xref: /openbmc/qemu/hw/ppc/spapr_caps.c (revision f101c9fe)
1 /*
2  * QEMU PowerPC pSeries Logical Partition capabilities handling
3  *
4  * Copyright (c) 2017 David Gibson, Red Hat Inc.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu/error-report.h"
27 #include "qapi/error.h"
28 #include "qapi/visitor.h"
29 #include "sysemu/hw_accel.h"
30 #include "exec/ram_addr.h"
31 #include "target/ppc/cpu.h"
32 #include "target/ppc/mmu-hash64.h"
33 #include "cpu-models.h"
34 #include "kvm_ppc.h"
35 #include "migration/vmstate.h"
36 #include "sysemu/tcg.h"
37 
38 #include "hw/ppc/spapr.h"
39 
40 typedef struct SpaprCapPossible {
41     int num;            /* size of vals array below */
42     const char *help;   /* help text for vals */
43     /*
44      * Note:
45      * - because of the way compatibility is determined vals MUST be ordered
46      *   such that later options are a superset of all preceding options.
47      * - the order of vals must be preserved, that is their index is important,
48      *   however vals may be added to the end of the list so long as the above
49      *   point is observed
50      */
51     const char *vals[];
52 } SpaprCapPossible;
53 
54 typedef struct SpaprCapabilityInfo {
55     const char *name;
56     const char *description;
57     int index;
58 
59     /* Getter and Setter Function Pointers */
60     ObjectPropertyAccessor *get;
61     ObjectPropertyAccessor *set;
62     const char *type;
63     /* Possible values if this is a custom string type */
64     SpaprCapPossible *possible;
65     /* Make sure the virtual hardware can support this capability */
66     void (*apply)(SpaprMachineState *spapr, uint8_t val, Error **errp);
67     void (*cpu_apply)(SpaprMachineState *spapr, PowerPCCPU *cpu,
68                       uint8_t val, Error **errp);
69     bool (*migrate_needed)(void *opaque);
70 } SpaprCapabilityInfo;
71 
72 static void spapr_cap_get_bool(Object *obj, Visitor *v, const char *name,
73                                void *opaque, Error **errp)
74 {
75     SpaprCapabilityInfo *cap = opaque;
76     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
77     bool value = spapr_get_cap(spapr, cap->index) == SPAPR_CAP_ON;
78 
79     visit_type_bool(v, name, &value, errp);
80 }
81 
82 static void spapr_cap_set_bool(Object *obj, Visitor *v, const char *name,
83                                void *opaque, Error **errp)
84 {
85     SpaprCapabilityInfo *cap = opaque;
86     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
87     bool value;
88 
89     if (!visit_type_bool(v, name, &value, errp)) {
90         return;
91     }
92 
93     spapr->cmd_line_caps[cap->index] = true;
94     spapr->eff.caps[cap->index] = value ? SPAPR_CAP_ON : SPAPR_CAP_OFF;
95 }
96 
97 
98 static void  spapr_cap_get_string(Object *obj, Visitor *v, const char *name,
99                                   void *opaque, Error **errp)
100 {
101     SpaprCapabilityInfo *cap = opaque;
102     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
103     char *val = NULL;
104     uint8_t value = spapr_get_cap(spapr, cap->index);
105 
106     if (value >= cap->possible->num) {
107         error_setg(errp, "Invalid value (%d) for cap-%s", value, cap->name);
108         return;
109     }
110 
111     val = g_strdup(cap->possible->vals[value]);
112 
113     visit_type_str(v, name, &val, errp);
114     g_free(val);
115 }
116 
117 static void spapr_cap_set_string(Object *obj, Visitor *v, const char *name,
118                                  void *opaque, Error **errp)
119 {
120     SpaprCapabilityInfo *cap = opaque;
121     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
122     uint8_t i;
123     char *val;
124 
125     if (!visit_type_str(v, name, &val, errp)) {
126         return;
127     }
128 
129     if (!strcmp(val, "?")) {
130         error_setg(errp, "%s", cap->possible->help);
131         goto out;
132     }
133     for (i = 0; i < cap->possible->num; i++) {
134         if (!strcasecmp(val, cap->possible->vals[i])) {
135             spapr->cmd_line_caps[cap->index] = true;
136             spapr->eff.caps[cap->index] = i;
137             goto out;
138         }
139     }
140 
141     error_setg(errp, "Invalid capability mode \"%s\" for cap-%s", val,
142                cap->name);
143 out:
144     g_free(val);
145 }
146 
147 static void spapr_cap_get_pagesize(Object *obj, Visitor *v, const char *name,
148                                    void *opaque, Error **errp)
149 {
150     SpaprCapabilityInfo *cap = opaque;
151     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
152     uint8_t val = spapr_get_cap(spapr, cap->index);
153     uint64_t pagesize = (1ULL << val);
154 
155     visit_type_size(v, name, &pagesize, errp);
156 }
157 
158 static void spapr_cap_set_pagesize(Object *obj, Visitor *v, const char *name,
159                                    void *opaque, Error **errp)
160 {
161     SpaprCapabilityInfo *cap = opaque;
162     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
163     uint64_t pagesize;
164     uint8_t val;
165 
166     if (!visit_type_size(v, name, &pagesize, errp)) {
167         return;
168     }
169 
170     if (!is_power_of_2(pagesize)) {
171         error_setg(errp, "cap-%s must be a power of 2", cap->name);
172         return;
173     }
174 
175     val = ctz64(pagesize);
176     spapr->cmd_line_caps[cap->index] = true;
177     spapr->eff.caps[cap->index] = val;
178 }
179 
180 static void cap_htm_apply(SpaprMachineState *spapr, uint8_t val, Error **errp)
181 {
182     ERRP_GUARD();
183     if (!val) {
184         /* TODO: We don't support disabling htm yet */
185         return;
186     }
187     if (tcg_enabled()) {
188         error_setg(errp, "No Transactional Memory support in TCG");
189         error_append_hint(errp, "Try appending -machine cap-htm=off\n");
190     } else if (kvm_enabled() && !kvmppc_has_cap_htm()) {
191         error_setg(errp,
192                    "KVM implementation does not support Transactional Memory");
193         error_append_hint(errp, "Try appending -machine cap-htm=off\n");
194     }
195 }
196 
197 static void cap_vsx_apply(SpaprMachineState *spapr, uint8_t val, Error **errp)
198 {
199     ERRP_GUARD();
200     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
201     CPUPPCState *env = &cpu->env;
202 
203     if (!val) {
204         /* TODO: We don't support disabling vsx yet */
205         return;
206     }
207     /* Allowable CPUs in spapr_cpu_core.c should already have gotten
208      * rid of anything that doesn't do VMX */
209     g_assert(env->insns_flags & PPC_ALTIVEC);
210     if (!(env->insns_flags2 & PPC2_VSX)) {
211         error_setg(errp, "VSX support not available");
212         error_append_hint(errp, "Try appending -machine cap-vsx=off\n");
213     }
214 }
215 
216 static void cap_dfp_apply(SpaprMachineState *spapr, uint8_t val, Error **errp)
217 {
218     ERRP_GUARD();
219     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
220     CPUPPCState *env = &cpu->env;
221 
222     if (!val) {
223         /* TODO: We don't support disabling dfp yet */
224         return;
225     }
226     if (!(env->insns_flags2 & PPC2_DFP)) {
227         error_setg(errp, "DFP support not available");
228         error_append_hint(errp, "Try appending -machine cap-dfp=off\n");
229     }
230 }
231 
232 SpaprCapPossible cap_cfpc_possible = {
233     .num = 3,
234     .vals = {"broken", "workaround", "fixed"},
235     .help = "broken - no protection, workaround - workaround available,"
236             " fixed - fixed in hardware",
237 };
238 
239 static void cap_safe_cache_apply(SpaprMachineState *spapr, uint8_t val,
240                                  Error **errp)
241 {
242     ERRP_GUARD();
243     uint8_t kvm_val =  kvmppc_get_cap_safe_cache();
244 
245     if (tcg_enabled() && val) {
246         /* TCG only supports broken, allow other values and print a warning */
247         warn_report("TCG doesn't support requested feature, cap-cfpc=%s",
248                     cap_cfpc_possible.vals[val]);
249     } else if (kvm_enabled() && (val > kvm_val)) {
250         error_setg(errp,
251                    "Requested safe cache capability level not supported by KVM");
252         error_append_hint(errp, "Try appending -machine cap-cfpc=%s\n",
253                           cap_cfpc_possible.vals[kvm_val]);
254     }
255 }
256 
257 SpaprCapPossible cap_sbbc_possible = {
258     .num = 3,
259     .vals = {"broken", "workaround", "fixed"},
260     .help = "broken - no protection, workaround - workaround available,"
261             " fixed - fixed in hardware",
262 };
263 
264 static void cap_safe_bounds_check_apply(SpaprMachineState *spapr, uint8_t val,
265                                         Error **errp)
266 {
267     ERRP_GUARD();
268     uint8_t kvm_val =  kvmppc_get_cap_safe_bounds_check();
269 
270     if (tcg_enabled() && val) {
271         /* TCG only supports broken, allow other values and print a warning */
272         warn_report("TCG doesn't support requested feature, cap-sbbc=%s",
273                     cap_sbbc_possible.vals[val]);
274     } else if (kvm_enabled() && (val > kvm_val)) {
275         error_setg(errp,
276 "Requested safe bounds check capability level not supported by KVM");
277         error_append_hint(errp, "Try appending -machine cap-sbbc=%s\n",
278                           cap_sbbc_possible.vals[kvm_val]);
279     }
280 }
281 
282 SpaprCapPossible cap_ibs_possible = {
283     .num = 5,
284     /* Note workaround only maintained for compatibility */
285     .vals = {"broken", "workaround", "fixed-ibs", "fixed-ccd", "fixed-na"},
286     .help = "broken - no protection, workaround - count cache flush"
287             ", fixed-ibs - indirect branch serialisation,"
288             " fixed-ccd - cache count disabled,"
289             " fixed-na - fixed in hardware (no longer applicable)",
290 };
291 
292 static void cap_safe_indirect_branch_apply(SpaprMachineState *spapr,
293                                            uint8_t val, Error **errp)
294 {
295     ERRP_GUARD();
296     uint8_t kvm_val = kvmppc_get_cap_safe_indirect_branch();
297 
298     if (tcg_enabled() && val) {
299         /* TCG only supports broken, allow other values and print a warning */
300         warn_report("TCG doesn't support requested feature, cap-ibs=%s",
301                     cap_ibs_possible.vals[val]);
302     } else if (kvm_enabled() && (val > kvm_val)) {
303         error_setg(errp,
304 "Requested safe indirect branch capability level not supported by KVM");
305         error_append_hint(errp, "Try appending -machine cap-ibs=%s\n",
306                           cap_ibs_possible.vals[kvm_val]);
307     }
308 }
309 
310 #define VALUE_DESC_TRISTATE     " (broken, workaround, fixed)"
311 
312 bool spapr_check_pagesize(SpaprMachineState *spapr, hwaddr pagesize,
313                           Error **errp)
314 {
315     hwaddr maxpagesize = (1ULL << spapr->eff.caps[SPAPR_CAP_HPT_MAXPAGESIZE]);
316 
317     if (!kvmppc_hpt_needs_host_contiguous_pages()) {
318         return true;
319     }
320 
321     if (maxpagesize > pagesize) {
322         error_setg(errp,
323                    "Can't support %"HWADDR_PRIu" kiB guest pages with %"
324                    HWADDR_PRIu" kiB host pages with this KVM implementation",
325                    maxpagesize >> 10, pagesize >> 10);
326         return false;
327     }
328 
329     return true;
330 }
331 
332 static void cap_hpt_maxpagesize_apply(SpaprMachineState *spapr,
333                                       uint8_t val, Error **errp)
334 {
335     if (val < 12) {
336         error_setg(errp, "Require at least 4kiB hpt-max-page-size");
337         return;
338     } else if (val < 16) {
339         warn_report("Many guests require at least 64kiB hpt-max-page-size");
340     }
341 
342     spapr_check_pagesize(spapr, qemu_minrampagesize(), errp);
343 }
344 
345 static bool cap_hpt_maxpagesize_migrate_needed(void *opaque)
346 {
347     return !SPAPR_MACHINE_GET_CLASS(opaque)->pre_4_1_migration;
348 }
349 
350 static bool spapr_pagesize_cb(void *opaque, uint32_t seg_pshift,
351                               uint32_t pshift)
352 {
353     unsigned maxshift = *((unsigned *)opaque);
354 
355     assert(pshift >= seg_pshift);
356 
357     /* Don't allow the guest to use pages bigger than the configured
358      * maximum size */
359     if (pshift > maxshift) {
360         return false;
361     }
362 
363     /* For whatever reason, KVM doesn't allow multiple pagesizes
364      * within a segment, *except* for the case of 16M pages in a 4k or
365      * 64k segment.  Always exclude other cases, so that TCG and KVM
366      * guests see a consistent environment */
367     if ((pshift != seg_pshift) && (pshift != 24)) {
368         return false;
369     }
370 
371     return true;
372 }
373 
374 static void ppc_hash64_filter_pagesizes(PowerPCCPU *cpu,
375                                  bool (*cb)(void *, uint32_t, uint32_t),
376                                  void *opaque)
377 {
378     PPCHash64Options *opts = cpu->hash64_opts;
379     int i;
380     int n = 0;
381     bool ci_largepage = false;
382 
383     assert(opts);
384 
385     n = 0;
386     for (i = 0; i < ARRAY_SIZE(opts->sps); i++) {
387         PPCHash64SegmentPageSizes *sps = &opts->sps[i];
388         int j;
389         int m = 0;
390 
391         assert(n <= i);
392 
393         if (!sps->page_shift) {
394             break;
395         }
396 
397         for (j = 0; j < ARRAY_SIZE(sps->enc); j++) {
398             PPCHash64PageSize *ps = &sps->enc[j];
399 
400             assert(m <= j);
401             if (!ps->page_shift) {
402                 break;
403             }
404 
405             if (cb(opaque, sps->page_shift, ps->page_shift)) {
406                 if (ps->page_shift >= 16) {
407                     ci_largepage = true;
408                 }
409                 sps->enc[m++] = *ps;
410             }
411         }
412 
413         /* Clear rest of the row */
414         for (j = m; j < ARRAY_SIZE(sps->enc); j++) {
415             memset(&sps->enc[j], 0, sizeof(sps->enc[j]));
416         }
417 
418         if (m) {
419             n++;
420         }
421     }
422 
423     /* Clear the rest of the table */
424     for (i = n; i < ARRAY_SIZE(opts->sps); i++) {
425         memset(&opts->sps[i], 0, sizeof(opts->sps[i]));
426     }
427 
428     if (!ci_largepage) {
429         opts->flags &= ~PPC_HASH64_CI_LARGEPAGE;
430     }
431 }
432 
433 static void cap_hpt_maxpagesize_cpu_apply(SpaprMachineState *spapr,
434                                           PowerPCCPU *cpu,
435                                           uint8_t val, Error **errp)
436 {
437     unsigned maxshift = val;
438 
439     ppc_hash64_filter_pagesizes(cpu, spapr_pagesize_cb, &maxshift);
440 }
441 
442 static void cap_nested_kvm_hv_apply(SpaprMachineState *spapr,
443                                     uint8_t val, Error **errp)
444 {
445     ERRP_GUARD();
446     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
447 
448     if (!val) {
449         /* capability disabled by default */
450         return;
451     }
452 
453     if (tcg_enabled()) {
454         error_setg(errp, "No Nested KVM-HV support in TCG");
455         error_append_hint(errp, "Try appending -machine cap-nested-hv=off\n");
456     } else if (kvm_enabled()) {
457         if (!ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0,
458                               spapr->max_compat_pvr)) {
459             error_setg(errp, "Nested KVM-HV only supported on POWER9");
460             error_append_hint(errp,
461                               "Try appending -machine max-cpu-compat=power9\n");
462             return;
463         }
464 
465         if (!kvmppc_has_cap_nested_kvm_hv()) {
466             error_setg(errp,
467                        "KVM implementation does not support Nested KVM-HV");
468             error_append_hint(errp,
469                               "Try appending -machine cap-nested-hv=off\n");
470         } else if (kvmppc_set_cap_nested_kvm_hv(val) < 0) {
471                 error_setg(errp, "Error enabling cap-nested-hv with KVM");
472                 error_append_hint(errp,
473                                   "Try appending -machine cap-nested-hv=off\n");
474         }
475     }
476 }
477 
478 static void cap_large_decr_apply(SpaprMachineState *spapr,
479                                  uint8_t val, Error **errp)
480 {
481     ERRP_GUARD();
482     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
483     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
484 
485     if (!val) {
486         return; /* Disabled by default */
487     }
488 
489     if (tcg_enabled()) {
490         if (!ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0,
491                               spapr->max_compat_pvr)) {
492             error_setg(errp, "Large decrementer only supported on POWER9");
493             error_append_hint(errp, "Try -cpu POWER9\n");
494             return;
495         }
496     } else if (kvm_enabled()) {
497         int kvm_nr_bits = kvmppc_get_cap_large_decr();
498 
499         if (!kvm_nr_bits) {
500             error_setg(errp, "No large decrementer support");
501             error_append_hint(errp,
502                               "Try appending -machine cap-large-decr=off\n");
503         } else if (pcc->lrg_decr_bits != kvm_nr_bits) {
504             error_setg(errp,
505                        "KVM large decrementer size (%d) differs to model (%d)",
506                        kvm_nr_bits, pcc->lrg_decr_bits);
507             error_append_hint(errp,
508                               "Try appending -machine cap-large-decr=off\n");
509         }
510     }
511 }
512 
513 static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
514                                      PowerPCCPU *cpu,
515                                      uint8_t val, Error **errp)
516 {
517     ERRP_GUARD();
518     CPUPPCState *env = &cpu->env;
519     target_ulong lpcr = env->spr[SPR_LPCR];
520 
521     if (kvm_enabled()) {
522         if (kvmppc_enable_cap_large_decr(cpu, val)) {
523             error_setg(errp, "No large decrementer support");
524             error_append_hint(errp,
525                               "Try appending -machine cap-large-decr=off\n");
526         }
527     }
528 
529     if (val) {
530         lpcr |= LPCR_LD;
531     } else {
532         lpcr &= ~LPCR_LD;
533     }
534     ppc_store_lpcr(cpu, lpcr);
535 }
536 
537 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
538                                  Error **errp)
539 {
540     ERRP_GUARD();
541     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
542 
543     if (tcg_enabled() && val) {
544         /* TCG doesn't implement anything here, but allow with a warning */
545         warn_report("TCG doesn't support requested feature, cap-ccf-assist=on");
546     } else if (kvm_enabled() && (val > kvm_val)) {
547         uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
548 
549         if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
550             /*
551              * If we don't have CCF assist on the host, the assist
552              * instruction is a harmless no-op.  It won't correctly
553              * implement the cache count flush *but* if we have
554              * count-cache-disabled in the host, that flush is
555              * unnnecessary.  So, specifically allow this case.  This
556              * allows us to have better performance on POWER9 DD2.3,
557              * while still working on POWER9 DD2.2 and POWER8 host
558              * cpus.
559              */
560             return;
561         }
562         error_setg(errp,
563                    "Requested count cache flush assist capability level not supported by KVM");
564         error_append_hint(errp, "Try appending -machine cap-ccf-assist=off\n");
565     }
566 }
567 
568 static void cap_fwnmi_apply(SpaprMachineState *spapr, uint8_t val,
569                                 Error **errp)
570 {
571     ERRP_GUARD();
572     if (!val) {
573         return; /* Disabled by default */
574     }
575 
576     if (kvm_enabled()) {
577         if (!kvmppc_get_fwnmi()) {
578             error_setg(errp,
579 "Firmware Assisted Non-Maskable Interrupts(FWNMI) not supported by KVM.");
580             error_append_hint(errp, "Try appending -machine cap-fwnmi=off\n");
581         }
582     }
583 }
584 
585 SpaprCapabilityInfo capability_table[SPAPR_CAP_NUM] = {
586     [SPAPR_CAP_HTM] = {
587         .name = "htm",
588         .description = "Allow Hardware Transactional Memory (HTM)",
589         .index = SPAPR_CAP_HTM,
590         .get = spapr_cap_get_bool,
591         .set = spapr_cap_set_bool,
592         .type = "bool",
593         .apply = cap_htm_apply,
594     },
595     [SPAPR_CAP_VSX] = {
596         .name = "vsx",
597         .description = "Allow Vector Scalar Extensions (VSX)",
598         .index = SPAPR_CAP_VSX,
599         .get = spapr_cap_get_bool,
600         .set = spapr_cap_set_bool,
601         .type = "bool",
602         .apply = cap_vsx_apply,
603     },
604     [SPAPR_CAP_DFP] = {
605         .name = "dfp",
606         .description = "Allow Decimal Floating Point (DFP)",
607         .index = SPAPR_CAP_DFP,
608         .get = spapr_cap_get_bool,
609         .set = spapr_cap_set_bool,
610         .type = "bool",
611         .apply = cap_dfp_apply,
612     },
613     [SPAPR_CAP_CFPC] = {
614         .name = "cfpc",
615         .description = "Cache Flush on Privilege Change" VALUE_DESC_TRISTATE,
616         .index = SPAPR_CAP_CFPC,
617         .get = spapr_cap_get_string,
618         .set = spapr_cap_set_string,
619         .type = "string",
620         .possible = &cap_cfpc_possible,
621         .apply = cap_safe_cache_apply,
622     },
623     [SPAPR_CAP_SBBC] = {
624         .name = "sbbc",
625         .description = "Speculation Barrier Bounds Checking" VALUE_DESC_TRISTATE,
626         .index = SPAPR_CAP_SBBC,
627         .get = spapr_cap_get_string,
628         .set = spapr_cap_set_string,
629         .type = "string",
630         .possible = &cap_sbbc_possible,
631         .apply = cap_safe_bounds_check_apply,
632     },
633     [SPAPR_CAP_IBS] = {
634         .name = "ibs",
635         .description =
636             "Indirect Branch Speculation (broken, workaround, fixed-ibs,"
637             "fixed-ccd, fixed-na)",
638         .index = SPAPR_CAP_IBS,
639         .get = spapr_cap_get_string,
640         .set = spapr_cap_set_string,
641         .type = "string",
642         .possible = &cap_ibs_possible,
643         .apply = cap_safe_indirect_branch_apply,
644     },
645     [SPAPR_CAP_HPT_MAXPAGESIZE] = {
646         .name = "hpt-max-page-size",
647         .description = "Maximum page size for Hash Page Table guests",
648         .index = SPAPR_CAP_HPT_MAXPAGESIZE,
649         .get = spapr_cap_get_pagesize,
650         .set = spapr_cap_set_pagesize,
651         .type = "int",
652         .apply = cap_hpt_maxpagesize_apply,
653         .cpu_apply = cap_hpt_maxpagesize_cpu_apply,
654         .migrate_needed = cap_hpt_maxpagesize_migrate_needed,
655     },
656     [SPAPR_CAP_NESTED_KVM_HV] = {
657         .name = "nested-hv",
658         .description = "Allow Nested KVM-HV",
659         .index = SPAPR_CAP_NESTED_KVM_HV,
660         .get = spapr_cap_get_bool,
661         .set = spapr_cap_set_bool,
662         .type = "bool",
663         .apply = cap_nested_kvm_hv_apply,
664     },
665     [SPAPR_CAP_LARGE_DECREMENTER] = {
666         .name = "large-decr",
667         .description = "Allow Large Decrementer",
668         .index = SPAPR_CAP_LARGE_DECREMENTER,
669         .get = spapr_cap_get_bool,
670         .set = spapr_cap_set_bool,
671         .type = "bool",
672         .apply = cap_large_decr_apply,
673         .cpu_apply = cap_large_decr_cpu_apply,
674     },
675     [SPAPR_CAP_CCF_ASSIST] = {
676         .name = "ccf-assist",
677         .description = "Count Cache Flush Assist via HW Instruction",
678         .index = SPAPR_CAP_CCF_ASSIST,
679         .get = spapr_cap_get_bool,
680         .set = spapr_cap_set_bool,
681         .type = "bool",
682         .apply = cap_ccf_assist_apply,
683     },
684     [SPAPR_CAP_FWNMI] = {
685         .name = "fwnmi",
686         .description = "Implements PAPR FWNMI option",
687         .index = SPAPR_CAP_FWNMI,
688         .get = spapr_cap_get_bool,
689         .set = spapr_cap_set_bool,
690         .type = "bool",
691         .apply = cap_fwnmi_apply,
692     },
693 };
694 
695 static SpaprCapabilities default_caps_with_cpu(SpaprMachineState *spapr,
696                                                const char *cputype)
697 {
698     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
699     SpaprCapabilities caps;
700 
701     caps = smc->default_caps;
702 
703     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_3_00,
704                                0, spapr->max_compat_pvr)) {
705         caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
706     }
707 
708     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_2_07,
709                                0, spapr->max_compat_pvr)) {
710         caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
711         caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
712     }
713 
714     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_2_06_PLUS,
715                                0, spapr->max_compat_pvr)) {
716         caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
717     }
718 
719     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_2_06,
720                                0, spapr->max_compat_pvr)) {
721         caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_OFF;
722         caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_OFF;
723         caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
724     }
725 
726     /* This is for pseries-2.12 and older */
727     if (smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] == 0) {
728         uint8_t mps;
729 
730         if (kvmppc_hpt_needs_host_contiguous_pages()) {
731             mps = ctz64(qemu_minrampagesize());
732         } else {
733             mps = 34; /* allow everything up to 16GiB, i.e. everything */
734         }
735 
736         caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = mps;
737     }
738 
739     return caps;
740 }
741 
742 int spapr_caps_pre_load(void *opaque)
743 {
744     SpaprMachineState *spapr = opaque;
745 
746     /* Set to default so we can tell if this came in with the migration */
747     spapr->mig = spapr->def;
748     return 0;
749 }
750 
751 int spapr_caps_pre_save(void *opaque)
752 {
753     SpaprMachineState *spapr = opaque;
754 
755     spapr->mig = spapr->eff;
756     return 0;
757 }
758 
759 /* This has to be called from the top-level spapr post_load, not the
760  * caps specific one.  Otherwise it wouldn't be called when the source
761  * caps are all defaults, which could still conflict with overridden
762  * caps on the destination */
763 int spapr_caps_post_migration(SpaprMachineState *spapr)
764 {
765     int i;
766     bool ok = true;
767     SpaprCapabilities dstcaps = spapr->eff;
768     SpaprCapabilities srccaps;
769 
770     srccaps = default_caps_with_cpu(spapr, MACHINE(spapr)->cpu_type);
771     for (i = 0; i < SPAPR_CAP_NUM; i++) {
772         /* If not default value then assume came in with the migration */
773         if (spapr->mig.caps[i] != spapr->def.caps[i]) {
774             srccaps.caps[i] = spapr->mig.caps[i];
775         }
776     }
777 
778     for (i = 0; i < SPAPR_CAP_NUM; i++) {
779         SpaprCapabilityInfo *info = &capability_table[i];
780 
781         if (srccaps.caps[i] > dstcaps.caps[i]) {
782             error_report("cap-%s higher level (%d) in incoming stream than on destination (%d)",
783                          info->name, srccaps.caps[i], dstcaps.caps[i]);
784             ok = false;
785         }
786 
787         if (srccaps.caps[i] < dstcaps.caps[i]) {
788             warn_report("cap-%s lower level (%d) in incoming stream than on destination (%d)",
789                          info->name, srccaps.caps[i], dstcaps.caps[i]);
790         }
791     }
792 
793     return ok ? 0 : -EINVAL;
794 }
795 
796 /* Used to generate the migration field and needed function for a spapr cap */
797 #define SPAPR_CAP_MIG_STATE(sname, cap)                 \
798 static bool spapr_cap_##sname##_needed(void *opaque)    \
799 {                                                       \
800     SpaprMachineState *spapr = opaque;                  \
801     bool (*needed)(void *opaque) =                      \
802         capability_table[cap].migrate_needed;           \
803                                                         \
804     return needed ? needed(opaque) : true &&            \
805            spapr->cmd_line_caps[cap] &&                 \
806            (spapr->eff.caps[cap] !=                     \
807             spapr->def.caps[cap]);                      \
808 }                                                       \
809                                                         \
810 const VMStateDescription vmstate_spapr_cap_##sname = {  \
811     .name = "spapr/cap/" #sname,                        \
812     .version_id = 1,                                    \
813     .minimum_version_id = 1,                            \
814     .needed = spapr_cap_##sname##_needed,               \
815     .fields = (VMStateField[]) {                        \
816         VMSTATE_UINT8(mig.caps[cap],                    \
817                       SpaprMachineState),               \
818         VMSTATE_END_OF_LIST()                           \
819     },                                                  \
820 }
821 
822 SPAPR_CAP_MIG_STATE(htm, SPAPR_CAP_HTM);
823 SPAPR_CAP_MIG_STATE(vsx, SPAPR_CAP_VSX);
824 SPAPR_CAP_MIG_STATE(dfp, SPAPR_CAP_DFP);
825 SPAPR_CAP_MIG_STATE(cfpc, SPAPR_CAP_CFPC);
826 SPAPR_CAP_MIG_STATE(sbbc, SPAPR_CAP_SBBC);
827 SPAPR_CAP_MIG_STATE(ibs, SPAPR_CAP_IBS);
828 SPAPR_CAP_MIG_STATE(hpt_maxpagesize, SPAPR_CAP_HPT_MAXPAGESIZE);
829 SPAPR_CAP_MIG_STATE(nested_kvm_hv, SPAPR_CAP_NESTED_KVM_HV);
830 SPAPR_CAP_MIG_STATE(large_decr, SPAPR_CAP_LARGE_DECREMENTER);
831 SPAPR_CAP_MIG_STATE(ccf_assist, SPAPR_CAP_CCF_ASSIST);
832 SPAPR_CAP_MIG_STATE(fwnmi, SPAPR_CAP_FWNMI);
833 
834 void spapr_caps_init(SpaprMachineState *spapr)
835 {
836     SpaprCapabilities default_caps;
837     int i;
838 
839     /* Compute the actual set of caps we should run with */
840     default_caps = default_caps_with_cpu(spapr, MACHINE(spapr)->cpu_type);
841 
842     for (i = 0; i < SPAPR_CAP_NUM; i++) {
843         /* Store the defaults */
844         spapr->def.caps[i] = default_caps.caps[i];
845         /* If not set on the command line then apply the default value */
846         if (!spapr->cmd_line_caps[i]) {
847             spapr->eff.caps[i] = default_caps.caps[i];
848         }
849     }
850 }
851 
852 void spapr_caps_apply(SpaprMachineState *spapr)
853 {
854     int i;
855 
856     for (i = 0; i < SPAPR_CAP_NUM; i++) {
857         SpaprCapabilityInfo *info = &capability_table[i];
858 
859         /*
860          * If the apply function can't set the desired level and thinks it's
861          * fatal, it should cause that.
862          */
863         info->apply(spapr, spapr->eff.caps[i], &error_fatal);
864     }
865 }
866 
867 void spapr_caps_cpu_apply(SpaprMachineState *spapr, PowerPCCPU *cpu)
868 {
869     int i;
870 
871     for (i = 0; i < SPAPR_CAP_NUM; i++) {
872         SpaprCapabilityInfo *info = &capability_table[i];
873 
874         /*
875          * If the apply function can't set the desired level and thinks it's
876          * fatal, it should cause that.
877          */
878         if (info->cpu_apply) {
879             info->cpu_apply(spapr, cpu, spapr->eff.caps[i], &error_fatal);
880         }
881     }
882 }
883 
884 void spapr_caps_add_properties(SpaprMachineClass *smc)
885 {
886     ObjectClass *klass = OBJECT_CLASS(smc);
887     int i;
888 
889     for (i = 0; i < ARRAY_SIZE(capability_table); i++) {
890         SpaprCapabilityInfo *cap = &capability_table[i];
891         char *name = g_strdup_printf("cap-%s", cap->name);
892         char *desc;
893 
894         object_class_property_add(klass, name, cap->type,
895                                   cap->get, cap->set,
896                                   NULL, cap);
897 
898         desc = g_strdup_printf("%s", cap->description);
899         object_class_property_set_description(klass, name, desc);
900         g_free(name);
901         g_free(desc);
902     }
903 }
904