ppc/xive: Add support for PQ state bits offload
[qemu.git] / hw / ppc / spapr_hcall.c
blobf0082907872caa1bb6c137f5372425b86f973fee
1 #include "qemu/osdep.h"
2 #include "qemu/cutils.h"
3 #include "qapi/error.h"
4 #include "sysemu/hw_accel.h"
5 #include "sysemu/runstate.h"
6 #include "qemu/log.h"
7 #include "qemu/main-loop.h"
8 #include "qemu/module.h"
9 #include "qemu/error-report.h"
10 #include "exec/exec-all.h"
11 #include "helper_regs.h"
12 #include "hw/ppc/ppc.h"
13 #include "hw/ppc/spapr.h"
14 #include "hw/ppc/spapr_cpu_core.h"
15 #include "mmu-hash64.h"
16 #include "cpu-models.h"
17 #include "trace.h"
18 #include "kvm_ppc.h"
19 #include "hw/ppc/fdt.h"
20 #include "hw/ppc/spapr_ovec.h"
21 #include "hw/ppc/spapr_numa.h"
22 #include "mmu-book3s-v3.h"
23 #include "hw/mem/memory-device.h"
25 bool is_ram_address(SpaprMachineState *spapr, hwaddr addr)
27 MachineState *machine = MACHINE(spapr);
28 DeviceMemoryState *dms = machine->device_memory;
30 if (addr < machine->ram_size) {
31 return true;
33 if ((addr >= dms->base)
34 && ((addr - dms->base) < memory_region_size(&dms->mr))) {
35 return true;
38 return false;
41 /* Convert a return code from the KVM ioctl()s implementing resize HPT
42 * into a PAPR hypercall return code */
43 static target_ulong resize_hpt_convert_rc(int ret)
45 if (ret >= 100000) {
46 return H_LONG_BUSY_ORDER_100_SEC;
47 } else if (ret >= 10000) {
48 return H_LONG_BUSY_ORDER_10_SEC;
49 } else if (ret >= 1000) {
50 return H_LONG_BUSY_ORDER_1_SEC;
51 } else if (ret >= 100) {
52 return H_LONG_BUSY_ORDER_100_MSEC;
53 } else if (ret >= 10) {
54 return H_LONG_BUSY_ORDER_10_MSEC;
55 } else if (ret > 0) {
56 return H_LONG_BUSY_ORDER_1_MSEC;
59 switch (ret) {
60 case 0:
61 return H_SUCCESS;
62 case -EPERM:
63 return H_AUTHORITY;
64 case -EINVAL:
65 return H_PARAMETER;
66 case -ENXIO:
67 return H_CLOSED;
68 case -ENOSPC:
69 return H_PTEG_FULL;
70 case -EBUSY:
71 return H_BUSY;
72 case -ENOMEM:
73 return H_NO_MEM;
74 default:
75 return H_HARDWARE;
79 static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu,
80 SpaprMachineState *spapr,
81 target_ulong opcode,
82 target_ulong *args)
84 target_ulong flags = args[0];
85 int shift = args[1];
86 uint64_t current_ram_size;
87 int rc;
89 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
90 return H_AUTHORITY;
93 if (!spapr->htab_shift) {
94 /* Radix guest, no HPT */
95 return H_NOT_AVAILABLE;
98 trace_spapr_h_resize_hpt_prepare(flags, shift);
100 if (flags != 0) {
101 return H_PARAMETER;
104 if (shift && ((shift < 18) || (shift > 46))) {
105 return H_PARAMETER;
108 current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
110 /* We only allow the guest to allocate an HPT one order above what
111 * we'd normally give them (to stop a small guest claiming a huge
112 * chunk of resources in the HPT */
113 if (shift > (spapr_hpt_shift_for_ramsize(current_ram_size) + 1)) {
114 return H_RESOURCE;
117 rc = kvmppc_resize_hpt_prepare(cpu, flags, shift);
118 if (rc != -ENOSYS) {
119 return resize_hpt_convert_rc(rc);
122 if (kvm_enabled()) {
123 return H_HARDWARE;
126 return softmmu_resize_hpt_prepare(cpu, spapr, shift);
129 static void do_push_sregs_to_kvm_pr(CPUState *cs, run_on_cpu_data data)
131 int ret;
133 cpu_synchronize_state(cs);
135 ret = kvmppc_put_books_sregs(POWERPC_CPU(cs));
136 if (ret < 0) {
137 error_report("failed to push sregs to KVM: %s", strerror(-ret));
138 exit(1);
142 void push_sregs_to_kvm_pr(SpaprMachineState *spapr)
144 CPUState *cs;
147 * This is a hack for the benefit of KVM PR - it abuses the SDR1
148 * slot in kvm_sregs to communicate the userspace address of the
149 * HPT
151 if (!kvm_enabled() || !spapr->htab) {
152 return;
155 CPU_FOREACH(cs) {
156 run_on_cpu(cs, do_push_sregs_to_kvm_pr, RUN_ON_CPU_NULL);
160 static target_ulong h_resize_hpt_commit(PowerPCCPU *cpu,
161 SpaprMachineState *spapr,
162 target_ulong opcode,
163 target_ulong *args)
165 target_ulong flags = args[0];
166 target_ulong shift = args[1];
167 int rc;
169 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
170 return H_AUTHORITY;
173 if (!spapr->htab_shift) {
174 /* Radix guest, no HPT */
175 return H_NOT_AVAILABLE;
178 trace_spapr_h_resize_hpt_commit(flags, shift);
180 rc = kvmppc_resize_hpt_commit(cpu, flags, shift);
181 if (rc != -ENOSYS) {
182 rc = resize_hpt_convert_rc(rc);
183 if (rc == H_SUCCESS) {
184 /* Need to set the new htab_shift in the machine state */
185 spapr->htab_shift = shift;
187 return rc;
190 if (kvm_enabled()) {
191 return H_HARDWARE;
194 return softmmu_resize_hpt_commit(cpu, spapr, flags, shift);
199 static target_ulong h_set_sprg0(PowerPCCPU *cpu, SpaprMachineState *spapr,
200 target_ulong opcode, target_ulong *args)
202 cpu_synchronize_state(CPU(cpu));
203 cpu->env.spr[SPR_SPRG0] = args[0];
205 return H_SUCCESS;
208 static target_ulong h_set_dabr(PowerPCCPU *cpu, SpaprMachineState *spapr,
209 target_ulong opcode, target_ulong *args)
211 if (!ppc_has_spr(cpu, SPR_DABR)) {
212 return H_HARDWARE; /* DABR register not available */
214 cpu_synchronize_state(CPU(cpu));
216 if (ppc_has_spr(cpu, SPR_DABRX)) {
217 cpu->env.spr[SPR_DABRX] = 0x3; /* Use Problem and Privileged state */
218 } else if (!(args[0] & 0x4)) { /* Breakpoint Translation set? */
219 return H_RESERVED_DABR;
222 cpu->env.spr[SPR_DABR] = args[0];
223 return H_SUCCESS;
226 static target_ulong h_set_xdabr(PowerPCCPU *cpu, SpaprMachineState *spapr,
227 target_ulong opcode, target_ulong *args)
229 target_ulong dabrx = args[1];
231 if (!ppc_has_spr(cpu, SPR_DABR) || !ppc_has_spr(cpu, SPR_DABRX)) {
232 return H_HARDWARE;
235 if ((dabrx & ~0xfULL) != 0 || (dabrx & H_DABRX_HYPERVISOR) != 0
236 || (dabrx & (H_DABRX_KERNEL | H_DABRX_USER)) == 0) {
237 return H_PARAMETER;
240 cpu_synchronize_state(CPU(cpu));
241 cpu->env.spr[SPR_DABRX] = dabrx;
242 cpu->env.spr[SPR_DABR] = args[0];
244 return H_SUCCESS;
247 static target_ulong h_page_init(PowerPCCPU *cpu, SpaprMachineState *spapr,
248 target_ulong opcode, target_ulong *args)
250 target_ulong flags = args[0];
251 hwaddr dst = args[1];
252 hwaddr src = args[2];
253 hwaddr len = TARGET_PAGE_SIZE;
254 uint8_t *pdst, *psrc;
255 target_long ret = H_SUCCESS;
257 if (flags & ~(H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE
258 | H_COPY_PAGE | H_ZERO_PAGE)) {
259 qemu_log_mask(LOG_UNIMP, "h_page_init: Bad flags (" TARGET_FMT_lx "\n",
260 flags);
261 return H_PARAMETER;
264 /* Map-in destination */
265 if (!is_ram_address(spapr, dst) || (dst & ~TARGET_PAGE_MASK) != 0) {
266 return H_PARAMETER;
268 pdst = cpu_physical_memory_map(dst, &len, true);
269 if (!pdst || len != TARGET_PAGE_SIZE) {
270 return H_PARAMETER;
273 if (flags & H_COPY_PAGE) {
274 /* Map-in source, copy to destination, and unmap source again */
275 if (!is_ram_address(spapr, src) || (src & ~TARGET_PAGE_MASK) != 0) {
276 ret = H_PARAMETER;
277 goto unmap_out;
279 psrc = cpu_physical_memory_map(src, &len, false);
280 if (!psrc || len != TARGET_PAGE_SIZE) {
281 ret = H_PARAMETER;
282 goto unmap_out;
284 memcpy(pdst, psrc, len);
285 cpu_physical_memory_unmap(psrc, len, 0, len);
286 } else if (flags & H_ZERO_PAGE) {
287 memset(pdst, 0, len); /* Just clear the destination page */
290 if (kvm_enabled() && (flags & H_ICACHE_SYNCHRONIZE) != 0) {
291 kvmppc_dcbst_range(cpu, pdst, len);
293 if (flags & (H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE)) {
294 if (kvm_enabled()) {
295 kvmppc_icbi_range(cpu, pdst, len);
296 } else {
297 tb_flush(CPU(cpu));
301 unmap_out:
302 cpu_physical_memory_unmap(pdst, TARGET_PAGE_SIZE, 1, len);
303 return ret;
306 #define FLAGS_REGISTER_VPA 0x0000200000000000ULL
307 #define FLAGS_REGISTER_DTL 0x0000400000000000ULL
308 #define FLAGS_REGISTER_SLBSHADOW 0x0000600000000000ULL
309 #define FLAGS_DEREGISTER_VPA 0x0000a00000000000ULL
310 #define FLAGS_DEREGISTER_DTL 0x0000c00000000000ULL
311 #define FLAGS_DEREGISTER_SLBSHADOW 0x0000e00000000000ULL
313 static target_ulong register_vpa(PowerPCCPU *cpu, target_ulong vpa)
315 CPUState *cs = CPU(cpu);
316 CPUPPCState *env = &cpu->env;
317 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
318 uint16_t size;
319 uint8_t tmp;
321 if (vpa == 0) {
322 hcall_dprintf("Can't cope with registering a VPA at logical 0\n");
323 return H_HARDWARE;
326 if (vpa % env->dcache_line_size) {
327 return H_PARAMETER;
329 /* FIXME: bounds check the address */
331 size = lduw_be_phys(cs->as, vpa + 0x4);
333 if (size < VPA_MIN_SIZE) {
334 return H_PARAMETER;
337 /* VPA is not allowed to cross a page boundary */
338 if ((vpa / 4096) != ((vpa + size - 1) / 4096)) {
339 return H_PARAMETER;
342 spapr_cpu->vpa_addr = vpa;
344 tmp = ldub_phys(cs->as, spapr_cpu->vpa_addr + VPA_SHARED_PROC_OFFSET);
345 tmp |= VPA_SHARED_PROC_VAL;
346 stb_phys(cs->as, spapr_cpu->vpa_addr + VPA_SHARED_PROC_OFFSET, tmp);
348 return H_SUCCESS;
351 static target_ulong deregister_vpa(PowerPCCPU *cpu, target_ulong vpa)
353 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
355 if (spapr_cpu->slb_shadow_addr) {
356 return H_RESOURCE;
359 if (spapr_cpu->dtl_addr) {
360 return H_RESOURCE;
363 spapr_cpu->vpa_addr = 0;
364 return H_SUCCESS;
367 static target_ulong register_slb_shadow(PowerPCCPU *cpu, target_ulong addr)
369 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
370 uint32_t size;
372 if (addr == 0) {
373 hcall_dprintf("Can't cope with SLB shadow at logical 0\n");
374 return H_HARDWARE;
377 size = ldl_be_phys(CPU(cpu)->as, addr + 0x4);
378 if (size < 0x8) {
379 return H_PARAMETER;
382 if ((addr / 4096) != ((addr + size - 1) / 4096)) {
383 return H_PARAMETER;
386 if (!spapr_cpu->vpa_addr) {
387 return H_RESOURCE;
390 spapr_cpu->slb_shadow_addr = addr;
391 spapr_cpu->slb_shadow_size = size;
393 return H_SUCCESS;
396 static target_ulong deregister_slb_shadow(PowerPCCPU *cpu, target_ulong addr)
398 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
400 spapr_cpu->slb_shadow_addr = 0;
401 spapr_cpu->slb_shadow_size = 0;
402 return H_SUCCESS;
405 static target_ulong register_dtl(PowerPCCPU *cpu, target_ulong addr)
407 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
408 uint32_t size;
410 if (addr == 0) {
411 hcall_dprintf("Can't cope with DTL at logical 0\n");
412 return H_HARDWARE;
415 size = ldl_be_phys(CPU(cpu)->as, addr + 0x4);
417 if (size < 48) {
418 return H_PARAMETER;
421 if (!spapr_cpu->vpa_addr) {
422 return H_RESOURCE;
425 spapr_cpu->dtl_addr = addr;
426 spapr_cpu->dtl_size = size;
428 return H_SUCCESS;
431 static target_ulong deregister_dtl(PowerPCCPU *cpu, target_ulong addr)
433 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
435 spapr_cpu->dtl_addr = 0;
436 spapr_cpu->dtl_size = 0;
438 return H_SUCCESS;
441 static target_ulong h_register_vpa(PowerPCCPU *cpu, SpaprMachineState *spapr,
442 target_ulong opcode, target_ulong *args)
444 target_ulong flags = args[0];
445 target_ulong procno = args[1];
446 target_ulong vpa = args[2];
447 target_ulong ret = H_PARAMETER;
448 PowerPCCPU *tcpu;
450 tcpu = spapr_find_cpu(procno);
451 if (!tcpu) {
452 return H_PARAMETER;
455 switch (flags) {
456 case FLAGS_REGISTER_VPA:
457 ret = register_vpa(tcpu, vpa);
458 break;
460 case FLAGS_DEREGISTER_VPA:
461 ret = deregister_vpa(tcpu, vpa);
462 break;
464 case FLAGS_REGISTER_SLBSHADOW:
465 ret = register_slb_shadow(tcpu, vpa);
466 break;
468 case FLAGS_DEREGISTER_SLBSHADOW:
469 ret = deregister_slb_shadow(tcpu, vpa);
470 break;
472 case FLAGS_REGISTER_DTL:
473 ret = register_dtl(tcpu, vpa);
474 break;
476 case FLAGS_DEREGISTER_DTL:
477 ret = deregister_dtl(tcpu, vpa);
478 break;
481 return ret;
484 static target_ulong h_cede(PowerPCCPU *cpu, SpaprMachineState *spapr,
485 target_ulong opcode, target_ulong *args)
487 CPUPPCState *env = &cpu->env;
488 CPUState *cs = CPU(cpu);
489 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
491 env->msr |= (1ULL << MSR_EE);
492 hreg_compute_hflags(env);
494 if (spapr_cpu->prod) {
495 spapr_cpu->prod = false;
496 return H_SUCCESS;
499 if (!cpu_has_work(cs)) {
500 cs->halted = 1;
501 cs->exception_index = EXCP_HLT;
502 cs->exit_request = 1;
505 return H_SUCCESS;
509 * Confer to self, aka join. Cede could use the same pattern as well, if
510 * EXCP_HLT can be changed to ECXP_HALTED.
512 static target_ulong h_confer_self(PowerPCCPU *cpu)
514 CPUState *cs = CPU(cpu);
515 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
517 if (spapr_cpu->prod) {
518 spapr_cpu->prod = false;
519 return H_SUCCESS;
521 cs->halted = 1;
522 cs->exception_index = EXCP_HALTED;
523 cs->exit_request = 1;
525 return H_SUCCESS;
528 static target_ulong h_join(PowerPCCPU *cpu, SpaprMachineState *spapr,
529 target_ulong opcode, target_ulong *args)
531 CPUPPCState *env = &cpu->env;
532 CPUState *cs;
533 bool last_unjoined = true;
535 if (env->msr & (1ULL << MSR_EE)) {
536 return H_BAD_MODE;
540 * Must not join the last CPU running. Interestingly, no such restriction
541 * for H_CONFER-to-self, but that is probably not intended to be used
542 * when H_JOIN is available.
544 CPU_FOREACH(cs) {
545 PowerPCCPU *c = POWERPC_CPU(cs);
546 CPUPPCState *e = &c->env;
547 if (c == cpu) {
548 continue;
551 /* Don't have a way to indicate joined, so use halted && MSR[EE]=0 */
552 if (!cs->halted || (e->msr & (1ULL << MSR_EE))) {
553 last_unjoined = false;
554 break;
557 if (last_unjoined) {
558 return H_CONTINUE;
561 return h_confer_self(cpu);
564 static target_ulong h_confer(PowerPCCPU *cpu, SpaprMachineState *spapr,
565 target_ulong opcode, target_ulong *args)
567 target_long target = args[0];
568 uint32_t dispatch = args[1];
569 CPUState *cs = CPU(cpu);
570 SpaprCpuState *spapr_cpu;
573 * -1 means confer to all other CPUs without dispatch counter check,
574 * otherwise it's a targeted confer.
576 if (target != -1) {
577 PowerPCCPU *target_cpu = spapr_find_cpu(target);
578 uint32_t target_dispatch;
580 if (!target_cpu) {
581 return H_PARAMETER;
585 * target == self is a special case, we wait until prodded, without
586 * dispatch counter check.
588 if (cpu == target_cpu) {
589 return h_confer_self(cpu);
592 spapr_cpu = spapr_cpu_state(target_cpu);
593 if (!spapr_cpu->vpa_addr || ((dispatch & 1) == 0)) {
594 return H_SUCCESS;
597 target_dispatch = ldl_be_phys(cs->as,
598 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
599 if (target_dispatch != dispatch) {
600 return H_SUCCESS;
604 * The targeted confer does not do anything special beyond yielding
605 * the current vCPU, but even this should be better than nothing.
606 * At least for single-threaded tcg, it gives the target a chance to
607 * run before we run again. Multi-threaded tcg does not really do
608 * anything with EXCP_YIELD yet.
612 cs->exception_index = EXCP_YIELD;
613 cs->exit_request = 1;
614 cpu_loop_exit(cs);
616 return H_SUCCESS;
619 static target_ulong h_prod(PowerPCCPU *cpu, SpaprMachineState *spapr,
620 target_ulong opcode, target_ulong *args)
622 target_long target = args[0];
623 PowerPCCPU *tcpu;
624 CPUState *cs;
625 SpaprCpuState *spapr_cpu;
627 tcpu = spapr_find_cpu(target);
628 cs = CPU(tcpu);
629 if (!cs) {
630 return H_PARAMETER;
633 spapr_cpu = spapr_cpu_state(tcpu);
634 spapr_cpu->prod = true;
635 cs->halted = 0;
636 qemu_cpu_kick(cs);
638 return H_SUCCESS;
641 static target_ulong h_rtas(PowerPCCPU *cpu, SpaprMachineState *spapr,
642 target_ulong opcode, target_ulong *args)
644 target_ulong rtas_r3 = args[0];
645 uint32_t token = rtas_ld(rtas_r3, 0);
646 uint32_t nargs = rtas_ld(rtas_r3, 1);
647 uint32_t nret = rtas_ld(rtas_r3, 2);
649 return spapr_rtas_call(cpu, spapr, token, nargs, rtas_r3 + 12,
650 nret, rtas_r3 + 12 + 4*nargs);
653 static target_ulong h_logical_load(PowerPCCPU *cpu, SpaprMachineState *spapr,
654 target_ulong opcode, target_ulong *args)
656 CPUState *cs = CPU(cpu);
657 target_ulong size = args[0];
658 target_ulong addr = args[1];
660 switch (size) {
661 case 1:
662 args[0] = ldub_phys(cs->as, addr);
663 return H_SUCCESS;
664 case 2:
665 args[0] = lduw_phys(cs->as, addr);
666 return H_SUCCESS;
667 case 4:
668 args[0] = ldl_phys(cs->as, addr);
669 return H_SUCCESS;
670 case 8:
671 args[0] = ldq_phys(cs->as, addr);
672 return H_SUCCESS;
674 return H_PARAMETER;
677 static target_ulong h_logical_store(PowerPCCPU *cpu, SpaprMachineState *spapr,
678 target_ulong opcode, target_ulong *args)
680 CPUState *cs = CPU(cpu);
682 target_ulong size = args[0];
683 target_ulong addr = args[1];
684 target_ulong val = args[2];
686 switch (size) {
687 case 1:
688 stb_phys(cs->as, addr, val);
689 return H_SUCCESS;
690 case 2:
691 stw_phys(cs->as, addr, val);
692 return H_SUCCESS;
693 case 4:
694 stl_phys(cs->as, addr, val);
695 return H_SUCCESS;
696 case 8:
697 stq_phys(cs->as, addr, val);
698 return H_SUCCESS;
700 return H_PARAMETER;
703 static target_ulong h_logical_memop(PowerPCCPU *cpu, SpaprMachineState *spapr,
704 target_ulong opcode, target_ulong *args)
706 CPUState *cs = CPU(cpu);
708 target_ulong dst = args[0]; /* Destination address */
709 target_ulong src = args[1]; /* Source address */
710 target_ulong esize = args[2]; /* Element size (0=1,1=2,2=4,3=8) */
711 target_ulong count = args[3]; /* Element count */
712 target_ulong op = args[4]; /* 0 = copy, 1 = invert */
713 uint64_t tmp;
714 unsigned int mask = (1 << esize) - 1;
715 int step = 1 << esize;
717 if (count > 0x80000000) {
718 return H_PARAMETER;
721 if ((dst & mask) || (src & mask) || (op > 1)) {
722 return H_PARAMETER;
725 if (dst >= src && dst < (src + (count << esize))) {
726 dst = dst + ((count - 1) << esize);
727 src = src + ((count - 1) << esize);
728 step = -step;
731 while (count--) {
732 switch (esize) {
733 case 0:
734 tmp = ldub_phys(cs->as, src);
735 break;
736 case 1:
737 tmp = lduw_phys(cs->as, src);
738 break;
739 case 2:
740 tmp = ldl_phys(cs->as, src);
741 break;
742 case 3:
743 tmp = ldq_phys(cs->as, src);
744 break;
745 default:
746 return H_PARAMETER;
748 if (op == 1) {
749 tmp = ~tmp;
751 switch (esize) {
752 case 0:
753 stb_phys(cs->as, dst, tmp);
754 break;
755 case 1:
756 stw_phys(cs->as, dst, tmp);
757 break;
758 case 2:
759 stl_phys(cs->as, dst, tmp);
760 break;
761 case 3:
762 stq_phys(cs->as, dst, tmp);
763 break;
765 dst = dst + step;
766 src = src + step;
769 return H_SUCCESS;
772 static target_ulong h_logical_icbi(PowerPCCPU *cpu, SpaprMachineState *spapr,
773 target_ulong opcode, target_ulong *args)
775 /* Nothing to do on emulation, KVM will trap this in the kernel */
776 return H_SUCCESS;
779 static target_ulong h_logical_dcbf(PowerPCCPU *cpu, SpaprMachineState *spapr,
780 target_ulong opcode, target_ulong *args)
782 /* Nothing to do on emulation, KVM will trap this in the kernel */
783 return H_SUCCESS;
786 static target_ulong h_set_mode_resource_le(PowerPCCPU *cpu,
787 SpaprMachineState *spapr,
788 target_ulong mflags,
789 target_ulong value1,
790 target_ulong value2)
792 if (value1) {
793 return H_P3;
795 if (value2) {
796 return H_P4;
799 switch (mflags) {
800 case H_SET_MODE_ENDIAN_BIG:
801 spapr_set_all_lpcrs(0, LPCR_ILE);
802 spapr_pci_switch_vga(spapr, true);
803 return H_SUCCESS;
805 case H_SET_MODE_ENDIAN_LITTLE:
806 spapr_set_all_lpcrs(LPCR_ILE, LPCR_ILE);
807 spapr_pci_switch_vga(spapr, false);
808 return H_SUCCESS;
811 return H_UNSUPPORTED_FLAG;
814 static target_ulong h_set_mode_resource_addr_trans_mode(PowerPCCPU *cpu,
815 target_ulong mflags,
816 target_ulong value1,
817 target_ulong value2)
819 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
821 if (!(pcc->insns_flags2 & PPC2_ISA207S)) {
822 return H_P2;
824 if (value1) {
825 return H_P3;
827 if (value2) {
828 return H_P4;
831 if (mflags == 1) {
832 /* AIL=1 is reserved in POWER8/POWER9/POWER10 */
833 return H_UNSUPPORTED_FLAG;
836 if (mflags == 2 && (pcc->insns_flags2 & PPC2_ISA310)) {
837 /* AIL=2 is reserved in POWER10 (ISA v3.1) */
838 return H_UNSUPPORTED_FLAG;
841 spapr_set_all_lpcrs(mflags << LPCR_AIL_SHIFT, LPCR_AIL);
843 return H_SUCCESS;
846 static target_ulong h_set_mode(PowerPCCPU *cpu, SpaprMachineState *spapr,
847 target_ulong opcode, target_ulong *args)
849 target_ulong resource = args[1];
850 target_ulong ret = H_P2;
852 switch (resource) {
853 case H_SET_MODE_RESOURCE_LE:
854 ret = h_set_mode_resource_le(cpu, spapr, args[0], args[2], args[3]);
855 break;
856 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
857 ret = h_set_mode_resource_addr_trans_mode(cpu, args[0],
858 args[2], args[3]);
859 break;
862 return ret;
865 static target_ulong h_clean_slb(PowerPCCPU *cpu, SpaprMachineState *spapr,
866 target_ulong opcode, target_ulong *args)
868 qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n",
869 opcode, " (H_CLEAN_SLB)");
870 return H_FUNCTION;
873 static target_ulong h_invalidate_pid(PowerPCCPU *cpu, SpaprMachineState *spapr,
874 target_ulong opcode, target_ulong *args)
876 qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n",
877 opcode, " (H_INVALIDATE_PID)");
878 return H_FUNCTION;
881 static void spapr_check_setup_free_hpt(SpaprMachineState *spapr,
882 uint64_t patbe_old, uint64_t patbe_new)
885 * We have 4 Options:
886 * HASH->HASH || RADIX->RADIX || NOTHING->RADIX : Do Nothing
887 * HASH->RADIX : Free HPT
888 * RADIX->HASH : Allocate HPT
889 * NOTHING->HASH : Allocate HPT
890 * Note: NOTHING implies the case where we said the guest could choose
891 * later and so assumed radix and now it's called H_REG_PROC_TBL
894 if ((patbe_old & PATE1_GR) == (patbe_new & PATE1_GR)) {
895 /* We assume RADIX, so this catches all the "Do Nothing" cases */
896 } else if (!(patbe_old & PATE1_GR)) {
897 /* HASH->RADIX : Free HPT */
898 spapr_free_hpt(spapr);
899 } else if (!(patbe_new & PATE1_GR)) {
900 /* RADIX->HASH || NOTHING->HASH : Allocate HPT */
901 spapr_setup_hpt(spapr);
903 return;
906 #define FLAGS_MASK 0x01FULL
907 #define FLAG_MODIFY 0x10
908 #define FLAG_REGISTER 0x08
909 #define FLAG_RADIX 0x04
910 #define FLAG_HASH_PROC_TBL 0x02
911 #define FLAG_GTSE 0x01
913 static target_ulong h_register_process_table(PowerPCCPU *cpu,
914 SpaprMachineState *spapr,
915 target_ulong opcode,
916 target_ulong *args)
918 target_ulong flags = args[0];
919 target_ulong proc_tbl = args[1];
920 target_ulong page_size = args[2];
921 target_ulong table_size = args[3];
922 target_ulong update_lpcr = 0;
923 uint64_t cproc;
925 if (flags & ~FLAGS_MASK) { /* Check no reserved bits are set */
926 return H_PARAMETER;
928 if (flags & FLAG_MODIFY) {
929 if (flags & FLAG_REGISTER) {
930 if (flags & FLAG_RADIX) { /* Register new RADIX process table */
931 if (proc_tbl & 0xfff || proc_tbl >> 60) {
932 return H_P2;
933 } else if (page_size) {
934 return H_P3;
935 } else if (table_size > 24) {
936 return H_P4;
938 cproc = PATE1_GR | proc_tbl | table_size;
939 } else { /* Register new HPT process table */
940 if (flags & FLAG_HASH_PROC_TBL) { /* Hash with Segment Tables */
941 /* TODO - Not Supported */
942 /* Technically caused by flag bits => H_PARAMETER */
943 return H_PARAMETER;
944 } else { /* Hash with SLB */
945 if (proc_tbl >> 38) {
946 return H_P2;
947 } else if (page_size & ~0x7) {
948 return H_P3;
949 } else if (table_size > 24) {
950 return H_P4;
953 cproc = (proc_tbl << 25) | page_size << 5 | table_size;
956 } else { /* Deregister current process table */
958 * Set to benign value: (current GR) | 0. This allows
959 * deregistration in KVM to succeed even if the radix bit
960 * in flags doesn't match the radix bit in the old PATE.
962 cproc = spapr->patb_entry & PATE1_GR;
964 } else { /* Maintain current registration */
965 if (!(flags & FLAG_RADIX) != !(spapr->patb_entry & PATE1_GR)) {
966 /* Technically caused by flag bits => H_PARAMETER */
967 return H_PARAMETER; /* Existing Process Table Mismatch */
969 cproc = spapr->patb_entry;
972 /* Check if we need to setup OR free the hpt */
973 spapr_check_setup_free_hpt(spapr, spapr->patb_entry, cproc);
975 spapr->patb_entry = cproc; /* Save new process table */
977 /* Update the UPRT, HR and GTSE bits in the LPCR for all cpus */
978 if (flags & FLAG_RADIX) /* Radix must use process tables, also set HR */
979 update_lpcr |= (LPCR_UPRT | LPCR_HR);
980 else if (flags & FLAG_HASH_PROC_TBL) /* Hash with process tables */
981 update_lpcr |= LPCR_UPRT;
982 if (flags & FLAG_GTSE) /* Guest translation shootdown enable */
983 update_lpcr |= LPCR_GTSE;
985 spapr_set_all_lpcrs(update_lpcr, LPCR_UPRT | LPCR_HR | LPCR_GTSE);
987 if (kvm_enabled()) {
988 return kvmppc_configure_v3_mmu(cpu, flags & FLAG_RADIX,
989 flags & FLAG_GTSE, cproc);
991 return H_SUCCESS;
994 #define H_SIGNAL_SYS_RESET_ALL -1
995 #define H_SIGNAL_SYS_RESET_ALLBUTSELF -2
997 static target_ulong h_signal_sys_reset(PowerPCCPU *cpu,
998 SpaprMachineState *spapr,
999 target_ulong opcode, target_ulong *args)
1001 target_long target = args[0];
1002 CPUState *cs;
1004 if (target < 0) {
1005 /* Broadcast */
1006 if (target < H_SIGNAL_SYS_RESET_ALLBUTSELF) {
1007 return H_PARAMETER;
1010 CPU_FOREACH(cs) {
1011 PowerPCCPU *c = POWERPC_CPU(cs);
1013 if (target == H_SIGNAL_SYS_RESET_ALLBUTSELF) {
1014 if (c == cpu) {
1015 continue;
1018 run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
1020 return H_SUCCESS;
1022 } else {
1023 /* Unicast */
1024 cs = CPU(spapr_find_cpu(target));
1025 if (cs) {
1026 run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
1027 return H_SUCCESS;
1029 return H_PARAMETER;
1033 /* Returns either a logical PVR or zero if none was found */
1034 static uint32_t cas_check_pvr(PowerPCCPU *cpu, uint32_t max_compat,
1035 target_ulong *addr, bool *raw_mode_supported)
1037 bool explicit_match = false; /* Matched the CPU's real PVR */
1038 uint32_t best_compat = 0;
1039 int i;
1042 * We scan the supplied table of PVRs looking for two things
1043 * 1. Is our real CPU PVR in the list?
1044 * 2. What's the "best" listed logical PVR
1046 for (i = 0; i < 512; ++i) {
1047 uint32_t pvr, pvr_mask;
1049 pvr_mask = ldl_be_phys(&address_space_memory, *addr);
1050 pvr = ldl_be_phys(&address_space_memory, *addr + 4);
1051 *addr += 8;
1053 if (~pvr_mask & pvr) {
1054 break; /* Terminator record */
1057 if ((cpu->env.spr[SPR_PVR] & pvr_mask) == (pvr & pvr_mask)) {
1058 explicit_match = true;
1059 } else {
1060 if (ppc_check_compat(cpu, pvr, best_compat, max_compat)) {
1061 best_compat = pvr;
1066 *raw_mode_supported = explicit_match;
1068 /* Parsing finished */
1069 trace_spapr_cas_pvr(cpu->compat_pvr, explicit_match, best_compat);
1071 return best_compat;
1074 static
1075 target_ulong do_client_architecture_support(PowerPCCPU *cpu,
1076 SpaprMachineState *spapr,
1077 target_ulong vec,
1078 target_ulong fdt_bufsize)
1080 target_ulong ov_table; /* Working address in data buffer */
1081 uint32_t cas_pvr;
1082 SpaprOptionVector *ov1_guest, *ov5_guest;
1083 bool guest_radix;
1084 bool raw_mode_supported = false;
1085 bool guest_xive;
1086 CPUState *cs;
1087 void *fdt;
1088 uint32_t max_compat = spapr->max_compat_pvr;
1090 /* CAS is supposed to be called early when only the boot vCPU is active. */
1091 CPU_FOREACH(cs) {
1092 if (cs == CPU(cpu)) {
1093 continue;
1095 if (!cs->halted) {
1096 warn_report("guest has multiple active vCPUs at CAS, which is not allowed");
1097 return H_MULTI_THREADS_ACTIVE;
1101 cas_pvr = cas_check_pvr(cpu, max_compat, &vec, &raw_mode_supported);
1102 if (!cas_pvr && (!raw_mode_supported || max_compat)) {
1104 * We couldn't find a suitable compatibility mode, and either
1105 * the guest doesn't support "raw" mode for this CPU, or "raw"
1106 * mode is disabled because a maximum compat mode is set.
1108 error_report("Couldn't negotiate a suitable PVR during CAS");
1109 return H_HARDWARE;
1112 /* Update CPUs */
1113 if (cpu->compat_pvr != cas_pvr) {
1114 Error *local_err = NULL;
1116 if (ppc_set_compat_all(cas_pvr, &local_err) < 0) {
1117 /* We fail to set compat mode (likely because running with KVM PR),
1118 * but maybe we can fallback to raw mode if the guest supports it.
1120 if (!raw_mode_supported) {
1121 error_report_err(local_err);
1122 return H_HARDWARE;
1124 error_free(local_err);
1128 /* For the future use: here @ov_table points to the first option vector */
1129 ov_table = vec;
1131 ov1_guest = spapr_ovec_parse_vector(ov_table, 1);
1132 if (!ov1_guest) {
1133 warn_report("guest didn't provide option vector 1");
1134 return H_PARAMETER;
1136 ov5_guest = spapr_ovec_parse_vector(ov_table, 5);
1137 if (!ov5_guest) {
1138 spapr_ovec_cleanup(ov1_guest);
1139 warn_report("guest didn't provide option vector 5");
1140 return H_PARAMETER;
1142 if (spapr_ovec_test(ov5_guest, OV5_MMU_BOTH)) {
1143 error_report("guest requested hash and radix MMU, which is invalid.");
1144 exit(EXIT_FAILURE);
1146 if (spapr_ovec_test(ov5_guest, OV5_XIVE_BOTH)) {
1147 error_report("guest requested an invalid interrupt mode");
1148 exit(EXIT_FAILURE);
1151 guest_radix = spapr_ovec_test(ov5_guest, OV5_MMU_RADIX_300);
1153 guest_xive = spapr_ovec_test(ov5_guest, OV5_XIVE_EXPLOIT);
1156 * HPT resizing is a bit of a special case, because when enabled
1157 * we assume an HPT guest will support it until it says it
1158 * doesn't, instead of assuming it won't support it until it says
1159 * it does. Strictly speaking that approach could break for
1160 * guests which don't make a CAS call, but those are so old we
1161 * don't care about them. Without that assumption we'd have to
1162 * make at least a temporary allocation of an HPT sized for max
1163 * memory, which could be impossibly difficult under KVM HV if
1164 * maxram is large.
1166 if (!guest_radix && !spapr_ovec_test(ov5_guest, OV5_HPT_RESIZE)) {
1167 int maxshift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1169 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_REQUIRED) {
1170 error_report(
1171 "h_client_architecture_support: Guest doesn't support HPT resizing, but resize-hpt=required");
1172 exit(1);
1175 if (spapr->htab_shift < maxshift) {
1176 /* Guest doesn't know about HPT resizing, so we
1177 * pre-emptively resize for the maximum permitted RAM. At
1178 * the point this is called, nothing should have been
1179 * entered into the existing HPT */
1180 spapr_reallocate_hpt(spapr, maxshift, &error_fatal);
1181 push_sregs_to_kvm_pr(spapr);
1185 /* NOTE: there are actually a number of ov5 bits where input from the
1186 * guest is always zero, and the platform/QEMU enables them independently
1187 * of guest input. To model these properly we'd want some sort of mask,
1188 * but since they only currently apply to memory migration as defined
1189 * by LoPAPR 1.1, 14.5.4.8, which QEMU doesn't implement, we don't need
1190 * to worry about this for now.
1193 /* full range of negotiated ov5 capabilities */
1194 spapr_ovec_intersect(spapr->ov5_cas, spapr->ov5, ov5_guest);
1195 spapr_ovec_cleanup(ov5_guest);
1197 spapr_check_mmu_mode(guest_radix);
1199 spapr->cas_pre_isa3_guest = !spapr_ovec_test(ov1_guest, OV1_PPC_3_00);
1200 spapr_ovec_cleanup(ov1_guest);
1203 * Check for NUMA affinity conditions now that we know which NUMA
1204 * affinity the guest will use.
1206 spapr_numa_associativity_check(spapr);
1209 * Ensure the guest asks for an interrupt mode we support;
1210 * otherwise terminate the boot.
1212 if (guest_xive) {
1213 if (!spapr->irq->xive) {
1214 error_report(
1215 "Guest requested unavailable interrupt mode (XIVE), try the ic-mode=xive or ic-mode=dual machine property");
1216 exit(EXIT_FAILURE);
1218 } else {
1219 if (!spapr->irq->xics) {
1220 error_report(
1221 "Guest requested unavailable interrupt mode (XICS), either don't set the ic-mode machine property or try ic-mode=xics or ic-mode=dual");
1222 exit(EXIT_FAILURE);
1226 spapr_irq_update_active_intc(spapr);
1229 * Process all pending hot-plug/unplug requests now. An updated full
1230 * rendered FDT will be returned to the guest.
1232 spapr_drc_reset_all(spapr);
1233 spapr_clear_pending_hotplug_events(spapr);
1236 * If spapr_machine_reset() did not set up a HPT but one is necessary
1237 * (because the guest isn't going to use radix) then set it up here.
1239 if ((spapr->patb_entry & PATE1_GR) && !guest_radix) {
1240 /* legacy hash or new hash: */
1241 spapr_setup_hpt(spapr);
1244 fdt = spapr_build_fdt(spapr, spapr->vof != NULL, fdt_bufsize);
1245 g_free(spapr->fdt_blob);
1246 spapr->fdt_size = fdt_totalsize(fdt);
1247 spapr->fdt_initial_size = spapr->fdt_size;
1248 spapr->fdt_blob = fdt;
1250 return H_SUCCESS;
1253 static target_ulong h_client_architecture_support(PowerPCCPU *cpu,
1254 SpaprMachineState *spapr,
1255 target_ulong opcode,
1256 target_ulong *args)
1258 target_ulong vec = ppc64_phys_to_real(args[0]);
1259 target_ulong fdt_buf = args[1];
1260 target_ulong fdt_bufsize = args[2];
1261 target_ulong ret;
1262 SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
1264 if (fdt_bufsize < sizeof(hdr)) {
1265 error_report("SLOF provided insufficient CAS buffer "
1266 TARGET_FMT_lu " (min: %zu)", fdt_bufsize, sizeof(hdr));
1267 exit(EXIT_FAILURE);
1270 fdt_bufsize -= sizeof(hdr);
1272 ret = do_client_architecture_support(cpu, spapr, vec, fdt_bufsize);
1273 if (ret == H_SUCCESS) {
1274 _FDT((fdt_pack(spapr->fdt_blob)));
1275 spapr->fdt_size = fdt_totalsize(spapr->fdt_blob);
1276 spapr->fdt_initial_size = spapr->fdt_size;
1278 cpu_physical_memory_write(fdt_buf, &hdr, sizeof(hdr));
1279 cpu_physical_memory_write(fdt_buf + sizeof(hdr), spapr->fdt_blob,
1280 spapr->fdt_size);
1281 trace_spapr_cas_continue(spapr->fdt_size + sizeof(hdr));
1284 return ret;
1287 target_ulong spapr_vof_client_architecture_support(MachineState *ms,
1288 CPUState *cs,
1289 target_ulong ovec_addr)
1291 SpaprMachineState *spapr = SPAPR_MACHINE(ms);
1293 target_ulong ret = do_client_architecture_support(POWERPC_CPU(cs), spapr,
1294 ovec_addr, FDT_MAX_SIZE);
1297 * This adds stdout and generates phandles for boottime and CAS FDTs.
1298 * It is alright to update the FDT here as do_client_architecture_support()
1299 * does not pack it.
1301 spapr_vof_client_dt_finalize(spapr, spapr->fdt_blob);
1303 return ret;
1306 static target_ulong h_get_cpu_characteristics(PowerPCCPU *cpu,
1307 SpaprMachineState *spapr,
1308 target_ulong opcode,
1309 target_ulong *args)
1311 uint64_t characteristics = H_CPU_CHAR_HON_BRANCH_HINTS &
1312 ~H_CPU_CHAR_THR_RECONF_TRIG;
1313 uint64_t behaviour = H_CPU_BEHAV_FAVOUR_SECURITY;
1314 uint8_t safe_cache = spapr_get_cap(spapr, SPAPR_CAP_CFPC);
1315 uint8_t safe_bounds_check = spapr_get_cap(spapr, SPAPR_CAP_SBBC);
1316 uint8_t safe_indirect_branch = spapr_get_cap(spapr, SPAPR_CAP_IBS);
1317 uint8_t count_cache_flush_assist = spapr_get_cap(spapr,
1318 SPAPR_CAP_CCF_ASSIST);
1320 switch (safe_cache) {
1321 case SPAPR_CAP_WORKAROUND:
1322 characteristics |= H_CPU_CHAR_L1D_FLUSH_ORI30;
1323 characteristics |= H_CPU_CHAR_L1D_FLUSH_TRIG2;
1324 characteristics |= H_CPU_CHAR_L1D_THREAD_PRIV;
1325 behaviour |= H_CPU_BEHAV_L1D_FLUSH_PR;
1326 break;
1327 case SPAPR_CAP_FIXED:
1328 behaviour |= H_CPU_BEHAV_NO_L1D_FLUSH_ENTRY;
1329 behaviour |= H_CPU_BEHAV_NO_L1D_FLUSH_UACCESS;
1330 break;
1331 default: /* broken */
1332 assert(safe_cache == SPAPR_CAP_BROKEN);
1333 behaviour |= H_CPU_BEHAV_L1D_FLUSH_PR;
1334 break;
1337 switch (safe_bounds_check) {
1338 case SPAPR_CAP_WORKAROUND:
1339 characteristics |= H_CPU_CHAR_SPEC_BAR_ORI31;
1340 behaviour |= H_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1341 break;
1342 case SPAPR_CAP_FIXED:
1343 break;
1344 default: /* broken */
1345 assert(safe_bounds_check == SPAPR_CAP_BROKEN);
1346 behaviour |= H_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1347 break;
1350 switch (safe_indirect_branch) {
1351 case SPAPR_CAP_FIXED_NA:
1352 break;
1353 case SPAPR_CAP_FIXED_CCD:
1354 characteristics |= H_CPU_CHAR_CACHE_COUNT_DIS;
1355 break;
1356 case SPAPR_CAP_FIXED_IBS:
1357 characteristics |= H_CPU_CHAR_BCCTRL_SERIALISED;
1358 break;
1359 case SPAPR_CAP_WORKAROUND:
1360 behaviour |= H_CPU_BEHAV_FLUSH_COUNT_CACHE;
1361 if (count_cache_flush_assist) {
1362 characteristics |= H_CPU_CHAR_BCCTR_FLUSH_ASSIST;
1364 break;
1365 default: /* broken */
1366 assert(safe_indirect_branch == SPAPR_CAP_BROKEN);
1367 break;
1370 args[0] = characteristics;
1371 args[1] = behaviour;
1372 return H_SUCCESS;
1375 static target_ulong h_update_dt(PowerPCCPU *cpu, SpaprMachineState *spapr,
1376 target_ulong opcode, target_ulong *args)
1378 target_ulong dt = ppc64_phys_to_real(args[0]);
1379 struct fdt_header hdr = { 0 };
1380 unsigned cb;
1381 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1382 void *fdt;
1384 cpu_physical_memory_read(dt, &hdr, sizeof(hdr));
1385 cb = fdt32_to_cpu(hdr.totalsize);
1387 if (!smc->update_dt_enabled) {
1388 return H_SUCCESS;
1391 /* Check that the fdt did not grow out of proportion */
1392 if (cb > spapr->fdt_initial_size * 2) {
1393 trace_spapr_update_dt_failed_size(spapr->fdt_initial_size, cb,
1394 fdt32_to_cpu(hdr.magic));
1395 return H_PARAMETER;
1398 fdt = g_malloc0(cb);
1399 cpu_physical_memory_read(dt, fdt, cb);
1401 /* Check the fdt consistency */
1402 if (fdt_check_full(fdt, cb)) {
1403 trace_spapr_update_dt_failed_check(spapr->fdt_initial_size, cb,
1404 fdt32_to_cpu(hdr.magic));
1405 return H_PARAMETER;
1408 g_free(spapr->fdt_blob);
1409 spapr->fdt_size = cb;
1410 spapr->fdt_blob = fdt;
1411 trace_spapr_update_dt(cb);
1413 return H_SUCCESS;
1416 static spapr_hcall_fn papr_hypercall_table[(MAX_HCALL_OPCODE / 4) + 1];
1417 static spapr_hcall_fn kvmppc_hypercall_table[KVMPPC_HCALL_MAX - KVMPPC_HCALL_BASE + 1];
1418 static spapr_hcall_fn svm_hypercall_table[(SVM_HCALL_MAX - SVM_HCALL_BASE) / 4 + 1];
1420 void spapr_register_hypercall(target_ulong opcode, spapr_hcall_fn fn)
1422 spapr_hcall_fn *slot;
1424 if (opcode <= MAX_HCALL_OPCODE) {
1425 assert((opcode & 0x3) == 0);
1427 slot = &papr_hypercall_table[opcode / 4];
1428 } else if (opcode >= SVM_HCALL_BASE && opcode <= SVM_HCALL_MAX) {
1429 /* we only have SVM-related hcall numbers assigned in multiples of 4 */
1430 assert((opcode & 0x3) == 0);
1432 slot = &svm_hypercall_table[(opcode - SVM_HCALL_BASE) / 4];
1433 } else {
1434 assert((opcode >= KVMPPC_HCALL_BASE) && (opcode <= KVMPPC_HCALL_MAX));
1436 slot = &kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE];
1439 assert(!(*slot));
1440 *slot = fn;
1443 target_ulong spapr_hypercall(PowerPCCPU *cpu, target_ulong opcode,
1444 target_ulong *args)
1446 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1448 if ((opcode <= MAX_HCALL_OPCODE)
1449 && ((opcode & 0x3) == 0)) {
1450 spapr_hcall_fn fn = papr_hypercall_table[opcode / 4];
1452 if (fn) {
1453 return fn(cpu, spapr, opcode, args);
1455 } else if ((opcode >= SVM_HCALL_BASE) &&
1456 (opcode <= SVM_HCALL_MAX)) {
1457 spapr_hcall_fn fn = svm_hypercall_table[(opcode - SVM_HCALL_BASE) / 4];
1459 if (fn) {
1460 return fn(cpu, spapr, opcode, args);
1462 } else if ((opcode >= KVMPPC_HCALL_BASE) &&
1463 (opcode <= KVMPPC_HCALL_MAX)) {
1464 spapr_hcall_fn fn = kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE];
1466 if (fn) {
1467 return fn(cpu, spapr, opcode, args);
1471 qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x" TARGET_FMT_lx "\n",
1472 opcode);
1473 return H_FUNCTION;
1476 #ifndef CONFIG_TCG
1477 static target_ulong h_softmmu(PowerPCCPU *cpu, SpaprMachineState *spapr,
1478 target_ulong opcode, target_ulong *args)
1480 g_assert_not_reached();
1483 static void hypercall_register_softmmu(void)
1485 /* hcall-pft */
1486 spapr_register_hypercall(H_ENTER, h_softmmu);
1487 spapr_register_hypercall(H_REMOVE, h_softmmu);
1488 spapr_register_hypercall(H_PROTECT, h_softmmu);
1489 spapr_register_hypercall(H_READ, h_softmmu);
1491 /* hcall-bulk */
1492 spapr_register_hypercall(H_BULK_REMOVE, h_softmmu);
1494 #else
1495 static void hypercall_register_softmmu(void)
1497 /* DO NOTHING */
1499 #endif
1501 /* TCG only */
1502 #define PRTS_MASK 0x1f
1504 static target_ulong h_set_ptbl(PowerPCCPU *cpu,
1505 SpaprMachineState *spapr,
1506 target_ulong opcode,
1507 target_ulong *args)
1509 target_ulong ptcr = args[0];
1511 if (!spapr_get_cap(spapr, SPAPR_CAP_NESTED_KVM_HV)) {
1512 return H_FUNCTION;
1515 if ((ptcr & PRTS_MASK) + 12 - 4 > 12) {
1516 return H_PARAMETER;
1519 spapr->nested_ptcr = ptcr; /* Save new partition table */
1521 return H_SUCCESS;
1524 static target_ulong h_tlb_invalidate(PowerPCCPU *cpu,
1525 SpaprMachineState *spapr,
1526 target_ulong opcode,
1527 target_ulong *args)
1530 * The spapr virtual hypervisor nested HV implementation retains no L2
1531 * translation state except for TLB. And the TLB is always invalidated
1532 * across L1<->L2 transitions, so nothing is required here.
1535 return H_SUCCESS;
1538 static target_ulong h_copy_tofrom_guest(PowerPCCPU *cpu,
1539 SpaprMachineState *spapr,
1540 target_ulong opcode,
1541 target_ulong *args)
1544 * This HCALL is not required, L1 KVM will take a slow path and walk the
1545 * page tables manually to do the data copy.
1547 return H_FUNCTION;
1551 * When this handler returns, the environment is switched to the L2 guest
1552 * and TCG begins running that. spapr_exit_nested() performs the switch from
1553 * L2 back to L1 and returns from the H_ENTER_NESTED hcall.
1555 static target_ulong h_enter_nested(PowerPCCPU *cpu,
1556 SpaprMachineState *spapr,
1557 target_ulong opcode,
1558 target_ulong *args)
1560 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
1561 CPUState *cs = CPU(cpu);
1562 CPUPPCState *env = &cpu->env;
1563 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1564 target_ulong hv_ptr = args[0];
1565 target_ulong regs_ptr = args[1];
1566 target_ulong hdec, now = cpu_ppc_load_tbl(env);
1567 target_ulong lpcr, lpcr_mask;
1568 struct kvmppc_hv_guest_state *hvstate;
1569 struct kvmppc_hv_guest_state hv_state;
1570 struct kvmppc_pt_regs *regs;
1571 hwaddr len;
1572 uint64_t cr;
1573 int i;
1575 if (spapr->nested_ptcr == 0) {
1576 return H_NOT_AVAILABLE;
1579 len = sizeof(*hvstate);
1580 hvstate = address_space_map(CPU(cpu)->as, hv_ptr, &len, false,
1581 MEMTXATTRS_UNSPECIFIED);
1582 if (len != sizeof(*hvstate)) {
1583 address_space_unmap(CPU(cpu)->as, hvstate, len, 0, false);
1584 return H_PARAMETER;
1587 memcpy(&hv_state, hvstate, len);
1589 address_space_unmap(CPU(cpu)->as, hvstate, len, len, false);
1592 * We accept versions 1 and 2. Version 2 fields are unused because TCG
1593 * does not implement DAWR*.
1595 if (hv_state.version > HV_GUEST_STATE_VERSION) {
1596 return H_PARAMETER;
1599 spapr_cpu->nested_host_state = g_try_malloc(sizeof(CPUPPCState));
1600 if (!spapr_cpu->nested_host_state) {
1601 return H_NO_MEM;
1604 memcpy(spapr_cpu->nested_host_state, env, sizeof(CPUPPCState));
1606 len = sizeof(*regs);
1607 regs = address_space_map(CPU(cpu)->as, regs_ptr, &len, false,
1608 MEMTXATTRS_UNSPECIFIED);
1609 if (!regs || len != sizeof(*regs)) {
1610 address_space_unmap(CPU(cpu)->as, regs, len, 0, false);
1611 g_free(spapr_cpu->nested_host_state);
1612 return H_P2;
1615 len = sizeof(env->gpr);
1616 assert(len == sizeof(regs->gpr));
1617 memcpy(env->gpr, regs->gpr, len);
1619 env->lr = regs->link;
1620 env->ctr = regs->ctr;
1621 cpu_write_xer(env, regs->xer);
1623 cr = regs->ccr;
1624 for (i = 7; i >= 0; i--) {
1625 env->crf[i] = cr & 15;
1626 cr >>= 4;
1629 env->msr = regs->msr;
1630 env->nip = regs->nip;
1632 address_space_unmap(CPU(cpu)->as, regs, len, len, false);
1634 env->cfar = hv_state.cfar;
1636 assert(env->spr[SPR_LPIDR] == 0);
1637 env->spr[SPR_LPIDR] = hv_state.lpid;
1639 lpcr_mask = LPCR_DPFD | LPCR_ILE | LPCR_AIL | LPCR_LD | LPCR_MER;
1640 lpcr = (env->spr[SPR_LPCR] & ~lpcr_mask) | (hv_state.lpcr & lpcr_mask);
1641 lpcr |= LPCR_HR | LPCR_UPRT | LPCR_GTSE | LPCR_HVICE | LPCR_HDICE;
1642 lpcr &= ~LPCR_LPES0;
1643 env->spr[SPR_LPCR] = lpcr & pcc->lpcr_mask;
1645 env->spr[SPR_PCR] = hv_state.pcr;
1646 /* hv_state.amor is not used */
1647 env->spr[SPR_DPDES] = hv_state.dpdes;
1648 env->spr[SPR_HFSCR] = hv_state.hfscr;
1649 hdec = hv_state.hdec_expiry - now;
1650 spapr_cpu->nested_tb_offset = hv_state.tb_offset;
1651 /* TCG does not implement DAWR*, CIABR, PURR, SPURR, IC, VTB, HEIR SPRs*/
1652 env->spr[SPR_SRR0] = hv_state.srr0;
1653 env->spr[SPR_SRR1] = hv_state.srr1;
1654 env->spr[SPR_SPRG0] = hv_state.sprg[0];
1655 env->spr[SPR_SPRG1] = hv_state.sprg[1];
1656 env->spr[SPR_SPRG2] = hv_state.sprg[2];
1657 env->spr[SPR_SPRG3] = hv_state.sprg[3];
1658 env->spr[SPR_BOOKS_PID] = hv_state.pidr;
1659 env->spr[SPR_PPR] = hv_state.ppr;
1661 cpu_ppc_hdecr_init(env);
1662 cpu_ppc_store_hdecr(env, hdec);
1665 * The hv_state.vcpu_token is not needed. It is used by the KVM
1666 * implementation to remember which L2 vCPU last ran on which physical
1667 * CPU so as to invalidate process scope translations if it is moved
1668 * between physical CPUs. For now TLBs are always flushed on L1<->L2
1669 * transitions so this is not a problem.
1671 * Could validate that the same vcpu_token does not attempt to run on
1672 * different L1 vCPUs at the same time, but that would be a L1 KVM bug
1673 * and it's not obviously worth a new data structure to do it.
1676 env->tb_env->tb_offset += spapr_cpu->nested_tb_offset;
1677 spapr_cpu->in_nested = true;
1679 hreg_compute_hflags(env);
1680 tlb_flush(cs);
1681 env->reserve_addr = -1; /* Reset the reservation */
1684 * The spapr hcall helper sets env->gpr[3] to the return value, but at
1685 * this point the L1 is not returning from the hcall but rather we
1686 * start running the L2, so r3 must not be clobbered, so return env->gpr[3]
1687 * to leave it unchanged.
1689 return env->gpr[3];
1692 void spapr_exit_nested(PowerPCCPU *cpu, int excp)
1694 CPUState *cs = CPU(cpu);
1695 CPUPPCState *env = &cpu->env;
1696 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1697 target_ulong r3_return = env->excp_vectors[excp]; /* hcall return value */
1698 target_ulong hv_ptr = spapr_cpu->nested_host_state->gpr[4];
1699 target_ulong regs_ptr = spapr_cpu->nested_host_state->gpr[5];
1700 struct kvmppc_hv_guest_state *hvstate;
1701 struct kvmppc_pt_regs *regs;
1702 hwaddr len;
1703 uint64_t cr;
1704 int i;
1706 assert(spapr_cpu->in_nested);
1708 cpu_ppc_hdecr_exit(env);
1710 len = sizeof(*hvstate);
1711 hvstate = address_space_map(CPU(cpu)->as, hv_ptr, &len, true,
1712 MEMTXATTRS_UNSPECIFIED);
1713 if (len != sizeof(*hvstate)) {
1714 address_space_unmap(CPU(cpu)->as, hvstate, len, 0, true);
1715 r3_return = H_PARAMETER;
1716 goto out_restore_l1;
1719 hvstate->cfar = env->cfar;
1720 hvstate->lpcr = env->spr[SPR_LPCR];
1721 hvstate->pcr = env->spr[SPR_PCR];
1722 hvstate->dpdes = env->spr[SPR_DPDES];
1723 hvstate->hfscr = env->spr[SPR_HFSCR];
1725 if (excp == POWERPC_EXCP_HDSI) {
1726 hvstate->hdar = env->spr[SPR_HDAR];
1727 hvstate->hdsisr = env->spr[SPR_HDSISR];
1728 hvstate->asdr = env->spr[SPR_ASDR];
1729 } else if (excp == POWERPC_EXCP_HISI) {
1730 hvstate->asdr = env->spr[SPR_ASDR];
1733 /* HEIR should be implemented for HV mode and saved here. */
1734 hvstate->srr0 = env->spr[SPR_SRR0];
1735 hvstate->srr1 = env->spr[SPR_SRR1];
1736 hvstate->sprg[0] = env->spr[SPR_SPRG0];
1737 hvstate->sprg[1] = env->spr[SPR_SPRG1];
1738 hvstate->sprg[2] = env->spr[SPR_SPRG2];
1739 hvstate->sprg[3] = env->spr[SPR_SPRG3];
1740 hvstate->pidr = env->spr[SPR_BOOKS_PID];
1741 hvstate->ppr = env->spr[SPR_PPR];
1743 /* Is it okay to specify write length larger than actual data written? */
1744 address_space_unmap(CPU(cpu)->as, hvstate, len, len, true);
1746 len = sizeof(*regs);
1747 regs = address_space_map(CPU(cpu)->as, regs_ptr, &len, true,
1748 MEMTXATTRS_UNSPECIFIED);
1749 if (!regs || len != sizeof(*regs)) {
1750 address_space_unmap(CPU(cpu)->as, regs, len, 0, true);
1751 r3_return = H_P2;
1752 goto out_restore_l1;
1755 len = sizeof(env->gpr);
1756 assert(len == sizeof(regs->gpr));
1757 memcpy(regs->gpr, env->gpr, len);
1759 regs->link = env->lr;
1760 regs->ctr = env->ctr;
1761 regs->xer = cpu_read_xer(env);
1763 cr = 0;
1764 for (i = 0; i < 8; i++) {
1765 cr |= (env->crf[i] & 15) << (4 * (7 - i));
1767 regs->ccr = cr;
1769 if (excp == POWERPC_EXCP_MCHECK ||
1770 excp == POWERPC_EXCP_RESET ||
1771 excp == POWERPC_EXCP_SYSCALL) {
1772 regs->nip = env->spr[SPR_SRR0];
1773 regs->msr = env->spr[SPR_SRR1] & env->msr_mask;
1774 } else {
1775 regs->nip = env->spr[SPR_HSRR0];
1776 regs->msr = env->spr[SPR_HSRR1] & env->msr_mask;
1779 /* Is it okay to specify write length larger than actual data written? */
1780 address_space_unmap(CPU(cpu)->as, regs, len, len, true);
1782 out_restore_l1:
1783 memcpy(env->gpr, spapr_cpu->nested_host_state->gpr, sizeof(env->gpr));
1784 env->lr = spapr_cpu->nested_host_state->lr;
1785 env->ctr = spapr_cpu->nested_host_state->ctr;
1786 memcpy(env->crf, spapr_cpu->nested_host_state->crf, sizeof(env->crf));
1787 env->cfar = spapr_cpu->nested_host_state->cfar;
1788 env->xer = spapr_cpu->nested_host_state->xer;
1789 env->so = spapr_cpu->nested_host_state->so;
1790 env->ov = spapr_cpu->nested_host_state->ov;
1791 env->ov32 = spapr_cpu->nested_host_state->ov32;
1792 env->ca32 = spapr_cpu->nested_host_state->ca32;
1793 env->msr = spapr_cpu->nested_host_state->msr;
1794 env->nip = spapr_cpu->nested_host_state->nip;
1796 assert(env->spr[SPR_LPIDR] != 0);
1797 env->spr[SPR_LPCR] = spapr_cpu->nested_host_state->spr[SPR_LPCR];
1798 env->spr[SPR_LPIDR] = spapr_cpu->nested_host_state->spr[SPR_LPIDR];
1799 env->spr[SPR_PCR] = spapr_cpu->nested_host_state->spr[SPR_PCR];
1800 env->spr[SPR_DPDES] = 0;
1801 env->spr[SPR_HFSCR] = spapr_cpu->nested_host_state->spr[SPR_HFSCR];
1802 env->spr[SPR_SRR0] = spapr_cpu->nested_host_state->spr[SPR_SRR0];
1803 env->spr[SPR_SRR1] = spapr_cpu->nested_host_state->spr[SPR_SRR1];
1804 env->spr[SPR_SPRG0] = spapr_cpu->nested_host_state->spr[SPR_SPRG0];
1805 env->spr[SPR_SPRG1] = spapr_cpu->nested_host_state->spr[SPR_SPRG1];
1806 env->spr[SPR_SPRG2] = spapr_cpu->nested_host_state->spr[SPR_SPRG2];
1807 env->spr[SPR_SPRG3] = spapr_cpu->nested_host_state->spr[SPR_SPRG3];
1808 env->spr[SPR_BOOKS_PID] = spapr_cpu->nested_host_state->spr[SPR_BOOKS_PID];
1809 env->spr[SPR_PPR] = spapr_cpu->nested_host_state->spr[SPR_PPR];
1812 * Return the interrupt vector address from H_ENTER_NESTED to the L1
1813 * (or error code).
1815 env->gpr[3] = r3_return;
1817 env->tb_env->tb_offset -= spapr_cpu->nested_tb_offset;
1818 spapr_cpu->in_nested = false;
1820 hreg_compute_hflags(env);
1821 tlb_flush(cs);
1822 env->reserve_addr = -1; /* Reset the reservation */
1824 g_free(spapr_cpu->nested_host_state);
1825 spapr_cpu->nested_host_state = NULL;
1828 static void hypercall_register_types(void)
1830 hypercall_register_softmmu();
1832 /* hcall-hpt-resize */
1833 spapr_register_hypercall(H_RESIZE_HPT_PREPARE, h_resize_hpt_prepare);
1834 spapr_register_hypercall(H_RESIZE_HPT_COMMIT, h_resize_hpt_commit);
1836 /* hcall-splpar */
1837 spapr_register_hypercall(H_REGISTER_VPA, h_register_vpa);
1838 spapr_register_hypercall(H_CEDE, h_cede);
1839 spapr_register_hypercall(H_CONFER, h_confer);
1840 spapr_register_hypercall(H_PROD, h_prod);
1842 /* hcall-join */
1843 spapr_register_hypercall(H_JOIN, h_join);
1845 spapr_register_hypercall(H_SIGNAL_SYS_RESET, h_signal_sys_reset);
1847 /* processor register resource access h-calls */
1848 spapr_register_hypercall(H_SET_SPRG0, h_set_sprg0);
1849 spapr_register_hypercall(H_SET_DABR, h_set_dabr);
1850 spapr_register_hypercall(H_SET_XDABR, h_set_xdabr);
1851 spapr_register_hypercall(H_PAGE_INIT, h_page_init);
1852 spapr_register_hypercall(H_SET_MODE, h_set_mode);
1854 /* In Memory Table MMU h-calls */
1855 spapr_register_hypercall(H_CLEAN_SLB, h_clean_slb);
1856 spapr_register_hypercall(H_INVALIDATE_PID, h_invalidate_pid);
1857 spapr_register_hypercall(H_REGISTER_PROC_TBL, h_register_process_table);
1859 /* hcall-get-cpu-characteristics */
1860 spapr_register_hypercall(H_GET_CPU_CHARACTERISTICS,
1861 h_get_cpu_characteristics);
1863 /* "debugger" hcalls (also used by SLOF). Note: We do -not- differenciate
1864 * here between the "CI" and the "CACHE" variants, they will use whatever
1865 * mapping attributes qemu is using. When using KVM, the kernel will
1866 * enforce the attributes more strongly
1868 spapr_register_hypercall(H_LOGICAL_CI_LOAD, h_logical_load);
1869 spapr_register_hypercall(H_LOGICAL_CI_STORE, h_logical_store);
1870 spapr_register_hypercall(H_LOGICAL_CACHE_LOAD, h_logical_load);
1871 spapr_register_hypercall(H_LOGICAL_CACHE_STORE, h_logical_store);
1872 spapr_register_hypercall(H_LOGICAL_ICBI, h_logical_icbi);
1873 spapr_register_hypercall(H_LOGICAL_DCBF, h_logical_dcbf);
1874 spapr_register_hypercall(KVMPPC_H_LOGICAL_MEMOP, h_logical_memop);
1876 /* qemu/KVM-PPC specific hcalls */
1877 spapr_register_hypercall(KVMPPC_H_RTAS, h_rtas);
1879 /* ibm,client-architecture-support support */
1880 spapr_register_hypercall(KVMPPC_H_CAS, h_client_architecture_support);
1882 spapr_register_hypercall(KVMPPC_H_UPDATE_DT, h_update_dt);
1884 spapr_register_hypercall(KVMPPC_H_SET_PARTITION_TABLE, h_set_ptbl);
1885 spapr_register_hypercall(KVMPPC_H_ENTER_NESTED, h_enter_nested);
1886 spapr_register_hypercall(KVMPPC_H_TLB_INVALIDATE, h_tlb_invalidate);
1887 spapr_register_hypercall(KVMPPC_H_COPY_TOFROM_GUEST, h_copy_tofrom_guest);
1890 type_init(hypercall_register_types)