target_ulong procno = args[1];
1036     target_ulong vpa = args[2];
1037     target_ulong ret = H_PARAMETER;
1038     CPUPPCState *tenv;
1039     PowerPCCPU *tcpu;
1040 
1041     tcpu = spapr_find_cpu(procno);
1042     if (!tcpu) {
1043         return H_PARAMETER;
1044     }
1045     tenv = &tcpu->env;
1046 
1047     switch (flags) {
1048     case FLAGS_REGISTER_VPA:
1049         ret = register_vpa(tenv, vpa);
1050         break;
1051 
1052     case FLAGS_DEREGISTER_VPA:
1053         ret = deregister_vpa(tenv, vpa);
1054         break;
1055 
1056     case FLAGS_REGISTER_SLBSHADOW:
1057         ret = register_slb_shadow(tenv, vpa);
1058         break;
1059 
1060     case FLAGS_DEREGISTER_SLBSHADOW:
1061         ret = deregister_slb_shadow(tenv, vpa);
1062         break;
1063 
1064     case FLAGS_REGISTER_DTL:
1065         ret = register_dtl(tenv, vpa);
1066         break;
1067 
1068     case FLAGS_DEREGISTER_DTL:
1069         ret = deregister_dtl(tenv, vpa);
1070         break;
1071     }
1072 
1073     return ret;
1074 }
1075 
1076 static target_ulong h_cede(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1077                            target_ulong opcode, target_ulong *args)
1078 {
1079     CPUPPCState *env = &cpu->env;
1080     CPUState *cs = CPU(cpu);
1081 
1082     env->msr |= (1ULL << MSR_EE);
1083     hreg_compute_hflags(env);
1084     if (!cpu_has_work(cs)) {
1085         cs->halted = 1;
1086         cs->exception_index = EXCP_HLT;
1087         cs->exit_request = 1;
1088     }
1089     return H_SUCCESS;
1090 }
1091 
1092 static target_ulong h_rtas(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1093                            target_ulong opcode, target_ulong *args)
1094 {
1095     target_ulong rtas_r3 = args[0];
1096     uint32_t token = rtas_ld(rtas_r3, 0);
1097     uint32_t nargs = rtas_ld(rtas_r3, 1);
1098     uint32_t nret = rtas_ld(rtas_r3, 2);
1099 
1100     return spapr_rtas_call(cpu, spapr, token, nargs, rtas_r3 + 12,
1101                            nret, rtas_r3 + 12 + 4*nargs);
1102 }
1103 
1104 static target_ulong h_logical_load(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1105                                    target_ulong opcode, target_ulong *args)
1106 {
1107     CPUState *cs = CPU(cpu);
1108     target_ulong size = args[0];
1109     target_ulong addr = args[1];
1110 
1111     switch (size) {
1112     case 1:
1113         args[0] = ldub_phys(cs->as, addr);
1114         return H_SUCCESS;
1115     case 2:
1116         args[0] = lduw_phys(cs->as, addr);
1117         return H_SUCCESS;
1118     case 4:
1119         args[0] = ldl_phys(cs->as, addr);
1120         return H_SUCCESS;
1121     case 8:
1122         args[0] = ldq_phys(cs->as, addr);
1123         return H_SUCCESS;
1124     }
1125     return H_PARAMETER;
1126 }
1127 
1128 static target_ulong h_logical_store(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1129                                     target_ulong opcode, target_ulong *args)
1130 {
1131     CPUState *cs = CPU(cpu);
1132 
1133     target_ulong size = args[0];
1134     target_ulong addr = args[1];
1135     target_ulong val  = args[2];
1136 
1137     switch (size) {
1138     case 1:
1139         stb_phys(cs->as, addr, val);
1140         return H_SUCCESS;
1141     case 2:
1142         stw_phys(cs->as, addr, val);
1143         return H_SUCCESS;
1144     case 4:
1145         stl_phys(cs->as, addr, val);
1146         return H_SUCCESS;
1147     case 8:
1148         stq_phys(cs->as, addr, val);
1149         return H_SUCCESS;
1150     }
1151     return H_PARAMETER;
1152 }
1153 
1154 static target_ulong h_logical_memop(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1155                                     target_ulong opcode, target_ulong *args)
1156 {
1157     CPUState *cs = CPU(cpu);
1158 
1159     target_ulong dst   = args[0]; /* Destination address */
1160     target_ulong src   = args[1]; /* Source address */
1161     target_ulong esize = args[2]; /* Element size (0=1,1=2,2=4,3=8) */
1162     target_ulong count = args[3]; /* Element count */
1163     target_ulong op    = args[4]; /* 0 = copy, 1 = invert */
1164     uint64_t tmp;
1165     unsigned int mask = (1 << esize) - 1;
1166     int step = 1 << esize;
1167 
1168     if (count > 0x80000000) {
1169         return H_PARAMETER;
1170     }
1171 
1172     if ((dst & mask) || (src & mask) || (op > 1)) {
1173         return H_PARAMETER;
1174     }
1175 
1176     if (dst >= src && dst < (src + (count << esize))) {
1177             dst = dst + ((count - 1) << esize);
1178             src = src + ((count - 1) << esize);
1179             step = -step;
1180     }
1181 
1182     while (count--) {
1183         switch (esize) {
1184         case 0:
1185             tmp = ldub_phys(cs->as, src);
1186             break;
1187         case 1:
1188             tmp = lduw_phys(cs->as, src);
1189             break;
1190         case 2:
1191             tmp = ldl_phys(cs->as, src);
1192             break;
1193         case 3:
1194             tmp = ldq_phys(cs->as, src);
1195             break;
1196         default:
1197             return H_PARAMETER;
1198         }
1199         if (op == 1) {
1200             tmp = ~tmp;
1201         }
1202         switch (esize) {
1203         case 0:
1204             stb_phys(cs->as, dst, tmp);
1205             break;
1206         case 1:
1207             stw_phys(cs->as, dst, tmp);
1208             break;
1209         case 2:
1210             stl_phys(cs->as, dst, tmp);
1211             break;
1212         case 3:
1213             stq_phys(cs->as, dst, tmp);
1214             break;
1215         }
1216         dst = dst + step;
1217         src = src + step;
1218     }
1219 
1220     return H_SUCCESS;
1221 }
1222 
1223 static target_ulong h_logical_icbi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1224                                    target_ulong opcode, target_ulong *args)
1225 {
1226     /* Nothing to do on emulation, KVM will trap this in the kernel */
1227     return H_SUCCESS;
1228 }
1229 
1230 static target_ulong h_logical_dcbf(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1231                                    target_ulong opcode, target_ulong *args)
1232 {
1233     /* Nothing to do on emulation, KVM will trap this in the kernel */
1234     return H_SUCCESS;
1235 }
1236 
1237 static target_ulong h_set_mode_resource_le(PowerPCCPU *cpu,
1238                                            target_ulong mflags,
1239                                            target_ulong value1,
1240                                            target_ulong value2)
1241 {
1242     if (value1) {
1243         return H_P3;
1244     }
1245     if (value2) {
1246         return H_P4;
1247     }
1248 
1249     switch (mflags) {
1250     case H_SET_MODE_ENDIAN_BIG:
1251         set_all_lpcrs(0, LPCR_ILE);
1252         spapr_pci_switch_vga(true);
1253         return H_SUCCESS;
1254 
1255     case H_SET_MODE_ENDIAN_LITTLE:
1256         set_all_lpcrs(LPCR_ILE, LPCR_ILE);
1257         spapr_pci_switch_vga(false);
1258         return H_SUCCESS;
1259     }
1260 
1261     return H_UNSUPPORTED_FLAG;
1262 }
1263 
1264 static target_ulong h_set_mode_resource_addr_trans_mode(PowerPCCPU *cpu,
1265                                                         target_ulong mflags,
1266                                                         target_ulong value1,
1267                                                         target_ulong value2)
1268 {
1269     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
1270 
1271     if (!(pcc->insns_flags2 & PPC2_ISA207S)) {
1272         return H_P2;
1273     }
1274     if (value1) {
1275         return H_P3;
1276     }
1277     if (value2) {
1278         return H_P4;
1279     }
1280 
1281     if (mflags == AIL_RESERVED) {
1282         return H_UNSUPPORTED_FLAG;
1283     }
1284 
1285     set_all_lpcrs(mflags << LPCR_AIL_SHIFT, LPCR_AIL);
1286 
1287     return H_SUCCESS;
1288 }
1289 
1290 static target_ulong h_set_mode(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1291                                target_ulong opcode, target_ulong *args)
1292 {
1293     target_ulong resource = args[1];
1294     target_ulong ret = H_P2;
1295 
1296     switch (resource) {
1297     case H_SET_MODE_RESOURCE_LE:
1298         ret = h_set_mode_resource_le(cpu, args[0], args[2], args[3]);
1299         break;
1300     case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
1301         ret = h_set_mode_resource_addr_trans_mode(cpu, args[0],
1302                                                   args[2], args[3]);
1303         break;
1304     }
1305 
1306     return ret;
1307 }
1308 
1309 static target_ulong h_clean_slb(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1310                                 target_ulong opcode, target_ulong *args)
1311 {
1312     qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n",
1313                   opcode, " (H_CLEAN_SLB)");
1314     return H_FUNCTION;
1315 }
1316 
1317 static target_ulong h_invalidate_pid(PowerPCCPU *cpu, sPAPRMachineState *spapr,
1318                                      target_ulong opcode, target_ulong *args)
1319 {
1320     qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n",
1321                   opcode, " (H_INVALIDATE_PID)");
1322     return H_FUNCTION;
1323 }
1324 
1325 static void spapr_check_setup_free_hpt(sPAPRMachineState *spapr,
1326                                        uint64_t patbe_old, uint64_t patbe_new)
1327 {
1328     /*
1329      * We have 4 Options:
1330      * HASH->HASH || RADIX->RADIX || NOTHING->RADIX : Do Nothing
1331      * HASH->RADIX                                  : Free HPT
1332      * RADIX->HASH                                  : Allocate HPT
1333      * NOTHING->HASH                                : Allocate HPT
1334      * Note: NOTHING implies the case where we said the guest could choose
1335      *       later and so assumed radix and now it's called H_REG_PROC_TBL
1336      */
1337 
1338     if ((patbe_old & PATBE1_GR) == (patbe_new & PATBE1_GR)) {
1339         /* We assume RADIX, so this catches all the "Do Nothing" cases */
1340     } else if (!(patbe_old & PATBE1_GR)) {
1341         /* HASH->RADIX : Free HPT */
1342         spapr_free_hpt(spapr);
1343     } else if (!(patbe_new & PATBE1_GR)) {
1344         /* RADIX->HASH || NOTHING->HASH : Allocate HPT */
1345         spapr_setup_hpt_and_vrma(spapr);
1346     }
1347     return;
1348 }
1349 
1350 #define FLAGS_MASK              0x01FULL
1351 #define FLAG_MODIFY             0x10
1352 #define FLAG_REGISTER           0x08
1353 #define FLAG_RADIX              0x04
1354 #define FLAG_HASH_PROC_TBL      0x02
1355 #define FLAG_GTSE               0x01
1356 
1357 static target_ulong h_register_process_table(PowerPCCPU *cpu,
1358                                              sPAPRMachineState *spapr,
1359                                              target_ulong opcode,
1360                                              target_ulong *args)
1361 {
1362     target_ulong flags = args[0];
1363     target_ulong proc_tbl = args[1];
1364     target_ulong page_size = args[2];
1365     target_ulong table_size = args[3];
1366     uint64_t cproc;
1367 
1368     if (flags & ~FLAGS_MASK) { /* Check no reserved bits are set */
1369         return H_PARAMETER;
1370     }
1371     if (flags & FLAG_MODIFY) {
1372         if (flags & FLAG_REGISTER) {
1373             if (flags & FLAG_RADIX) { /* Register new RADIX process table */
1374                 if (proc_tbl & 0xfff || proc_tbl >> 60) {
1375                     return H_P2;
1376                 } else if (page_size) {
1377                     return H_P3;
1378                 } else if (table_size > 24) {
1379                     return H_P4;
1380                 }
1381                 cproc = PATBE1_GR | proc_tbl | table_size;
1382             } else { /* Register new HPT process table */
1383                 if (flags & FLAG_HASH_PROC_TBL) { /* Hash with Segment Tables */
1384                     /* TODO - Not Supported */
1385                     /* Technically caused by flag bits => H_PARAMETER */
1386                     return H_PARAMETER;
1387                 } else { /* Hash with SLB */
1388                     if (proc_tbl >> 38) {
1389                         return H_P2;
1390                     } else if (page_size & ~0x7) {
1391                         return H_P3;
1392                     } else if (table_size > 24) {
1393                         return H_P4;
1394                     }
1395                 }
1396                 cproc = (proc_tbl << 25) | page_size << 5 | table_size;
1397             }
1398 
1399         } else { /* Deregister current process table */
1400             /* Set to benign value: (current GR) | 0. This allows
1401              * deregistration in KVM to succeed even if the radix bit in flags
1402              * doesn't match the radix bit in the old PATB. */
1403             cproc = spapr->patb_entry & PATBE1_GR;
1404         }
1405     } else { /* Maintain current registration */
1406         if (!(flags & FLAG_RADIX) != !(spapr->patb_entry & PATBE1_GR)) {
1407             /* Technically caused by flag bits => H_PARAMETER */
1408             return H_PARAMETER; /* Existing Process Table Mismatch */
1409         }
1410         cproc = spapr->patb_entry;
1411     }
1412 
1413     /* Check if we need to setup OR free the hpt */
1414     spapr_check_setup_free_hpt(spapr, spapr->patb_entry, cproc);
1415 
1416     spapr->patb_entry = cproc; /* Save new process table */
1417 
1418     /* Update the UPRT and GTSE bits in the LPCR for all cpus */
1419     set_all_lpcrs(((flags & (FLAG_RADIX | FLAG_HASH_PROC_TBL)) ? LPCR_UPRT : 0) |
1420                   ((flags & FLAG_GTSE) ? LPCR_GTSE : 0),
1421                   LPCR_UPRT | LPCR_GTSE);
1422 
1423     if (kvm_enabled()) {
1424         return kvmppc_configure_v3_mmu(cpu, flags & FLAG_RADIX,
1425                                        flags & FLAG_GTSE, cproc);
1426     }
1427     return H_SUCCESS;
1428 }
1429 
1430 #define H_SIGNAL_SYS_RESET_ALL         -1
1431 #define H_SIGNAL_SYS_RESET_ALLBUTSELF  -2
1432 
1433 static target_ulong h_signal_sys_reset(PowerPCCPU *cpu,
1434                                        sPAPRMachineState *spapr,
1435                                        target_ulong opcode, target_ulong *args)
1436 {
1437     target_long target = args[0];
1438     CPUState *cs;
1439 
1440     if (target < 0) {
1441         /* Broadcast */
1442         if (target < H_SIGNAL_SYS_RESET_ALLBUTSELF) {
1443             return H_PARAMETER;
1444         }
1445 
1446         CPU_FOREACH(cs) {
1447             PowerPCCPU *c = POWERPC_CPU(cs);
1448 
1449             if (target == H_SIGNAL_SYS_RESET_ALLBUTSELF) {
1450                 if (c == cpu) {
1451                     continue;
1452                 }
1453             }
1454             run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
1455         }
1456         return H_SUCCESS;
1457 
1458     } else {
1459         /* Unicast */
1460         cs = CPU(spapr_find_cpu(target));
1461         if (cs) {
1462             run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
1463             return H_SUCCESS;
1464         }
1465         return H_PARAMETER;
1466     }
1467 }
1468 
1469 static uint32_t cas_check_pvr(sPAPRMachineState *spapr, PowerPCCPU *cpu,
1470                               target_ulong *addr, bool *raw_mode_supported,
1471                               Error **errp)
1472 {
1473     bool explicit_match = false; /* Matched the CPU's real PVR */
1474     uint32_t max_compat = spapr->max_compat_pvr;
1475     uint32_t best_compat = 0;
1476     int i;
1477 
1478     /*
1479      * We scan the supplied table of PVRs looking for two things
1480      *   1. Is our real CPU PVR in the list?
1481      *   2. What's the "best" listed logical PVR
1482      */
1483     for (i = 0; i < 512; ++i) {
1484         uint32_t pvr, pvr_mask;
1485 
1486         pvr_mask = ldl_be_phys(&address_space_memory, *addr);
1487         pvr = ldl_be_phys(&address_space_memory, *addr + 4);
1488         *addr += 8;
1489 
1490         if (~pvr_mask & pvr) {
1491             break; /* Terminator record */
1492         }
1493 
1494         if ((cpu->env.spr[SPR_PVR] & pvr_mask) == (pvr & pvr_mask)) {
1495             explicit_match = true;
1496         } else {
1497             if (ppc_check_compat(cpu, pvr, best_compat, max_compat)) {
1498                 best_compat = pvr;
1499             }
1500         }
1501     }
1502 
1503     if ((best_compat == 0) && (!explicit_match || max_compat)) {
1504         /* We couldn't find a suitable compatibility mode, and either
1505          * the guest doesn't support "raw" mode for this CPU, or raw
1506          * mode is disabled because a maximum compat mode is set */
1507         error_setg(errp, "Couldn't negotiate a suitable PVR during CAS");
1508         return 0;
1509     }
1510 
1511     *raw_mode_supported = explicit_match;
1512 
1513     /* Parsing finished */
1514     trace_spapr_cas_pvr(cpu->compat_pvr, explicit_match, best_compat);
1515 
1516     return best_compat;
1517 }
1518 
1519 static target_ulong h_client_architecture_support(PowerPCCPU *cpu,
1520                                                   sPAPRMachineState *spapr,
1521                                                   target_ulong opcode,
1522                                                   target_ulong *args)
1523 {
1524     /* Working address in data buffer */
1525     target_ulong addr = ppc64_phys_to_real(args[0]);
1526     target_ulong ov_table;
1527     uint32_t cas_pvr;
1528     sPAPROptionVector *ov1_guest, *ov5_guest, *ov5_cas_old, *ov5_updates;
1529     bool guest_radix;
1530     Error *local_err = NULL;
1531     bool raw_mode_supported = false;
1532 
1533     cas_pvr = cas_check_pvr(spapr, cpu, &addr, &raw_mode_supported, &local_err);
1534     if (local_err) {
1535         error_report_err(local_err);
1536         return H_HARDWARE;
1537     }
1538 
1539     /* Update CPUs */
1540     if (cpu->compat_pvr != cas_pvr) {
1541         ppc_set_compat_all(cas_pvr, &local_err);
1542         if (local_err) {
1543             /* We fail to set compat mode (likely because running with KVM PR),
1544              * but maybe we can fallback to raw mode if the guest supports it.
1545              */
1546             if (!raw_mode_supported) {
1547                 error_report_err(local_err);
1548                 return H_HARDWARE;
1549             }
1550             local_err = NULL;
1551         }
1552     }
1553 
1554     /* For the future use: here @ov_table points to the first option vector */
1555     ov_table = addr;
1556 
1557     ov1_guest = spapr_ovec_parse_vector(ov_table, 1);
1558     ov5_guest = spapr_ovec_parse_vector(ov_table, 5);
1559     if (spapr_ovec_test(ov5_guest, OV5_MMU_BOTH)) {
1560         error_report("guest requested hash and radix MMU, which is invalid.");
1561         exit(EXIT_FAILURE);
1562     }
1563     /* The radix/hash bit in byte 24 requires special handling: */
1564     guest_radix = spapr_ovec_test(ov5_guest, OV5_MMU_RADIX_300);
1565     spapr_ovec_clear(ov5_guest, OV5_MMU_RADIX_300);
1566 
1567     /*
1568      * HPT resizing is a bit of a special case, because when enabled
1569      * we assume an HPT guest will support it until it says it
1570      * doesn't, instead of assuming it won't support it until it says
1571      * it does.  Strictly speaking that approach could break for
1572      * guests which don't make a CAS call, but those are so old we
1573      * don't care about them.  Without that assumption we'd have to
1574      * make at least a temporary allocation of an HPT sized for max
1575      * memory, which could be impossibly difficult under KVM HV if
1576      * maxram is large.
1577      */
1578     if (!guest_radix && !spapr_ovec_test(ov5_guest, OV5_HPT_RESIZE)) {
1579         int maxshift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1580 
1581         if (spapr->resize_hpt == SPAPR_RESIZE_HPT_REQUIRED) {
1582             error_report(
1583                 "h_client_architecture_support: Guest doesn't support HPT resizing, but resize-hpt=required");
1584             exit(1);
1585         }
1586 
1587         if (spapr->htab_shift < maxshift) {
1588             /* Guest doesn't know about HPT resizing, so we
1589              * pre-emptively resize for the maximum permitted RAM.  At
1590              * the point this is called, nothing should have been
1591              * entered into the existing HPT */
1592             spapr_reallocate_hpt(spapr, maxshift, &error_fatal);
1593             push_sregs_to_kvm_pr(spapr);
1594         }
1595     }
1596 
1597     /* NOTE: there are actually a number of ov5 bits where input from the
1598      * guest is always zero, and the platform/QEMU enables them independently
1599      * of guest input. To model these properly we'd want some sort of mask,
1600      * but since they only currently apply to memory migration as defined
1601      * by LoPAPR 1.1, 14.5.4.8, which QEMU doesn't implement, we don't need
1602      * to worry about this for now.
1603      */
1604     ov5_cas_old = spapr_ovec_clone(spapr->ov5_cas);
1605 
1606     /* also clear the radix/hash bit from the current ov5_cas bits to
1607      * be in sync with the newly ov5 bits. Else the radix bit will be
1608      * seen as being removed and this will generate a reset loop
1609      */
1610     spapr_ovec_clear(ov5_cas_old, OV5_MMU_RADIX_300);
1611 
1612     /* full range of negotiated ov5 capabilities */
1613     spapr_ovec_intersect(spapr->ov5_cas, spapr->ov5, ov5_guest);
1614     spapr_ovec_cleanup(ov5_guest);
1615     /* capabilities that have been added since CAS-generated guest reset.
1616      * if capabilities have since been removed, generate another reset
1617      */
1618     ov5_updates = spapr_ovec_new();
1619     spapr->cas_reboot = spapr_ovec_diff(ov5_updates,
1620                                         ov5_cas_old, spapr->ov5_cas);
1621     /* Now that processing is finished, set the radix/hash bit for the
1622      * guest if it requested a valid mode; otherwise terminate the boot. */
1623     if (guest_radix) {
1624         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1625             error_report("Guest requested unavailable MMU mode (radix).");
1626             exit(EXIT_FAILURE);
1627         }
1628         spapr_ovec_set(spapr->ov5_cas, OV5_MMU_RADIX_300);
1629     } else {
1630         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1631             && !kvmppc_has_cap_mmu_hash_v3()) {
1632             error_report("Guest requested unavailable MMU mode (hash).");
1633             exit(EXIT_FAILURE);
1634         }
1635     }
1636     spapr->cas_legacy_guest_workaround = !spapr_ovec_test(ov1_guest,
1637                                                           OV1_PPC_3_00);
1638     if (!spapr->cas_reboot) {
1639         /* If spapr_machine_reset() did not set up a HPT but one is necessary
1640          * (because the guest isn't going to use radix) then set it up here. */
1641         if ((spapr->patb_entry & PATBE1_GR) && !guest_radix) {
1642             /* legacy hash or new hash: */
1643             spapr_setup_hpt_and_vrma(spapr);
1644         }
1645         spapr->cas_reboot =
1646             (spapr_h_cas_compose_response(spapr, args[1], args[2],
1647                                           ov5_updates) != 0);
1648     }
1649     spapr_ovec_cleanup(ov5_updates);
1650 
1651     if (spapr->cas_reboot) {
1652         qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET);
1653     }
1654 
1655     return H_SUCCESS;
1656 }
1657 
1658 static target_ulong h_get_cpu_characteristics(PowerPCCPU *cpu,
1659                                               sPAPRMachineState *spapr,
1660                                               target_ulong opcode,
1661                                               target_ulong *args)
1662 {
1663     uint64_t characteristics = H_CPU_CHAR_HON_BRANCH_HINTS &
1664                                ~H_CPU_CHAR_THR_RECONF_TRIG;
1665     uint64_t behaviour = H_CPU_BEHAV_FAVOUR_SECURITY;
1666     uint8_t safe_cache = spapr_get_cap(spapr, SPAPR_CAP_CFPC);
1667     uint8_t safe_bounds_check = spapr_get_cap(spapr, SPAPR_CAP_SBBC);
1668     uint8_t safe_indirect_branch = spapr_get_cap(spapr, SPAPR_CAP_IBS);
1669 
1670     switch (safe_cache) {
1671     case SPAPR_CAP_WORKAROUND:
1672         characteristics |= H_CPU_CHAR_L1D_FLUSH_ORI30;
1673         characteristics |= H_CPU_CHAR_L1D_FLUSH_TRIG2;
1674         characteristics |= H_CPU_CHAR_L1D_THREAD_PRIV;
1675         behaviour |= H_CPU_BEHAV_L1D_FLUSH_PR;
1676         break;
1677     case SPAPR_CAP_FIXED:
1678         break;
1679     default: /* broken */
1680         assert(safe_cache == SPAPR_CAP_BROKEN);
1681         behaviour |= H_CPU_BEHAV_L1D_FLUSH_PR;
1682         break;
1683     }
1684 
1685     switch (safe_bounds_check) {
1686     case SPAPR_CAP_WORKAROUND:
1687         characteristics |= H_CPU_CHAR_SPEC_BAR_ORI31;
1688         behaviour |= H_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1689         break;
1690     case SPAPR_CAP_FIXED:
1691         break;
1692     default: /* broken */
1693         assert(safe_bounds_check == SPAPR_CAP_BROKEN);
1694         behaviour |= H_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1695         break;
1696     }
1697 
1698     switch (safe_indirect_branch) {
1699     case SPAPR_CAP_FIXED_CCD:
1700         characteristics |= H_CPU_CHAR_CACHE_COUNT_DIS;
1701         break;
1702     case SPAPR_CAP_FIXED_IBS:
1703         characteristics |= H_CPU_CHAR_BCCTRL_SERIALISED;
1704         break;
1705     default: /* broken */
1706         assert(safe_indirect_branch == SPAPR_CAP_BROKEN);
1707         break;
1708     }
1709 
1710     args[0] = characteristics;
1711     args[1] = behaviour;
1712 
1713     return H_SUCCESS;
1714 }
1715 
1716 static spapr_hcall_fn papr_hypercall_table[(MAX_HCALL_OPCODE / 4) + 1];
1717 static spapr_hcall_fn kvmppc_hypercall_table[KVMPPC_HCALL_MAX - KVMPPC_HCALL_BASE + 1];
1718 
1719 void spapr_register_hypercall(target_ulong opcode, spapr_hcall_fn fn)
1720 {
1721     spapr_hcall_fn *slot;
1722 
1723     if (opcode <= MAX_HCALL_OPCODE) {
1724         assert((opcode & 0x3) == 0);
1725 
1726         slot = &papr_hypercall_table[opcode / 4];
1727     } else {
1728         assert((opcode >= KVMPPC_HCALL_BASE) && (opcode <= KVMPPC_HCALL_MAX));
1729 
1730         slot = &kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE];
1731     }
1732 
1733     assert(!(*slot));
1734     *slot = fn;
1735 }
1736 
1737 target_ulong spapr_hypercall(PowerPCCPU *cpu, target_ulong opcode,
1738                              target_ulong *args)
1739 {
1740     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1741 
1742     if ((opcode <= MAX_HCALL_OPCODE)
1743         && ((opcode & 0x3) == 0)) {
1744         spapr_hcall_fn fn = papr_hypercall_table[opcode / 4];
1745 
1746         if (fn) {
1747             return fn(cpu, spapr, opcode, args);
1748         }
1749     } else if ((opcode >= KVMPPC_HCALL_BASE) &&
1750                (opcode <= KVMPPC_HCALL_MAX)) {
1751         spapr_hcall_fn fn = kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE];
1752 
1753         if (fn) {
1754             return fn(cpu, spapr, opcode, args);
1755         }
1756     }
1757 
1758     qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x" TARGET_FMT_lx "\n",
1759                   opcode);
1760     return H_FUNCTION;
1761 }
1762 
1763 static void hypercall_register_types(void)
1764 {
1765     /* hcall-pft */
1766     spapr_register_hypercall(H_ENTER, h_enter);
1767     spapr_register_hypercall(H_REMOVE, h_remove);
1768     spapr_register_hypercall(H_PROTECT, h_protect);
1769     spapr_register_hypercall(H_READ, h_read);
1770 
1771     /* hcall-bulk */
1772     spapr_register_hypercall(H_BULK_REMOVE, h_bulk_remove);
1773 
1774     /* hcall-hpt-resize */
1775     spapr_register_hypercall(H_RESIZE_HPT_PREPARE, h_resize_hpt_prepare);
1776     spapr_register_hypercall(H_RESIZE_HPT_COMMIT, h_resize_hpt_commit);
1777 
1778     /* hcall-splpar */
1779     spapr_register_hypercall(H_REGISTER_VPA, h_register_vpa);
1780     spapr_register_hypercall(H_CEDE, h_cede);
1781     spapr_register_hypercall(H_SIGNAL_SYS_RESET, h_signal_sys_reset);
1782 
1783     /* processor register resource access h-calls */
1784     spapr_register_hypercall(H_SET_SPRG0, h_set_sprg0);
1785     spapr_register_hypercall(H_SET_DABR, h_set_dabr);
1786     spapr_register_hypercall(H_SET_XDABR, h_set_xdabr);
1787     spapr_register_hypercall(H_PAGE_INIT, h_page_init);
1788     spapr_register_hypercall(H_SET_MODE, h_set_mode);
1789 
1790     /* In Memory Table MMU h-calls */
1791     spapr_register_hypercall(H_CLEAN_SLB, h_clean_slb);
1792     spapr_register_hypercall(H_INVALIDATE_PID, h_invalidate_pid);
1793     spapr_register_hypercall(H_REGISTER_PROC_TBL, h_register_process_table);
1794 
1795     /* hcall-get-cpu-characteristics */
1796     spapr_register_hypercall(H_GET_CPU_CHARACTERISTICS,
1797                              h_get_cpu_characteristics);
1798 
1799     /* "debugger" hcalls (also used by SLOF). Note: We do -not- differenciate
1800      * here between the "CI" and the "CACHE" variants, they will use whatever
1801      * mapping attributes qemu is using. When using KVM, the kernel will
1802      * enforce the attributes more strongly
1803      */
1804     spapr_register_hypercall(H_LOGICAL_CI_LOAD, h_logical_load);
1805     spapr_register_hypercall(H_LOGICAL_CI_STORE, h_logical_store);
1806     spapr_register_hypercall(H_LOGICAL_CACHE_LOAD, h_logical_load);
1807     spapr_register_hypercall(H_LOGICAL_CACHE_STORE, h_logical_store);
1808     spapr_register_hypercall(H_LOGICAL_ICBI, h_logical_icbi);
1809     spapr_register_hypercall(H_LOGICAL_DCBF, h_logical_dcbf);
1810     spapr_register_hypercall(KVMPPC_H_LOGICAL_MEMOP, h_logical_memop);
1811 
1812     /* qemu/KVM-PPC specific hcalls */
1813     spapr_register_hypercall(KVMPPC_H_RTAS, h_rtas);
1814 
1815     /* ibm,client-architecture-support support */
1816     spapr_register_hypercall(KVMPPC_H_CAS, h_client_architecture_support);
1817 }
1818 
1819 type_init(hypercall_register_types)
1820