xref: /qemu/hw/ppc/spapr.c (revision c894bdf78b32ed825903f4f05c765eecf2e4f1cc)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  * Copyright (c) 2010-2024, IBM Corporation..
8  *
9  * SPDX-License-Identifier: GPL-2.0-or-later
10  *
11  * Permission is hereby granted, free of charge, to any person obtaining a copy
12  * of this software and associated documentation files (the "Software"), to deal
13  * in the Software without restriction, including without limitation the rights
14  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
15  * copies of the Software, and to permit persons to whom the Software is
16  * furnished to do so, subject to the following conditions:
17  *
18  * The above copyright notice and this permission notice shall be included in
19  * all copies or substantial portions of the Software.
20  *
21  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
22  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
23  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
24  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
25  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
26  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27  * THE SOFTWARE.
28  */
29 
30 #include "qemu/osdep.h"
31 #include "qemu/datadir.h"
32 #include "qemu/memalign.h"
33 #include "qemu/guest-random.h"
34 #include "qapi/error.h"
35 #include "qapi/qapi-events-machine.h"
36 #include "qapi/qapi-events-qdev.h"
37 #include "qapi/visitor.h"
38 #include "system/system.h"
39 #include "system/hostmem.h"
40 #include "system/numa.h"
41 #include "system/tcg.h"
42 #include "system/qtest.h"
43 #include "system/reset.h"
44 #include "system/runstate.h"
45 #include "qemu/log.h"
46 #include "hw/fw-path-provider.h"
47 #include "elf.h"
48 #include "net/net.h"
49 #include "system/device_tree.h"
50 #include "system/cpus.h"
51 #include "system/hw_accel.h"
52 #include "kvm_ppc.h"
53 #include "migration/misc.h"
54 #include "migration/qemu-file-types.h"
55 #include "migration/global_state.h"
56 #include "migration/register.h"
57 #include "migration/blocker.h"
58 #include "mmu-hash64.h"
59 #include "mmu-book3s-v3.h"
60 #include "cpu-models.h"
61 #include "hw/core/cpu.h"
62 
63 #include "hw/ppc/ppc.h"
64 #include "hw/loader.h"
65 
66 #include "hw/ppc/fdt.h"
67 #include "hw/ppc/spapr.h"
68 #include "hw/ppc/spapr_nested.h"
69 #include "hw/ppc/spapr_vio.h"
70 #include "hw/ppc/vof.h"
71 #include "hw/qdev-properties.h"
72 #include "hw/pci-host/spapr.h"
73 #include "hw/pci/msi.h"
74 
75 #include "hw/pci/pci.h"
76 #include "hw/scsi/scsi.h"
77 #include "hw/virtio/virtio-scsi.h"
78 #include "hw/virtio/vhost-scsi-common.h"
79 
80 #include "exec/ram_addr.h"
81 #include "system/confidential-guest-support.h"
82 #include "hw/usb.h"
83 #include "qemu/config-file.h"
84 #include "qemu/error-report.h"
85 #include "trace.h"
86 #include "hw/nmi.h"
87 #include "hw/intc/intc.h"
88 
89 #include "hw/ppc/spapr_cpu_core.h"
90 #include "hw/mem/memory-device.h"
91 #include "hw/ppc/spapr_tpm_proxy.h"
92 #include "hw/ppc/spapr_nvdimm.h"
93 #include "hw/ppc/spapr_numa.h"
94 
95 #include <libfdt.h>
96 
97 /* SLOF memory layout:
98  *
99  * SLOF raw image loaded at 0, copies its romfs right below the flat
100  * device-tree, then position SLOF itself 31M below that
101  *
102  * So we set FW_OVERHEAD to 40MB which should account for all of that
103  * and more
104  *
105  * We load our kernel at 4M, leaving space for SLOF initial image
106  */
107 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
108 #define FW_MAX_SIZE             0x400000
109 #define FW_FILE_NAME            "slof.bin"
110 #define FW_FILE_NAME_VOF        "vof.bin"
111 #define FW_OVERHEAD             0x2800000
112 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
113 
114 #define MIN_RMA_SLOF            (128 * MiB)
115 
116 #define PHANDLE_INTC            0x00001111
117 
118 /* These two functions implement the VCPU id numbering: one to compute them
119  * all and one to identify thread 0 of a VCORE. Any change to the first one
120  * is likely to have an impact on the second one, so let's keep them close.
121  */
122 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
123 {
124     MachineState *ms = MACHINE(spapr);
125     unsigned int smp_threads = ms->smp.threads;
126 
127     assert(spapr->vsmt);
128     return
129         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
130 }
131 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
132                                       PowerPCCPU *cpu)
133 {
134     assert(spapr->vsmt);
135     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
136 }
137 
138 int spapr_max_server_number(SpaprMachineState *spapr)
139 {
140     MachineState *ms = MACHINE(spapr);
141 
142     assert(spapr->vsmt);
143     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
144 }
145 
146 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
147                                   int smt_threads)
148 {
149     int i, ret = 0;
150     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
151     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
152     int index = spapr_get_vcpu_id(cpu);
153 
154     if (cpu->compat_pvr) {
155         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
156         if (ret < 0) {
157             return ret;
158         }
159     }
160 
161     /* Build interrupt servers and gservers properties */
162     for (i = 0; i < smt_threads; i++) {
163         servers_prop[i] = cpu_to_be32(index + i);
164         /* Hack, direct the group queues back to cpu 0 */
165         gservers_prop[i*2] = cpu_to_be32(index + i);
166         gservers_prop[i*2 + 1] = 0;
167     }
168     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
169                       servers_prop, sizeof(*servers_prop) * smt_threads);
170     if (ret < 0) {
171         return ret;
172     }
173     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
174                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
175 
176     return ret;
177 }
178 
179 static void spapr_dt_pa_features(SpaprMachineState *spapr,
180                                  PowerPCCPU *cpu,
181                                  void *fdt, int offset)
182 {
183     /*
184      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
185      * but not MTTCG, so disable it. To advertise it, a cap would have
186      * to be added, or support implemented for MTTCG.
187      *
188      * Copy/paste is not supported by TCG, so it is not advertised. KVM
189      * can execute them but it has no accelerator drivers which are usable,
190      * so there isn't much need for it anyway.
191      */
192 
193     /* These should be kept in sync with pnv */
194     uint8_t pa_features_206[] = { 6, 0,
195         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
196     uint8_t pa_features_207[] = { 24, 0,
197         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
198         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
199         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
200         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
201     uint8_t pa_features_300[] = { 66, 0,
202         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
203         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
204         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
205         /* 6: DS207 */
206         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
207         /* 16: Vector */
208         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
209         /* 18: Vec. Scalar, 20: Vec. XOR */
210         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
211         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
212         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
213         /* 32: LE atomic, 34: EBB + ext EBB */
214         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
215         /* 40: Radix MMU */
216         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
217         /* 42: PM, 44: PC RA, 46: SC vec'd */
218         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
219         /* 48: SIMD, 50: QP BFP, 52: String */
220         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
221         /* 54: DecFP, 56: DecI, 58: SHA */
222         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
223         /* 60: NM atomic, 62: RNG */
224         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
225     };
226     /* 3.1 removes SAO, HTM support */
227     uint8_t pa_features_31[] = { 74, 0,
228         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
229         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
230         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
231         /* 6: DS207 */
232         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
233         /* 16: Vector */
234         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
235         /* 18: Vec. Scalar, 20: Vec. XOR */
236         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
237         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
239         /* 32: LE atomic, 34: EBB + ext EBB */
240         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
241         /* 40: Radix MMU */
242         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
243         /* 42: PM, 44: PC RA, 46: SC vec'd */
244         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
245         /* 48: SIMD, 50: QP BFP, 52: String */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
247         /* 54: DecFP, 56: DecI, 58: SHA */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
249         /* 60: NM atomic, 62: RNG */
250         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
251         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
252         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
253         /* 72: [P]HASHST/[P]HASHCHK */
254         0x80, 0x00,                         /* 72 - 73 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
272         pa_features = pa_features_31;
273         pa_size = sizeof(pa_features_31);
274     }
275     if (!pa_features) {
276         return;
277     }
278 
279     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
280         /*
281          * Note: we keep CI large pages off by default because a 64K capable
282          * guest provisioned with large pages might otherwise try to map a qemu
283          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
284          * even if that qemu runs on a 4k host.
285          * We dd this bit back here if we are confident this is not an issue
286          */
287         pa_features[3] |= 0x20;
288     }
289     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
290         pa_features[24] |= 0x80;    /* Transactional memory support */
291     }
292     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
293         /* Workaround for broken kernels that attempt (guest) radix
294          * mode when they can't handle it, if they see the radix bit set
295          * in pa-features. So hide it from them. */
296         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
297     }
298 
299     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
300 }
301 
302 static void spapr_dt_pi_features(SpaprMachineState *spapr,
303                                  PowerPCCPU *cpu,
304                                  void *fdt, int offset)
305 {
306     uint8_t pi_features[] = { 1, 0,
307         0x00 };
308 
309     if (kvm_enabled() && ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00,
310                                           0, cpu->compat_pvr)) {
311         /*
312          * POWER9 and later CPUs with KVM run in LPAR-per-thread mode where
313          * all threads are essentially independent CPUs, and msgsndp does not
314          * work (because it is physically-addressed) and therefore is
315          * emulated by KVM, so disable it here to ensure XIVE will be used.
316          * This is both KVM and CPU implementation-specific behaviour so a KVM
317          * cap would be cleanest, but for now this works. If KVM ever permits
318          * native msgsndp execution by guests, a cap could be added at that
319          * time.
320          */
321         pi_features[2] |= 0x08; /* 4: No msgsndp */
322     }
323 
324     _FDT((fdt_setprop(fdt, offset, "ibm,pi-features", pi_features,
325                       sizeof(pi_features))));
326 }
327 
328 static hwaddr spapr_node0_size(MachineState *machine)
329 {
330     if (machine->numa_state->num_nodes) {
331         int i;
332         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
333             if (machine->numa_state->nodes[i].node_mem) {
334                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
335                            machine->ram_size);
336             }
337         }
338     }
339     return machine->ram_size;
340 }
341 
342 static void add_str(GString *s, const gchar *s1)
343 {
344     g_string_append_len(s, s1, strlen(s1) + 1);
345 }
346 
347 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
348                                 hwaddr start, hwaddr size)
349 {
350     char mem_name[32];
351     uint64_t mem_reg_property[2];
352     int off;
353 
354     mem_reg_property[0] = cpu_to_be64(start);
355     mem_reg_property[1] = cpu_to_be64(size);
356 
357     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
358     off = fdt_add_subnode(fdt, 0, mem_name);
359     _FDT(off);
360     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
361     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
362                       sizeof(mem_reg_property))));
363     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
364     return off;
365 }
366 
367 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
368 {
369     MemoryDeviceInfoList *info;
370 
371     for (info = list; info; info = info->next) {
372         MemoryDeviceInfo *value = info->value;
373 
374         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
375             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
376 
377             if (addr >= pcdimm_info->addr &&
378                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
379                 return pcdimm_info->node;
380             }
381         }
382     }
383 
384     return -1;
385 }
386 
387 struct sPAPRDrconfCellV2 {
388      uint32_t seq_lmbs;
389      uint64_t base_addr;
390      uint32_t drc_index;
391      uint32_t aa_index;
392      uint32_t flags;
393 } QEMU_PACKED;
394 
395 typedef struct DrconfCellQueue {
396     struct sPAPRDrconfCellV2 cell;
397     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
398 } DrconfCellQueue;
399 
400 static DrconfCellQueue *
401 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
402                       uint32_t drc_index, uint32_t aa_index,
403                       uint32_t flags)
404 {
405     DrconfCellQueue *elem;
406 
407     elem = g_malloc0(sizeof(*elem));
408     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
409     elem->cell.base_addr = cpu_to_be64(base_addr);
410     elem->cell.drc_index = cpu_to_be32(drc_index);
411     elem->cell.aa_index = cpu_to_be32(aa_index);
412     elem->cell.flags = cpu_to_be32(flags);
413 
414     return elem;
415 }
416 
417 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
418                                       int offset, MemoryDeviceInfoList *dimms)
419 {
420     MachineState *machine = MACHINE(spapr);
421     uint8_t *int_buf, *cur_index;
422     int ret;
423     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
424     uint64_t addr, cur_addr, size;
425     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
426     uint64_t mem_end = machine->device_memory->base +
427                        memory_region_size(&machine->device_memory->mr);
428     uint32_t node, buf_len, nr_entries = 0;
429     SpaprDrc *drc;
430     DrconfCellQueue *elem, *next;
431     MemoryDeviceInfoList *info;
432     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
433         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
434 
435     /* Entry to cover RAM and the gap area */
436     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
437                                  SPAPR_LMB_FLAGS_RESERVED |
438                                  SPAPR_LMB_FLAGS_DRC_INVALID);
439     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440     nr_entries++;
441 
442     cur_addr = machine->device_memory->base;
443     for (info = dimms; info; info = info->next) {
444         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
445 
446         addr = di->addr;
447         size = di->size;
448         node = di->node;
449 
450         /*
451          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
452          * area is marked hotpluggable in the next iteration for the bigger
453          * chunk including the NVDIMM occupied area.
454          */
455         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
456             continue;
457 
458         /* Entry for hot-pluggable area */
459         if (cur_addr < addr) {
460             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
461             g_assert(drc);
462             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
463                                          cur_addr, spapr_drc_index(drc), -1, 0);
464             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
465             nr_entries++;
466         }
467 
468         /* Entry for DIMM */
469         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
470         g_assert(drc);
471         elem = spapr_get_drconf_cell(size / lmb_size, addr,
472                                      spapr_drc_index(drc), node,
473                                      (SPAPR_LMB_FLAGS_ASSIGNED |
474                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
475         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
476         nr_entries++;
477         cur_addr = addr + size;
478     }
479 
480     /* Entry for remaining hotpluggable area */
481     if (cur_addr < mem_end) {
482         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
483         g_assert(drc);
484         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
485                                      cur_addr, spapr_drc_index(drc), -1, 0);
486         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
487         nr_entries++;
488     }
489 
490     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
491     int_buf = cur_index = g_malloc0(buf_len);
492     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
493     cur_index += sizeof(nr_entries);
494 
495     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
496         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
497         cur_index += sizeof(elem->cell);
498         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
499         g_free(elem);
500     }
501 
502     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
503     g_free(int_buf);
504     if (ret < 0) {
505         return -1;
506     }
507     return 0;
508 }
509 
510 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
511                                    int offset, MemoryDeviceInfoList *dimms)
512 {
513     MachineState *machine = MACHINE(spapr);
514     int i, ret;
515     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
516     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
517     uint32_t nr_lmbs = (machine->device_memory->base +
518                        memory_region_size(&machine->device_memory->mr)) /
519                        lmb_size;
520     uint32_t *int_buf, *cur_index, buf_len;
521 
522     /*
523      * Allocate enough buffer size to fit in ibm,dynamic-memory
524      */
525     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
526     cur_index = int_buf = g_malloc0(buf_len);
527     int_buf[0] = cpu_to_be32(nr_lmbs);
528     cur_index++;
529     for (i = 0; i < nr_lmbs; i++) {
530         uint64_t addr = i * lmb_size;
531         uint32_t *dynamic_memory = cur_index;
532 
533         if (i >= device_lmb_start) {
534             SpaprDrc *drc;
535 
536             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
537             g_assert(drc);
538 
539             dynamic_memory[0] = cpu_to_be32(addr >> 32);
540             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
541             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
542             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
543             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
544             if (memory_region_present(get_system_memory(), addr)) {
545                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
546             } else {
547                 dynamic_memory[5] = cpu_to_be32(0);
548             }
549         } else {
550             /*
551              * LMB information for RMA, boot time RAM and gap b/n RAM and
552              * device memory region -- all these are marked as reserved
553              * and as having no valid DRC.
554              */
555             dynamic_memory[0] = cpu_to_be32(addr >> 32);
556             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
557             dynamic_memory[2] = cpu_to_be32(0);
558             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
559             dynamic_memory[4] = cpu_to_be32(-1);
560             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
561                                             SPAPR_LMB_FLAGS_DRC_INVALID);
562         }
563 
564         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
565     }
566     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
567     g_free(int_buf);
568     if (ret < 0) {
569         return -1;
570     }
571     return 0;
572 }
573 
574 /*
575  * Adds ibm,dynamic-reconfiguration-memory node.
576  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
577  * of this device tree node.
578  */
579 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
580                                                    void *fdt)
581 {
582     MachineState *machine = MACHINE(spapr);
583     int ret, offset;
584     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
585     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
586                                 cpu_to_be32(lmb_size & 0xffffffff)};
587     MemoryDeviceInfoList *dimms = NULL;
588 
589     /* Don't create the node if there is no device memory. */
590     if (!machine->device_memory) {
591         return 0;
592     }
593 
594     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
595 
596     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
597                     sizeof(prop_lmb_size));
598     if (ret < 0) {
599         return ret;
600     }
601 
602     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
603     if (ret < 0) {
604         return ret;
605     }
606 
607     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
608     if (ret < 0) {
609         return ret;
610     }
611 
612     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
613     dimms = qmp_memory_device_list();
614     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
615         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
616     } else {
617         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
618     }
619     qapi_free_MemoryDeviceInfoList(dimms);
620 
621     if (ret < 0) {
622         return ret;
623     }
624 
625     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
626 
627     return ret;
628 }
629 
630 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
631 {
632     MachineState *machine = MACHINE(spapr);
633     hwaddr mem_start, node_size;
634     int i, nb_nodes = machine->numa_state->num_nodes;
635     NodeInfo *nodes = machine->numa_state->nodes;
636 
637     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
638         if (!nodes[i].node_mem) {
639             continue;
640         }
641         if (mem_start >= machine->ram_size) {
642             node_size = 0;
643         } else {
644             node_size = nodes[i].node_mem;
645             if (node_size > machine->ram_size - mem_start) {
646                 node_size = machine->ram_size - mem_start;
647             }
648         }
649         if (!mem_start) {
650             /* spapr_machine_init() checks for rma_size <= node0_size
651              * already */
652             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
653             mem_start += spapr->rma_size;
654             node_size -= spapr->rma_size;
655         }
656         for ( ; node_size; ) {
657             hwaddr sizetmp = pow2floor(node_size);
658 
659             /* mem_start != 0 here */
660             if (ctzl(mem_start) < ctzl(sizetmp)) {
661                 sizetmp = 1ULL << ctzl(mem_start);
662             }
663 
664             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
665             node_size -= sizetmp;
666             mem_start += sizetmp;
667         }
668     }
669 
670     /* Generate ibm,dynamic-reconfiguration-memory node if required */
671     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
672         int ret;
673 
674         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
675         if (ret) {
676             return ret;
677         }
678     }
679 
680     return 0;
681 }
682 
683 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
684                          SpaprMachineState *spapr)
685 {
686     MachineState *ms = MACHINE(spapr);
687     PowerPCCPU *cpu = POWERPC_CPU(cs);
688     CPUPPCState *env = &cpu->env;
689     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
690     int index = spapr_get_vcpu_id(cpu);
691     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
692                        0xffffffff, 0xffffffff};
693     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
694         : SPAPR_TIMEBASE_FREQ;
695     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
696     uint32_t page_sizes_prop[64];
697     size_t page_sizes_prop_size;
698     unsigned int smp_threads = ms->smp.threads;
699     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
700     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
701     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
702     SpaprDrc *drc;
703     int drc_index;
704     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
705     int i;
706 
707     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, env->core_index);
708     if (drc) {
709         drc_index = spapr_drc_index(drc);
710         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
711     }
712 
713     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
714     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
715 
716     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
717     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
718                            env->dcache_line_size)));
719     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
720                            env->dcache_line_size)));
721     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
722                            env->icache_line_size)));
723     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
724                            env->icache_line_size)));
725 
726     if (pcc->l1_dcache_size) {
727         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
728                                pcc->l1_dcache_size)));
729     } else {
730         warn_report("Unknown L1 dcache size for cpu");
731     }
732     if (pcc->l1_icache_size) {
733         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
734                                pcc->l1_icache_size)));
735     } else {
736         warn_report("Unknown L1 icache size for cpu");
737     }
738 
739     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
740     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
741     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
742     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
743     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
744     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
745 
746     if (ppc_has_spr(cpu, SPR_PURR)) {
747         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
748     }
749     if (ppc_has_spr(cpu, SPR_PURR)) {
750         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
751     }
752 
753     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
754         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
755                           segs, sizeof(segs))));
756     }
757 
758     /* Advertise VSX (vector extensions) if available
759      *   1               == VMX / Altivec available
760      *   2               == VSX available
761      *
762      * Only CPUs for which we create core types in spapr_cpu_core.c
763      * are possible, and all of those have VMX */
764     if (env->insns_flags & PPC_ALTIVEC) {
765         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
766             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
767         } else {
768             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
769         }
770     }
771 
772     /* Advertise DFP (Decimal Floating Point) if available
773      *   0 / no property == no DFP
774      *   1               == DFP available */
775     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
776         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
777     }
778 
779     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
780                                                       sizeof(page_sizes_prop));
781     if (page_sizes_prop_size) {
782         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
783                           page_sizes_prop, page_sizes_prop_size)));
784     }
785 
786     spapr_dt_pa_features(spapr, cpu, fdt, offset);
787 
788     spapr_dt_pi_features(spapr, cpu, fdt, offset);
789 
790     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
791                            cs->cpu_index / vcpus_per_socket)));
792 
793     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
794                       pft_size_prop, sizeof(pft_size_prop))));
795 
796     if (ms->numa_state->num_nodes > 1) {
797         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
798     }
799 
800     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
801 
802     if (pcc->radix_page_info) {
803         for (i = 0; i < pcc->radix_page_info->count; i++) {
804             radix_AP_encodings[i] =
805                 cpu_to_be32(pcc->radix_page_info->entries[i]);
806         }
807         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
808                           radix_AP_encodings,
809                           pcc->radix_page_info->count *
810                           sizeof(radix_AP_encodings[0]))));
811     }
812 
813     /*
814      * We set this property to let the guest know that it can use the large
815      * decrementer and its width in bits.
816      */
817     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
818         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
819                               pcc->lrg_decr_bits)));
820 }
821 
822 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
823                              int cpus_offset)
824 {
825     PowerPCCPU *cpu = POWERPC_CPU(cs);
826     int index = spapr_get_vcpu_id(cpu);
827     DeviceClass *dc = DEVICE_GET_CLASS(cs);
828     g_autofree char *nodename = NULL;
829     int offset;
830 
831     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
832         return;
833     }
834 
835     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
836     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
837     _FDT(offset);
838     spapr_dt_cpu(cs, fdt, offset, spapr);
839 }
840 
841 
842 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
843 {
844     CPUState **rev;
845     CPUState *cs;
846     int n_cpus;
847     int cpus_offset;
848     int i;
849 
850     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
851     _FDT(cpus_offset);
852     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
853     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
854 
855     /*
856      * We walk the CPUs in reverse order to ensure that CPU DT nodes
857      * created by fdt_add_subnode() end up in the right order in FDT
858      * for the guest kernel the enumerate the CPUs correctly.
859      *
860      * The CPU list cannot be traversed in reverse order, so we need
861      * to do extra work.
862      */
863     n_cpus = 0;
864     rev = NULL;
865     CPU_FOREACH(cs) {
866         rev = g_renew(CPUState *, rev, n_cpus + 1);
867         rev[n_cpus++] = cs;
868     }
869 
870     for (i = n_cpus - 1; i >= 0; i--) {
871         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
872     }
873 
874     g_free(rev);
875 }
876 
877 static int spapr_dt_rng(void *fdt)
878 {
879     int node;
880     int ret;
881 
882     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
883     if (node <= 0) {
884         return -1;
885     }
886     ret = fdt_setprop_string(fdt, node, "device_type",
887                              "ibm,platform-facilities");
888     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
889     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
890 
891     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
892     if (node <= 0) {
893         return -1;
894     }
895     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
896 
897     return ret ? -1 : 0;
898 }
899 
900 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
901 {
902     MachineState *ms = MACHINE(spapr);
903     int rtas;
904     GString *hypertas = g_string_sized_new(256);
905     GString *qemu_hypertas = g_string_sized_new(256);
906     uint32_t lrdr_capacity[] = {
907         0,
908         0,
909         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
910         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
911         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
912     };
913 
914     /* Do we have device memory? */
915     if (MACHINE(spapr)->device_memory) {
916         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
917             memory_region_size(&MACHINE(spapr)->device_memory->mr);
918 
919         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
920         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
921     }
922 
923     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
924 
925     /* hypertas */
926     add_str(hypertas, "hcall-pft");
927     add_str(hypertas, "hcall-term");
928     add_str(hypertas, "hcall-dabr");
929     add_str(hypertas, "hcall-interrupt");
930     add_str(hypertas, "hcall-tce");
931     add_str(hypertas, "hcall-vio");
932     add_str(hypertas, "hcall-splpar");
933     add_str(hypertas, "hcall-join");
934     add_str(hypertas, "hcall-bulk");
935     add_str(hypertas, "hcall-set-mode");
936     add_str(hypertas, "hcall-sprg0");
937     add_str(hypertas, "hcall-copy");
938     add_str(hypertas, "hcall-debug");
939     add_str(hypertas, "hcall-vphn");
940     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
941         add_str(hypertas, "hcall-rpt-invalidate");
942     }
943 
944     add_str(qemu_hypertas, "hcall-memop1");
945 
946     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
947         add_str(hypertas, "hcall-multi-tce");
948     }
949 
950     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
951         add_str(hypertas, "hcall-hpt-resize");
952     }
953 
954     add_str(hypertas, "hcall-watchdog");
955 
956     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
957                      hypertas->str, hypertas->len));
958     g_string_free(hypertas, TRUE);
959     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
960                      qemu_hypertas->str, qemu_hypertas->len));
961     g_string_free(qemu_hypertas, TRUE);
962 
963     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
964 
965     /*
966      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
967      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
968      *
969      * The system reset requirements are driven by existing Linux and PowerVM
970      * implementation which (contrary to PAPR) saves r3 in the error log
971      * structure like machine check, so Linux expects to find the saved r3
972      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
973      * does not look at the error value).
974      *
975      * System reset interrupts are not subject to interlock like machine
976      * check, so this memory area could be corrupted if the sreset is
977      * interrupted by a machine check (or vice versa) if it was shared. To
978      * prevent this, system reset uses per-CPU areas for the sreset save
979      * area. A system reset that interrupts a system reset handler could
980      * still overwrite this area, but Linux doesn't try to recover in that
981      * case anyway.
982      *
983      * The extra 8 bytes is required because Linux's FWNMI error log check
984      * is off-by-one.
985      *
986      * RTAS_MIN_SIZE is required for the RTAS blob itself.
987      */
988     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
989                           RTAS_ERROR_LOG_MAX +
990                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
991                           sizeof(uint64_t)));
992     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
993                           RTAS_ERROR_LOG_MAX));
994     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
995                           RTAS_EVENT_SCAN_RATE));
996 
997     g_assert(msi_nonbroken);
998     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
999 
1000     /*
1001      * According to PAPR, rtas ibm,os-term does not guarantee a return
1002      * back to the guest cpu.
1003      *
1004      * While an additional ibm,extended-os-term property indicates
1005      * that rtas call return will always occur. Set this property.
1006      */
1007     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1008 
1009     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1010                      lrdr_capacity, sizeof(lrdr_capacity)));
1011 
1012     spapr_dt_rtas_tokens(fdt, rtas);
1013 }
1014 
1015 /*
1016  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1017  * and the XIVE features that the guest may request and thus the valid
1018  * values for bytes 23..26 of option vector 5:
1019  */
1020 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1021                                           int chosen)
1022 {
1023     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1024 
1025     char val[2 * 4] = {
1026         23, 0x00, /* XICS / XIVE mode */
1027         24, 0x00, /* Hash/Radix, filled in below. */
1028         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1029         26, 0x40, /* Radix options: GTSE == yes. */
1030     };
1031 
1032     if (spapr->irq->xics && spapr->irq->xive) {
1033         val[1] = SPAPR_OV5_XIVE_BOTH;
1034     } else if (spapr->irq->xive) {
1035         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1036     } else {
1037         assert(spapr->irq->xics);
1038         val[1] = SPAPR_OV5_XIVE_LEGACY;
1039     }
1040 
1041     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1042                           first_ppc_cpu->compat_pvr)) {
1043         /*
1044          * If we're in a pre POWER9 compat mode then the guest should
1045          * do hash and use the legacy interrupt mode
1046          */
1047         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1048         val[3] = 0x00; /* Hash */
1049         spapr_check_mmu_mode(false);
1050     } else if (kvm_enabled()) {
1051         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1052             val[3] = 0x80; /* OV5_MMU_BOTH */
1053         } else if (kvmppc_has_cap_mmu_radix()) {
1054             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1055         } else {
1056             val[3] = 0x00; /* Hash */
1057         }
1058     } else {
1059         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1060         val[3] = 0xC0;
1061     }
1062     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1063                      val, sizeof(val)));
1064 }
1065 
1066 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1067 {
1068     MachineState *machine = MACHINE(spapr);
1069     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1070     int chosen;
1071 
1072     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1073 
1074     if (reset) {
1075         const char *boot_device = spapr->boot_device;
1076         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1077         size_t cb = 0;
1078         g_autofree char *bootlist = get_boot_devices_list(&cb);
1079 
1080         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1081             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1082                                     machine->kernel_cmdline));
1083         }
1084 
1085         if (spapr->initrd_size) {
1086             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1087                                   spapr->initrd_base));
1088             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1089                                   spapr->initrd_base + spapr->initrd_size));
1090         }
1091 
1092         if (spapr->kernel_size) {
1093             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1094                                   cpu_to_be64(spapr->kernel_size) };
1095 
1096             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1097                          &kprop, sizeof(kprop)));
1098             if (spapr->kernel_le) {
1099                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1100             }
1101         }
1102         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1103             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1104         }
1105         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1106         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1107         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1108 
1109         if (cb && bootlist) {
1110             int i;
1111 
1112             for (i = 0; i < cb; i++) {
1113                 if (bootlist[i] == '\n') {
1114                     bootlist[i] = ' ';
1115                 }
1116             }
1117             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1118         }
1119 
1120         if (boot_device && strlen(boot_device)) {
1121             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1122         }
1123 
1124         if (spapr->want_stdout_path && stdout_path) {
1125             /*
1126              * "linux,stdout-path" and "stdout" properties are
1127              * deprecated by linux kernel. New platforms should only
1128              * use the "stdout-path" property. Set the new property
1129              * and continue using older property to remain compatible
1130              * with the existing firmware.
1131              */
1132             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1133             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1134         }
1135 
1136         /*
1137          * We can deal with BAR reallocation just fine, advertise it
1138          * to the guest
1139          */
1140         if (smc->linux_pci_probe) {
1141             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1142         }
1143 
1144         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1145     }
1146 
1147     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1148 
1149     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1150 }
1151 
1152 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1153 {
1154     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1155      * KVM to work under pHyp with some guest co-operation */
1156     int hypervisor;
1157     uint8_t hypercall[16];
1158 
1159     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1160     /* indicate KVM hypercall interface */
1161     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1162     if (kvmppc_has_cap_fixup_hcalls()) {
1163         /*
1164          * Older KVM versions with older guest kernels were broken
1165          * with the magic page, don't allow the guest to map it.
1166          */
1167         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1168                                   sizeof(hypercall))) {
1169             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1170                              hypercall, sizeof(hypercall)));
1171         }
1172     }
1173 }
1174 
1175 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1176 {
1177     MachineState *machine = MACHINE(spapr);
1178     MachineClass *mc = MACHINE_GET_CLASS(machine);
1179     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1180     uint32_t root_drc_type_mask = 0;
1181     int ret;
1182     void *fdt;
1183     SpaprPhbState *phb;
1184     char *buf;
1185 
1186     fdt = g_malloc0(space);
1187     _FDT((fdt_create_empty_tree(fdt, space)));
1188 
1189     /* Root node */
1190     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1191     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1192     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1193 
1194     /* Guest UUID & Name*/
1195     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1196     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1197     if (qemu_uuid_set) {
1198         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1199     }
1200     g_free(buf);
1201 
1202     if (qemu_get_vm_name()) {
1203         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1204                                 qemu_get_vm_name()));
1205     }
1206 
1207     /* Host Model & Serial Number */
1208     if (spapr->host_model) {
1209         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1210     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1211         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1212         g_free(buf);
1213     }
1214 
1215     if (spapr->host_serial) {
1216         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1217     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1218         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1219         g_free(buf);
1220     }
1221 
1222     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1223     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1224 
1225     /* /interrupt controller */
1226     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1227 
1228     ret = spapr_dt_memory(spapr, fdt);
1229     if (ret < 0) {
1230         error_report("couldn't setup memory nodes in fdt");
1231         exit(1);
1232     }
1233 
1234     /* /vdevice */
1235     spapr_dt_vdevice(spapr->vio_bus, fdt);
1236 
1237     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1238         ret = spapr_dt_rng(fdt);
1239         if (ret < 0) {
1240             error_report("could not set up rng device in the fdt");
1241             exit(1);
1242         }
1243     }
1244 
1245     QLIST_FOREACH(phb, &spapr->phbs, list) {
1246         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1247         if (ret < 0) {
1248             error_report("couldn't setup PCI devices in fdt");
1249             exit(1);
1250         }
1251     }
1252 
1253     spapr_dt_cpus(fdt, spapr);
1254 
1255     /* ibm,drc-indexes and friends */
1256     root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1257     if (smc->dr_phb_enabled) {
1258         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1259     }
1260     if (mc->nvdimm_supported) {
1261         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1262     }
1263     if (root_drc_type_mask) {
1264         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1265     }
1266 
1267     if (mc->has_hotpluggable_cpus) {
1268         int offset = fdt_path_offset(fdt, "/cpus");
1269         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1270         if (ret < 0) {
1271             error_report("Couldn't set up CPU DR device tree properties");
1272             exit(1);
1273         }
1274     }
1275 
1276     /* /event-sources */
1277     spapr_dt_events(spapr, fdt);
1278 
1279     /* /rtas */
1280     spapr_dt_rtas(spapr, fdt);
1281 
1282     /* /chosen */
1283     spapr_dt_chosen(spapr, fdt, reset);
1284 
1285     /* /hypervisor */
1286     if (kvm_enabled()) {
1287         spapr_dt_hypervisor(spapr, fdt);
1288     }
1289 
1290     /* Build memory reserve map */
1291     if (reset) {
1292         if (spapr->kernel_size) {
1293             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1294                                   spapr->kernel_size)));
1295         }
1296         if (spapr->initrd_size) {
1297             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1298                                   spapr->initrd_size)));
1299         }
1300     }
1301 
1302     /* NVDIMM devices */
1303     if (mc->nvdimm_supported) {
1304         spapr_dt_persistent_memory(spapr, fdt);
1305     }
1306 
1307     return fdt;
1308 }
1309 
1310 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1311 {
1312     SpaprMachineState *spapr = opaque;
1313 
1314     return (addr & 0x0fffffff) + spapr->kernel_addr;
1315 }
1316 
1317 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1318                                     PowerPCCPU *cpu)
1319 {
1320     CPUPPCState *env = &cpu->env;
1321 
1322     /* The TCG path should also be holding the BQL at this point */
1323     g_assert(bql_locked());
1324 
1325     g_assert(!vhyp_cpu_in_nested(cpu));
1326 
1327     if (FIELD_EX64(env->msr, MSR, PR)) {
1328         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1329         env->gpr[3] = H_PRIVILEGE;
1330     } else {
1331         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1332     }
1333 }
1334 
1335 struct LPCRSyncState {
1336     target_ulong value;
1337     target_ulong mask;
1338 };
1339 
1340 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1341 {
1342     struct LPCRSyncState *s = arg.host_ptr;
1343     PowerPCCPU *cpu = POWERPC_CPU(cs);
1344     CPUPPCState *env = &cpu->env;
1345     target_ulong lpcr;
1346 
1347     cpu_synchronize_state(cs);
1348     lpcr = env->spr[SPR_LPCR];
1349     lpcr &= ~s->mask;
1350     lpcr |= s->value;
1351     ppc_store_lpcr(cpu, lpcr);
1352 }
1353 
1354 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1355 {
1356     CPUState *cs;
1357     struct LPCRSyncState s = {
1358         .value = value,
1359         .mask = mask
1360     };
1361     CPU_FOREACH(cs) {
1362         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1363     }
1364 }
1365 
1366 /* May be used when the machine is not running */
1367 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1368 {
1369     CPUState *cs;
1370     CPU_FOREACH(cs) {
1371         PowerPCCPU *cpu = POWERPC_CPU(cs);
1372         CPUPPCState *env = &cpu->env;
1373         target_ulong lpcr;
1374 
1375         lpcr = env->spr[SPR_LPCR];
1376         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1377         ppc_store_lpcr(cpu, lpcr);
1378     }
1379 }
1380 
1381 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1382                            target_ulong lpid, ppc_v3_pate_t *entry)
1383 {
1384     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1385     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1386 
1387     if (!spapr_cpu->in_nested) {
1388         assert(lpid == 0);
1389 
1390         /* Copy PATE1:GR into PATE0:HR */
1391         entry->dw0 = spapr->patb_entry & PATE0_HR;
1392         entry->dw1 = spapr->patb_entry;
1393         return true;
1394     } else {
1395         if (spapr_nested_api(spapr) == NESTED_API_KVM_HV) {
1396             return spapr_get_pate_nested_hv(spapr, cpu, lpid, entry);
1397         } else if (spapr_nested_api(spapr) == NESTED_API_PAPR) {
1398             return spapr_get_pate_nested_papr(spapr, cpu, lpid, entry);
1399         } else {
1400             g_assert_not_reached();
1401         }
1402     }
1403 }
1404 
1405 static uint64_t *hpte_get_ptr(SpaprMachineState *s, unsigned index)
1406 {
1407     uint64_t *table = s->htab;
1408 
1409     return &table[2 * index];
1410 }
1411 
1412 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1413 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1414 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1415 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1416 
1417 /*
1418  * Get the fd to access the kernel htab, re-opening it if necessary
1419  */
1420 static int get_htab_fd(SpaprMachineState *spapr)
1421 {
1422     Error *local_err = NULL;
1423 
1424     if (spapr->htab_fd >= 0) {
1425         return spapr->htab_fd;
1426     }
1427 
1428     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1429     if (spapr->htab_fd < 0) {
1430         error_report_err(local_err);
1431     }
1432 
1433     return spapr->htab_fd;
1434 }
1435 
1436 void close_htab_fd(SpaprMachineState *spapr)
1437 {
1438     if (spapr->htab_fd >= 0) {
1439         close(spapr->htab_fd);
1440     }
1441     spapr->htab_fd = -1;
1442 }
1443 
1444 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1445 {
1446     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1447 
1448     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1449 }
1450 
1451 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1452 {
1453     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1454 
1455     assert(kvm_enabled());
1456 
1457     if (!spapr->htab) {
1458         return 0;
1459     }
1460 
1461     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1462 }
1463 
1464 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1465                                                 hwaddr ptex, int n)
1466 {
1467     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1468     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1469 
1470     if (!spapr->htab) {
1471         /*
1472          * HTAB is controlled by KVM. Fetch into temporary buffer
1473          */
1474         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1475         kvmppc_read_hptes(hptes, ptex, n);
1476         return hptes;
1477     }
1478 
1479     /*
1480      * HTAB is controlled by QEMU. Just point to the internally
1481      * accessible PTEG.
1482      */
1483     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1484 }
1485 
1486 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1487                               const ppc_hash_pte64_t *hptes,
1488                               hwaddr ptex, int n)
1489 {
1490     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1491 
1492     if (!spapr->htab) {
1493         g_free((void *)hptes);
1494     }
1495 
1496     /* Nothing to do for qemu managed HPT */
1497 }
1498 
1499 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1500                       uint64_t pte0, uint64_t pte1)
1501 {
1502     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1503     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1504 
1505     if (!spapr->htab) {
1506         kvmppc_write_hpte(ptex, pte0, pte1);
1507     } else {
1508         if (pte0 & HPTE64_V_VALID) {
1509             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1510             /*
1511              * When setting valid, we write PTE1 first. This ensures
1512              * proper synchronization with the reading code in
1513              * ppc_hash64_pteg_search()
1514              */
1515             smp_wmb();
1516             stq_p(spapr->htab + offset, pte0);
1517         } else {
1518             stq_p(spapr->htab + offset, pte0);
1519             /*
1520              * When clearing it we set PTE0 first. This ensures proper
1521              * synchronization with the reading code in
1522              * ppc_hash64_pteg_search()
1523              */
1524             smp_wmb();
1525             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1526         }
1527     }
1528 }
1529 
1530 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1531                              uint64_t pte1)
1532 {
1533     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1534     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1535 
1536     if (!spapr->htab) {
1537         /* There should always be a hash table when this is called */
1538         error_report("spapr_hpte_set_c called with no hash table !");
1539         return;
1540     }
1541 
1542     /* The HW performs a non-atomic byte update */
1543     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1544 }
1545 
1546 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1547                              uint64_t pte1)
1548 {
1549     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1550     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1551 
1552     if (!spapr->htab) {
1553         /* There should always be a hash table when this is called */
1554         error_report("spapr_hpte_set_r called with no hash table !");
1555         return;
1556     }
1557 
1558     /* The HW performs a non-atomic byte update */
1559     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1560 }
1561 
1562 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1563 {
1564     int shift;
1565 
1566     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1567      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1568      * that's much more than is needed for Linux guests */
1569     shift = ctz64(pow2ceil(ramsize)) - 7;
1570     shift = MAX(shift, 18); /* Minimum architected size */
1571     shift = MIN(shift, 46); /* Maximum architected size */
1572     return shift;
1573 }
1574 
1575 void spapr_free_hpt(SpaprMachineState *spapr)
1576 {
1577     qemu_vfree(spapr->htab);
1578     spapr->htab = NULL;
1579     spapr->htab_shift = 0;
1580     close_htab_fd(spapr);
1581 }
1582 
1583 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1584 {
1585     ERRP_GUARD();
1586     long rc;
1587 
1588     /* Clean up any HPT info from a previous boot */
1589     spapr_free_hpt(spapr);
1590 
1591     rc = kvmppc_reset_htab(shift);
1592 
1593     if (rc == -EOPNOTSUPP) {
1594         error_setg(errp, "HPT not supported in nested guests");
1595         return -EOPNOTSUPP;
1596     }
1597 
1598     if (rc < 0) {
1599         /* kernel-side HPT needed, but couldn't allocate one */
1600         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1601                          shift);
1602         error_append_hint(errp, "Try smaller maxmem?\n");
1603         return -errno;
1604     } else if (rc > 0) {
1605         /* kernel-side HPT allocated */
1606         if (rc != shift) {
1607             error_setg(errp,
1608                        "Requested order %d HPT, but kernel allocated order %ld",
1609                        shift, rc);
1610             error_append_hint(errp, "Try smaller maxmem?\n");
1611             return -ENOSPC;
1612         }
1613 
1614         spapr->htab_shift = shift;
1615         spapr->htab = NULL;
1616     } else {
1617         /* kernel-side HPT not needed, allocate in userspace instead */
1618         size_t size = 1ULL << shift;
1619         int i;
1620 
1621         spapr->htab = qemu_memalign(size, size);
1622         memset(spapr->htab, 0, size);
1623         spapr->htab_shift = shift;
1624 
1625         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1626             DIRTY_HPTE(hpte_get_ptr(spapr, i));
1627         }
1628     }
1629     /* We're setting up a hash table, so that means we're not radix */
1630     spapr->patb_entry = 0;
1631     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1632     return 0;
1633 }
1634 
1635 void spapr_setup_hpt(SpaprMachineState *spapr)
1636 {
1637     int hpt_shift;
1638 
1639     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1640         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1641     } else {
1642         uint64_t current_ram_size;
1643 
1644         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1645         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1646     }
1647     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1648 
1649     if (kvm_enabled()) {
1650         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1651 
1652         /* Check our RMA fits in the possible VRMA */
1653         if (vrma_limit < spapr->rma_size) {
1654             error_report("Unable to create %" HWADDR_PRIu
1655                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1656                          spapr->rma_size / MiB, vrma_limit / MiB);
1657             exit(EXIT_FAILURE);
1658         }
1659     }
1660 }
1661 
1662 void spapr_check_mmu_mode(bool guest_radix)
1663 {
1664     if (guest_radix) {
1665         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1666             error_report("Guest requested unavailable MMU mode (radix).");
1667             exit(EXIT_FAILURE);
1668         }
1669     } else {
1670         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1671             && !kvmppc_has_cap_mmu_hash_v3()) {
1672             error_report("Guest requested unavailable MMU mode (hash).");
1673             exit(EXIT_FAILURE);
1674         }
1675     }
1676 }
1677 
1678 static void spapr_machine_reset(MachineState *machine, ResetType type)
1679 {
1680     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1681     PowerPCCPU *first_ppc_cpu;
1682     hwaddr fdt_addr;
1683     void *fdt;
1684     int rc;
1685 
1686     if (type != RESET_TYPE_SNAPSHOT_LOAD) {
1687         /*
1688          * Record-replay snapshot load must not consume random, this was
1689          * already replayed from initial machine reset.
1690          */
1691         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1692     }
1693 
1694     if (machine->cgs) {
1695         confidential_guest_kvm_reset(machine->cgs, &error_fatal);
1696     }
1697     spapr_caps_apply(spapr);
1698     spapr_nested_reset(spapr);
1699 
1700     first_ppc_cpu = POWERPC_CPU(first_cpu);
1701     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1702         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1703                               spapr->max_compat_pvr)) {
1704         /*
1705          * If using KVM with radix mode available, VCPUs can be started
1706          * without a HPT because KVM will start them in radix mode.
1707          * Set the GR bit in PATE so that we know there is no HPT.
1708          */
1709         spapr->patb_entry = PATE1_GR;
1710         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1711     } else {
1712         spapr_setup_hpt(spapr);
1713     }
1714 
1715     qemu_devices_reset(type);
1716 
1717     spapr_ovec_cleanup(spapr->ov5_cas);
1718     spapr->ov5_cas = spapr_ovec_new();
1719 
1720     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1721 
1722     /*
1723      * This is fixing some of the default configuration of the XIVE
1724      * devices. To be called after the reset of the machine devices.
1725      */
1726     spapr_irq_reset(spapr, &error_fatal);
1727 
1728     /*
1729      * There is no CAS under qtest. Simulate one to please the code that
1730      * depends on spapr->ov5_cas. This is especially needed to test device
1731      * unplug, so we do that before resetting the DRCs.
1732      */
1733     if (qtest_enabled()) {
1734         spapr_ovec_cleanup(spapr->ov5_cas);
1735         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1736     }
1737 
1738     spapr_nvdimm_finish_flushes();
1739 
1740     /* DRC reset may cause a device to be unplugged. This will cause troubles
1741      * if this device is used by another device (eg, a running vhost backend
1742      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1743      * situations, we reset DRCs after all devices have been reset.
1744      */
1745     spapr_drc_reset_all(spapr);
1746 
1747     spapr_clear_pending_events(spapr);
1748 
1749     /*
1750      * We place the device tree just below either the top of the RMA,
1751      * or just below 2GB, whichever is lower, so that it can be
1752      * processed with 32-bit real mode code if necessary
1753      */
1754     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1755 
1756     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1757     if (spapr->vof) {
1758         spapr_vof_reset(spapr, fdt, &error_fatal);
1759         /*
1760          * Do not pack the FDT as the client may change properties.
1761          * VOF client does not expect the FDT so we do not load it to the VM.
1762          */
1763     } else {
1764         rc = fdt_pack(fdt);
1765         /* Should only fail if we've built a corrupted tree */
1766         assert(rc == 0);
1767 
1768         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1769                                   0, fdt_addr, 0);
1770         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1771     }
1772 
1773     g_free(spapr->fdt_blob);
1774     spapr->fdt_size = fdt_totalsize(fdt);
1775     spapr->fdt_initial_size = spapr->fdt_size;
1776     spapr->fdt_blob = fdt;
1777 
1778     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1779     machine->fdt = fdt;
1780 
1781     /* Set up the entry state */
1782     first_ppc_cpu->env.gpr[5] = 0;
1783 
1784     spapr->fwnmi_system_reset_addr = -1;
1785     spapr->fwnmi_machine_check_addr = -1;
1786     spapr->fwnmi_machine_check_interlock = -1;
1787 
1788     /* Signal all vCPUs waiting on this condition */
1789     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1790 
1791     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1792 }
1793 
1794 static void spapr_create_nvram(SpaprMachineState *spapr)
1795 {
1796     DeviceState *dev = qdev_new("spapr-nvram");
1797     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1798 
1799     if (dinfo) {
1800         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1801                                 &error_fatal);
1802     }
1803 
1804     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1805 
1806     spapr->nvram = (struct SpaprNvram *)dev;
1807 }
1808 
1809 static void spapr_rtc_create(SpaprMachineState *spapr)
1810 {
1811     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1812                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1813                                        &error_fatal, NULL);
1814     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1815     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1816                               "date");
1817 }
1818 
1819 /* Returns whether we want to use VGA or not */
1820 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1821 {
1822     vga_interface_created = true;
1823     switch (vga_interface_type) {
1824     case VGA_NONE:
1825         return false;
1826     case VGA_DEVICE:
1827         return true;
1828     case VGA_STD:
1829     case VGA_VIRTIO:
1830     case VGA_CIRRUS:
1831         return pci_vga_init(pci_bus) != NULL;
1832     default:
1833         error_setg(errp,
1834                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1835         return false;
1836     }
1837 }
1838 
1839 static int spapr_pre_load(void *opaque)
1840 {
1841     int rc;
1842 
1843     rc = spapr_caps_pre_load(opaque);
1844     if (rc) {
1845         return rc;
1846     }
1847 
1848     return 0;
1849 }
1850 
1851 static int spapr_post_load(void *opaque, int version_id)
1852 {
1853     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1854     int err = 0;
1855 
1856     err = spapr_caps_post_migration(spapr);
1857     if (err) {
1858         return err;
1859     }
1860 
1861     /*
1862      * In earlier versions, there was no separate qdev for the PAPR
1863      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1864      * So when migrating from those versions, poke the incoming offset
1865      * value into the RTC device
1866      */
1867     if (version_id < 3) {
1868         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1869         if (err) {
1870             return err;
1871         }
1872     }
1873 
1874     if (kvm_enabled() && spapr->patb_entry) {
1875         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1876         bool radix = !!(spapr->patb_entry & PATE1_GR);
1877         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1878 
1879         /*
1880          * Update LPCR:HR and UPRT as they may not be set properly in
1881          * the stream
1882          */
1883         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1884                             LPCR_HR | LPCR_UPRT);
1885 
1886         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1887         if (err) {
1888             error_report("Process table config unsupported by the host");
1889             return -EINVAL;
1890         }
1891     }
1892 
1893     err = spapr_irq_post_load(spapr, version_id);
1894     if (err) {
1895         return err;
1896     }
1897 
1898     return err;
1899 }
1900 
1901 static int spapr_pre_save(void *opaque)
1902 {
1903     int rc;
1904 
1905     rc = spapr_caps_pre_save(opaque);
1906     if (rc) {
1907         return rc;
1908     }
1909 
1910     return 0;
1911 }
1912 
1913 static bool version_before_3(void *opaque, int version_id)
1914 {
1915     return version_id < 3;
1916 }
1917 
1918 static bool spapr_pending_events_needed(void *opaque)
1919 {
1920     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1921     return !QTAILQ_EMPTY(&spapr->pending_events);
1922 }
1923 
1924 static const VMStateDescription vmstate_spapr_event_entry = {
1925     .name = "spapr_event_log_entry",
1926     .version_id = 1,
1927     .minimum_version_id = 1,
1928     .fields = (const VMStateField[]) {
1929         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1930         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1931         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1932                                      NULL, extended_length),
1933         VMSTATE_END_OF_LIST()
1934     },
1935 };
1936 
1937 static const VMStateDescription vmstate_spapr_pending_events = {
1938     .name = "spapr_pending_events",
1939     .version_id = 1,
1940     .minimum_version_id = 1,
1941     .needed = spapr_pending_events_needed,
1942     .fields = (const VMStateField[]) {
1943         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1944                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1945         VMSTATE_END_OF_LIST()
1946     },
1947 };
1948 
1949 static bool spapr_ov5_cas_needed(void *opaque)
1950 {
1951     SpaprMachineState *spapr = opaque;
1952     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1953     bool cas_needed;
1954 
1955     /* Prior to the introduction of SpaprOptionVector, we had two option
1956      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1957      * Both of these options encode machine topology into the device-tree
1958      * in such a way that the now-booted OS should still be able to interact
1959      * appropriately with QEMU regardless of what options were actually
1960      * negotiatied on the source side.
1961      *
1962      * As such, we can avoid migrating the CAS-negotiated options if these
1963      * are the only options available on the current machine/platform.
1964      * Since these are the only options available for pseries-2.7 and
1965      * earlier, this allows us to maintain old->new/new->old migration
1966      * compatibility.
1967      *
1968      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1969      * via default pseries-2.8 machines and explicit command-line parameters.
1970      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1971      * of the actual CAS-negotiated values to continue working properly. For
1972      * example, availability of memory unplug depends on knowing whether
1973      * OV5_HP_EVT was negotiated via CAS.
1974      *
1975      * Thus, for any cases where the set of available CAS-negotiatable
1976      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1977      * include the CAS-negotiated options in the migration stream, unless
1978      * if they affect boot time behaviour only.
1979      */
1980     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1981     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1982     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1983 
1984     /* We need extra information if we have any bits outside the mask
1985      * defined above */
1986     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1987 
1988     spapr_ovec_cleanup(ov5_mask);
1989 
1990     return cas_needed;
1991 }
1992 
1993 static const VMStateDescription vmstate_spapr_ov5_cas = {
1994     .name = "spapr_option_vector_ov5_cas",
1995     .version_id = 1,
1996     .minimum_version_id = 1,
1997     .needed = spapr_ov5_cas_needed,
1998     .fields = (const VMStateField[]) {
1999         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2000                                  vmstate_spapr_ovec, SpaprOptionVector),
2001         VMSTATE_END_OF_LIST()
2002     },
2003 };
2004 
2005 static bool spapr_patb_entry_needed(void *opaque)
2006 {
2007     SpaprMachineState *spapr = opaque;
2008 
2009     return !!spapr->patb_entry;
2010 }
2011 
2012 static const VMStateDescription vmstate_spapr_patb_entry = {
2013     .name = "spapr_patb_entry",
2014     .version_id = 1,
2015     .minimum_version_id = 1,
2016     .needed = spapr_patb_entry_needed,
2017     .fields = (const VMStateField[]) {
2018         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2019         VMSTATE_END_OF_LIST()
2020     },
2021 };
2022 
2023 static bool spapr_irq_map_needed(void *opaque)
2024 {
2025     SpaprMachineState *spapr = opaque;
2026 
2027     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2028 }
2029 
2030 static const VMStateDescription vmstate_spapr_irq_map = {
2031     .name = "spapr_irq_map",
2032     .version_id = 1,
2033     .minimum_version_id = 1,
2034     .needed = spapr_irq_map_needed,
2035     .fields = (const VMStateField[]) {
2036         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2037         VMSTATE_END_OF_LIST()
2038     },
2039 };
2040 
2041 static bool spapr_dtb_needed(void *opaque)
2042 {
2043     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2044 
2045     return smc->update_dt_enabled;
2046 }
2047 
2048 static int spapr_dtb_pre_load(void *opaque)
2049 {
2050     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2051 
2052     g_free(spapr->fdt_blob);
2053     spapr->fdt_blob = NULL;
2054     spapr->fdt_size = 0;
2055 
2056     return 0;
2057 }
2058 
2059 static const VMStateDescription vmstate_spapr_dtb = {
2060     .name = "spapr_dtb",
2061     .version_id = 1,
2062     .minimum_version_id = 1,
2063     .needed = spapr_dtb_needed,
2064     .pre_load = spapr_dtb_pre_load,
2065     .fields = (const VMStateField[]) {
2066         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2067         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2068         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2069                                      fdt_size),
2070         VMSTATE_END_OF_LIST()
2071     },
2072 };
2073 
2074 static bool spapr_fwnmi_needed(void *opaque)
2075 {
2076     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2077 
2078     return spapr->fwnmi_machine_check_addr != -1;
2079 }
2080 
2081 static int spapr_fwnmi_pre_save(void *opaque)
2082 {
2083     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2084 
2085     /*
2086      * Check if machine check handling is in progress and print a
2087      * warning message.
2088      */
2089     if (spapr->fwnmi_machine_check_interlock != -1) {
2090         warn_report("A machine check is being handled during migration. The"
2091                 "handler may run and log hardware error on the destination");
2092     }
2093 
2094     return 0;
2095 }
2096 
2097 static const VMStateDescription vmstate_spapr_fwnmi = {
2098     .name = "spapr_fwnmi",
2099     .version_id = 1,
2100     .minimum_version_id = 1,
2101     .needed = spapr_fwnmi_needed,
2102     .pre_save = spapr_fwnmi_pre_save,
2103     .fields = (const VMStateField[]) {
2104         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2105         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2106         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2107         VMSTATE_END_OF_LIST()
2108     },
2109 };
2110 
2111 static const VMStateDescription vmstate_spapr = {
2112     .name = "spapr",
2113     .version_id = 3,
2114     .minimum_version_id = 1,
2115     .pre_load = spapr_pre_load,
2116     .post_load = spapr_post_load,
2117     .pre_save = spapr_pre_save,
2118     .fields = (const VMStateField[]) {
2119         /* used to be @next_irq */
2120         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2121 
2122         /* RTC offset */
2123         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2124 
2125         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2126         VMSTATE_END_OF_LIST()
2127     },
2128     .subsections = (const VMStateDescription * const []) {
2129         &vmstate_spapr_ov5_cas,
2130         &vmstate_spapr_patb_entry,
2131         &vmstate_spapr_pending_events,
2132         &vmstate_spapr_cap_htm,
2133         &vmstate_spapr_cap_vsx,
2134         &vmstate_spapr_cap_dfp,
2135         &vmstate_spapr_cap_cfpc,
2136         &vmstate_spapr_cap_sbbc,
2137         &vmstate_spapr_cap_ibs,
2138         &vmstate_spapr_cap_hpt_maxpagesize,
2139         &vmstate_spapr_irq_map,
2140         &vmstate_spapr_cap_nested_kvm_hv,
2141         &vmstate_spapr_dtb,
2142         &vmstate_spapr_cap_large_decr,
2143         &vmstate_spapr_cap_ccf_assist,
2144         &vmstate_spapr_cap_fwnmi,
2145         &vmstate_spapr_fwnmi,
2146         &vmstate_spapr_cap_rpt_invalidate,
2147         &vmstate_spapr_cap_ail_mode_3,
2148         &vmstate_spapr_cap_nested_papr,
2149         NULL
2150     }
2151 };
2152 
2153 static int htab_save_setup(QEMUFile *f, void *opaque, Error **errp)
2154 {
2155     SpaprMachineState *spapr = opaque;
2156 
2157     /* "Iteration" header */
2158     if (!spapr->htab_shift) {
2159         qemu_put_be32(f, -1);
2160     } else {
2161         qemu_put_be32(f, spapr->htab_shift);
2162     }
2163 
2164     if (spapr->htab) {
2165         spapr->htab_save_index = 0;
2166         spapr->htab_first_pass = true;
2167     } else {
2168         if (spapr->htab_shift) {
2169             assert(kvm_enabled());
2170         }
2171     }
2172 
2173 
2174     return 0;
2175 }
2176 
2177 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2178                             int chunkstart, int n_valid, int n_invalid)
2179 {
2180     qemu_put_be32(f, chunkstart);
2181     qemu_put_be16(f, n_valid);
2182     qemu_put_be16(f, n_invalid);
2183     qemu_put_buffer(f, (void *)hpte_get_ptr(spapr, chunkstart),
2184                     HASH_PTE_SIZE_64 * n_valid);
2185 }
2186 
2187 static void htab_save_end_marker(QEMUFile *f)
2188 {
2189     qemu_put_be32(f, 0);
2190     qemu_put_be16(f, 0);
2191     qemu_put_be16(f, 0);
2192 }
2193 
2194 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2195                                  int64_t max_ns)
2196 {
2197     bool has_timeout = max_ns != -1;
2198     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2199     int index = spapr->htab_save_index;
2200     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2201 
2202     assert(spapr->htab_first_pass);
2203 
2204     do {
2205         int chunkstart;
2206 
2207         /* Consume invalid HPTEs */
2208         while ((index < htabslots)
2209                && !HPTE_VALID(hpte_get_ptr(spapr, index))) {
2210             CLEAN_HPTE(hpte_get_ptr(spapr, index));
2211             index++;
2212         }
2213 
2214         /* Consume valid HPTEs */
2215         chunkstart = index;
2216         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2217                && HPTE_VALID(hpte_get_ptr(spapr, index))) {
2218             CLEAN_HPTE(hpte_get_ptr(spapr, index));
2219             index++;
2220         }
2221 
2222         if (index > chunkstart) {
2223             int n_valid = index - chunkstart;
2224 
2225             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2226 
2227             if (has_timeout &&
2228                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2229                 break;
2230             }
2231         }
2232     } while ((index < htabslots) && !migration_rate_exceeded(f));
2233 
2234     if (index >= htabslots) {
2235         assert(index == htabslots);
2236         index = 0;
2237         spapr->htab_first_pass = false;
2238     }
2239     spapr->htab_save_index = index;
2240 }
2241 
2242 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2243                                 int64_t max_ns)
2244 {
2245     bool final = max_ns < 0;
2246     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2247     int examined = 0, sent = 0;
2248     int index = spapr->htab_save_index;
2249     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2250 
2251     assert(!spapr->htab_first_pass);
2252 
2253     do {
2254         int chunkstart, invalidstart;
2255 
2256         /* Consume non-dirty HPTEs */
2257         while ((index < htabslots)
2258                && !HPTE_DIRTY(hpte_get_ptr(spapr, index))) {
2259             index++;
2260             examined++;
2261         }
2262 
2263         chunkstart = index;
2264         /* Consume valid dirty HPTEs */
2265         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2266                && HPTE_DIRTY(hpte_get_ptr(spapr, index))
2267                && HPTE_VALID(hpte_get_ptr(spapr, index))) {
2268             CLEAN_HPTE(hpte_get_ptr(spapr, index));
2269             index++;
2270             examined++;
2271         }
2272 
2273         invalidstart = index;
2274         /* Consume invalid dirty HPTEs */
2275         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2276                && HPTE_DIRTY(hpte_get_ptr(spapr, index))
2277                && !HPTE_VALID(hpte_get_ptr(spapr, index))) {
2278             CLEAN_HPTE(hpte_get_ptr(spapr, index));
2279             index++;
2280             examined++;
2281         }
2282 
2283         if (index > chunkstart) {
2284             int n_valid = invalidstart - chunkstart;
2285             int n_invalid = index - invalidstart;
2286 
2287             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2288             sent += index - chunkstart;
2289 
2290             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2291                 break;
2292             }
2293         }
2294 
2295         if (examined >= htabslots) {
2296             break;
2297         }
2298 
2299         if (index >= htabslots) {
2300             assert(index == htabslots);
2301             index = 0;
2302         }
2303     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2304 
2305     if (index >= htabslots) {
2306         assert(index == htabslots);
2307         index = 0;
2308     }
2309 
2310     spapr->htab_save_index = index;
2311 
2312     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2313 }
2314 
2315 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2316 #define MAX_KVM_BUF_SIZE    2048
2317 
2318 static int htab_save_iterate(QEMUFile *f, void *opaque)
2319 {
2320     SpaprMachineState *spapr = opaque;
2321     int fd;
2322     int rc = 0;
2323 
2324     /* Iteration header */
2325     if (!spapr->htab_shift) {
2326         qemu_put_be32(f, -1);
2327         return 1;
2328     } else {
2329         qemu_put_be32(f, 0);
2330     }
2331 
2332     if (!spapr->htab) {
2333         assert(kvm_enabled());
2334 
2335         fd = get_htab_fd(spapr);
2336         if (fd < 0) {
2337             return fd;
2338         }
2339 
2340         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2341         if (rc < 0) {
2342             return rc;
2343         }
2344     } else  if (spapr->htab_first_pass) {
2345         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2346     } else {
2347         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2348     }
2349 
2350     htab_save_end_marker(f);
2351 
2352     return rc;
2353 }
2354 
2355 static int htab_save_complete(QEMUFile *f, void *opaque)
2356 {
2357     SpaprMachineState *spapr = opaque;
2358     int fd;
2359 
2360     /* Iteration header */
2361     if (!spapr->htab_shift) {
2362         qemu_put_be32(f, -1);
2363         return 0;
2364     } else {
2365         qemu_put_be32(f, 0);
2366     }
2367 
2368     if (!spapr->htab) {
2369         int rc;
2370 
2371         assert(kvm_enabled());
2372 
2373         fd = get_htab_fd(spapr);
2374         if (fd < 0) {
2375             return fd;
2376         }
2377 
2378         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2379         if (rc < 0) {
2380             return rc;
2381         }
2382     } else {
2383         if (spapr->htab_first_pass) {
2384             htab_save_first_pass(f, spapr, -1);
2385         }
2386         htab_save_later_pass(f, spapr, -1);
2387     }
2388 
2389     /* End marker */
2390     htab_save_end_marker(f);
2391 
2392     return 0;
2393 }
2394 
2395 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2396 {
2397     SpaprMachineState *spapr = opaque;
2398     uint32_t section_hdr;
2399     int fd = -1;
2400     Error *local_err = NULL;
2401 
2402     if (version_id < 1 || version_id > 1) {
2403         error_report("htab_load() bad version");
2404         return -EINVAL;
2405     }
2406 
2407     section_hdr = qemu_get_be32(f);
2408 
2409     if (section_hdr == -1) {
2410         spapr_free_hpt(spapr);
2411         return 0;
2412     }
2413 
2414     if (section_hdr) {
2415         int ret;
2416 
2417         /* First section gives the htab size */
2418         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2419         if (ret < 0) {
2420             error_report_err(local_err);
2421             return ret;
2422         }
2423         return 0;
2424     }
2425 
2426     if (!spapr->htab) {
2427         assert(kvm_enabled());
2428 
2429         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2430         if (fd < 0) {
2431             error_report_err(local_err);
2432             return fd;
2433         }
2434     }
2435 
2436     while (true) {
2437         uint32_t index;
2438         uint16_t n_valid, n_invalid;
2439 
2440         index = qemu_get_be32(f);
2441         n_valid = qemu_get_be16(f);
2442         n_invalid = qemu_get_be16(f);
2443 
2444         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2445             /* End of Stream */
2446             break;
2447         }
2448 
2449         if ((index + n_valid + n_invalid) >
2450             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2451             /* Bad index in stream */
2452             error_report(
2453                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2454                 index, n_valid, n_invalid, spapr->htab_shift);
2455             return -EINVAL;
2456         }
2457 
2458         if (spapr->htab) {
2459             if (n_valid) {
2460                 qemu_get_buffer(f, (void *)hpte_get_ptr(spapr, index),
2461                                 HASH_PTE_SIZE_64 * n_valid);
2462             }
2463             if (n_invalid) {
2464                 memset(hpte_get_ptr(spapr, index + n_valid), 0,
2465                        HASH_PTE_SIZE_64 * n_invalid);
2466             }
2467         } else {
2468             int rc;
2469 
2470             assert(fd >= 0);
2471 
2472             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2473                                         &local_err);
2474             if (rc < 0) {
2475                 error_report_err(local_err);
2476                 return rc;
2477             }
2478         }
2479     }
2480 
2481     if (!spapr->htab) {
2482         assert(fd >= 0);
2483         close(fd);
2484     }
2485 
2486     return 0;
2487 }
2488 
2489 static void htab_save_cleanup(void *opaque)
2490 {
2491     SpaprMachineState *spapr = opaque;
2492 
2493     close_htab_fd(spapr);
2494 }
2495 
2496 static SaveVMHandlers savevm_htab_handlers = {
2497     .save_setup = htab_save_setup,
2498     .save_live_iterate = htab_save_iterate,
2499     .save_live_complete_precopy = htab_save_complete,
2500     .save_cleanup = htab_save_cleanup,
2501     .load_state = htab_load,
2502 };
2503 
2504 static void spapr_boot_set(void *opaque, const char *boot_device,
2505                            Error **errp)
2506 {
2507     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2508 
2509     g_free(spapr->boot_device);
2510     spapr->boot_device = g_strdup(boot_device);
2511 }
2512 
2513 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2514 {
2515     MachineState *machine = MACHINE(spapr);
2516     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2517     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2518     int i;
2519 
2520     g_assert(!nr_lmbs || machine->device_memory);
2521     for (i = 0; i < nr_lmbs; i++) {
2522         uint64_t addr;
2523 
2524         addr = i * lmb_size + machine->device_memory->base;
2525         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2526                                addr / lmb_size);
2527     }
2528 }
2529 
2530 /*
2531  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2532  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2533  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2534  */
2535 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2536 {
2537     int i;
2538 
2539     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2540         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2541                    " is not aligned to %" PRIu64 " MiB",
2542                    machine->ram_size,
2543                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2544         return;
2545     }
2546 
2547     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2548         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2549                    " is not aligned to %" PRIu64 " MiB",
2550                    machine->ram_size,
2551                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2552         return;
2553     }
2554 
2555     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2556         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2557             error_setg(errp,
2558                        "Node %d memory size 0x%" PRIx64
2559                        " is not aligned to %" PRIu64 " MiB",
2560                        i, machine->numa_state->nodes[i].node_mem,
2561                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2562             return;
2563         }
2564     }
2565 }
2566 
2567 /* find cpu slot in machine->possible_cpus by core_id */
2568 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2569 {
2570     int index = id / ms->smp.threads;
2571 
2572     if (index >= ms->possible_cpus->len) {
2573         return NULL;
2574     }
2575     if (idx) {
2576         *idx = index;
2577     }
2578     return &ms->possible_cpus->cpus[index];
2579 }
2580 
2581 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2582 {
2583     MachineState *ms = MACHINE(spapr);
2584     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2585     Error *local_err = NULL;
2586     bool vsmt_user = !!spapr->vsmt;
2587     int kvm_smt = kvmppc_smt_threads();
2588     int ret;
2589     unsigned int smp_threads = ms->smp.threads;
2590 
2591     if (tcg_enabled()) {
2592         if (smp_threads > 1 &&
2593             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2594                                    spapr->max_compat_pvr)) {
2595             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2596             return;
2597         }
2598 
2599         if (smp_threads > 8) {
2600             error_setg(errp, "TCG cannot support more than 8 threads/core "
2601                        "on a pseries machine");
2602             return;
2603         }
2604     }
2605     if (!is_power_of_2(smp_threads)) {
2606         error_setg(errp, "Cannot support %d threads/core on a pseries "
2607                    "machine because it must be a power of 2", smp_threads);
2608         return;
2609     }
2610 
2611     /* Determine the VSMT mode to use: */
2612     if (vsmt_user) {
2613         if (spapr->vsmt < smp_threads) {
2614             error_setg(errp, "Cannot support VSMT mode %d"
2615                        " because it must be >= threads/core (%d)",
2616                        spapr->vsmt, smp_threads);
2617             return;
2618         }
2619         /* In this case, spapr->vsmt has been set by the command line */
2620     } else if (!smc->smp_threads_vsmt) {
2621         /*
2622          * Default VSMT value is tricky, because we need it to be as
2623          * consistent as possible (for migration), but this requires
2624          * changing it for at least some existing cases.  We pick 8 as
2625          * the value that we'd get with KVM on POWER8, the
2626          * overwhelmingly common case in production systems.
2627          */
2628         spapr->vsmt = MAX(8, smp_threads);
2629     } else {
2630         spapr->vsmt = smp_threads;
2631     }
2632 
2633     /* KVM: If necessary, set the SMT mode: */
2634     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2635         ret = kvmppc_set_smt_threads(spapr->vsmt);
2636         if (ret) {
2637             /* Looks like KVM isn't able to change VSMT mode */
2638             error_setg(&local_err,
2639                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2640                        spapr->vsmt, ret);
2641             /* We can live with that if the default one is big enough
2642              * for the number of threads, and a submultiple of the one
2643              * we want.  In this case we'll waste some vcpu ids, but
2644              * behaviour will be correct */
2645             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2646                 warn_report_err(local_err);
2647             } else {
2648                 if (!vsmt_user) {
2649                     error_append_hint(&local_err,
2650                                       "On PPC, a VM with %d threads/core"
2651                                       " on a host with %d threads/core"
2652                                       " requires the use of VSMT mode %d.\n",
2653                                       smp_threads, kvm_smt, spapr->vsmt);
2654                 }
2655                 kvmppc_error_append_smt_possible_hint(&local_err);
2656                 error_propagate(errp, local_err);
2657             }
2658         }
2659     }
2660     /* else TCG: nothing to do currently */
2661 }
2662 
2663 static void spapr_init_cpus(SpaprMachineState *spapr)
2664 {
2665     MachineState *machine = MACHINE(spapr);
2666     MachineClass *mc = MACHINE_GET_CLASS(machine);
2667     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2668     const CPUArchIdList *possible_cpus;
2669     unsigned int smp_cpus = machine->smp.cpus;
2670     unsigned int smp_threads = machine->smp.threads;
2671     unsigned int max_cpus = machine->smp.max_cpus;
2672     int boot_cores_nr = smp_cpus / smp_threads;
2673     int i;
2674 
2675     possible_cpus = mc->possible_cpu_arch_ids(machine);
2676     if (mc->has_hotpluggable_cpus) {
2677         if (smp_cpus % smp_threads) {
2678             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2679                          smp_cpus, smp_threads);
2680             exit(1);
2681         }
2682         if (max_cpus % smp_threads) {
2683             error_report("max_cpus (%u) must be multiple of threads (%u)",
2684                          max_cpus, smp_threads);
2685             exit(1);
2686         }
2687     } else {
2688         if (max_cpus != smp_cpus) {
2689             error_report("This machine version does not support CPU hotplug");
2690             exit(1);
2691         }
2692         boot_cores_nr = possible_cpus->len;
2693     }
2694 
2695     for (i = 0; i < possible_cpus->len; i++) {
2696         int core_id = i * smp_threads;
2697 
2698         if (mc->has_hotpluggable_cpus) {
2699             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2700                                    spapr_vcpu_id(spapr, core_id));
2701         }
2702 
2703         if (i < boot_cores_nr) {
2704             Object *core  = object_new(type);
2705             int nr_threads = smp_threads;
2706 
2707             /* Handle the partially filled core for older machine types */
2708             if ((i + 1) * smp_threads >= smp_cpus) {
2709                 nr_threads = smp_cpus - i * smp_threads;
2710             }
2711 
2712             object_property_set_int(core, "nr-threads", nr_threads,
2713                                     &error_fatal);
2714             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2715                                     &error_fatal);
2716             qdev_realize(DEVICE(core), NULL, &error_fatal);
2717 
2718             object_unref(core);
2719         }
2720     }
2721 }
2722 
2723 static PCIHostState *spapr_create_default_phb(void)
2724 {
2725     DeviceState *dev;
2726 
2727     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2728     qdev_prop_set_uint32(dev, "index", 0);
2729     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2730 
2731     return PCI_HOST_BRIDGE(dev);
2732 }
2733 
2734 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2735 {
2736     MachineState *machine = MACHINE(spapr);
2737     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2738     hwaddr rma_size = machine->ram_size;
2739     hwaddr node0_size = spapr_node0_size(machine);
2740 
2741     /* RMA has to fit in the first NUMA node */
2742     rma_size = MIN(rma_size, node0_size);
2743 
2744     /*
2745      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2746      * never exceed that
2747      */
2748     rma_size = MIN(rma_size, 1 * TiB);
2749 
2750     /*
2751      * Clamp the RMA size based on machine type.  This is for
2752      * migration compatibility with older qemu versions, which limited
2753      * the RMA size for complicated and mostly bad reasons.
2754      */
2755     if (smc->rma_limit) {
2756         rma_size = MIN(rma_size, smc->rma_limit);
2757     }
2758 
2759     if (rma_size < MIN_RMA_SLOF) {
2760         error_setg(errp,
2761                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2762                    "ldMiB guest RMA (Real Mode Area memory)",
2763                    MIN_RMA_SLOF / MiB);
2764         return 0;
2765     }
2766 
2767     return rma_size;
2768 }
2769 
2770 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2771 {
2772     MachineState *machine = MACHINE(spapr);
2773     int i;
2774 
2775     for (i = 0; i < machine->ram_slots; i++) {
2776         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2777     }
2778 }
2779 
2780 /* pSeries LPAR / sPAPR hardware init */
2781 static void spapr_machine_init(MachineState *machine)
2782 {
2783     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2784     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2785     MachineClass *mc = MACHINE_GET_CLASS(machine);
2786     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2787     const char *bios_name = machine->firmware ?: bios_default;
2788     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2789     const char *kernel_filename = machine->kernel_filename;
2790     const char *initrd_filename = machine->initrd_filename;
2791     PCIHostState *phb;
2792     bool has_vga;
2793     int i;
2794     MemoryRegion *sysmem = get_system_memory();
2795     long load_limit, fw_size;
2796     Error *resize_hpt_err = NULL;
2797     NICInfo *nd;
2798 
2799     if (!filename) {
2800         error_report("Could not find LPAR firmware '%s'", bios_name);
2801         exit(1);
2802     }
2803     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2804     if (fw_size <= 0) {
2805         error_report("Could not load LPAR firmware '%s'", filename);
2806         exit(1);
2807     }
2808 
2809     /*
2810      * if Secure VM (PEF) support is configured, then initialize it
2811      */
2812     if (machine->cgs) {
2813         confidential_guest_kvm_init(machine->cgs, &error_fatal);
2814     }
2815 
2816     msi_nonbroken = true;
2817 
2818     QLIST_INIT(&spapr->phbs);
2819     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2820 
2821     /* Determine capabilities to run with */
2822     spapr_caps_init(spapr);
2823 
2824     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2825     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2826         /*
2827          * If the user explicitly requested a mode we should either
2828          * supply it, or fail completely (which we do below).  But if
2829          * it's not set explicitly, we reset our mode to something
2830          * that works
2831          */
2832         if (resize_hpt_err) {
2833             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2834             error_free(resize_hpt_err);
2835             resize_hpt_err = NULL;
2836         } else {
2837             spapr->resize_hpt = smc->resize_hpt_default;
2838         }
2839     }
2840 
2841     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2842 
2843     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2844         /*
2845          * User requested HPT resize, but this host can't supply it.  Bail out
2846          */
2847         error_report_err(resize_hpt_err);
2848         exit(1);
2849     }
2850     error_free(resize_hpt_err);
2851 
2852     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2853 
2854     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2855     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2856 
2857     /*
2858      * VSMT must be set in order to be able to compute VCPU ids, ie to
2859      * call spapr_max_server_number() or spapr_vcpu_id().
2860      */
2861     spapr_set_vsmt_mode(spapr, &error_fatal);
2862 
2863     /* Set up Interrupt Controller before we create the VCPUs */
2864     spapr_irq_init(spapr, &error_fatal);
2865 
2866     /* Set up containers for ibm,client-architecture-support negotiated options
2867      */
2868     spapr->ov5 = spapr_ovec_new();
2869     spapr->ov5_cas = spapr_ovec_new();
2870 
2871     spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2872     spapr_validate_node_memory(machine, &error_fatal);
2873 
2874     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2875 
2876     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2877     if (!smc->pre_6_2_numa_affinity) {
2878         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2879     }
2880 
2881     /* advertise support for dedicated HP event source to guests */
2882     if (spapr->use_hotplug_event_source) {
2883         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2884     }
2885 
2886     /* advertise support for HPT resizing */
2887     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2888         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2889     }
2890 
2891     /* advertise support for ibm,dyamic-memory-v2 */
2892     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2893 
2894     /* advertise XIVE on POWER9 machines */
2895     if (spapr->irq->xive) {
2896         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2897     }
2898 
2899     /* init CPUs */
2900     spapr_init_cpus(spapr);
2901 
2902     /* Init numa_assoc_array */
2903     spapr_numa_associativity_init(spapr, machine);
2904 
2905     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2906         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2907                               spapr->max_compat_pvr)) {
2908         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2909         /* KVM and TCG always allow GTSE with radix... */
2910         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2911     }
2912     /* ... but not with hash (currently). */
2913 
2914     if (kvm_enabled()) {
2915         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2916         kvmppc_enable_logical_ci_hcalls();
2917         kvmppc_enable_set_mode_hcall();
2918 
2919         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2920         kvmppc_enable_clear_ref_mod_hcalls();
2921 
2922         /* Enable H_PAGE_INIT */
2923         kvmppc_enable_h_page_init();
2924     }
2925 
2926     /* map RAM */
2927     memory_region_add_subregion(sysmem, 0, machine->ram);
2928 
2929     /* initialize hotplug memory address space */
2930     if (machine->ram_size < machine->maxram_size) {
2931         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2932         hwaddr device_mem_base;
2933 
2934         /*
2935          * Limit the number of hotpluggable memory slots to half the number
2936          * slots that KVM supports, leaving the other half for PCI and other
2937          * devices. However ensure that number of slots doesn't drop below 32.
2938          */
2939         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2940                            SPAPR_MAX_RAM_SLOTS;
2941 
2942         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2943             max_memslots = SPAPR_MAX_RAM_SLOTS;
2944         }
2945         if (machine->ram_slots > max_memslots) {
2946             error_report("Specified number of memory slots %"
2947                          PRIu64" exceeds max supported %d",
2948                          machine->ram_slots, max_memslots);
2949             exit(1);
2950         }
2951 
2952         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
2953         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
2954     }
2955 
2956     spapr_create_lmb_dr_connectors(spapr);
2957 
2958     if (mc->nvdimm_supported) {
2959         spapr_create_nvdimm_dr_connectors(spapr);
2960     }
2961 
2962     /* Set up RTAS event infrastructure */
2963     spapr_events_init(spapr);
2964 
2965     /* Set up the RTC RTAS interfaces */
2966     spapr_rtc_create(spapr);
2967 
2968     /* Set up VIO bus */
2969     spapr->vio_bus = spapr_vio_bus_init();
2970 
2971     for (i = 0; serial_hd(i); i++) {
2972         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2973     }
2974 
2975     /* We always have at least the nvram device on VIO */
2976     spapr_create_nvram(spapr);
2977 
2978     /*
2979      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2980      * connectors (described in root DT node's "ibm,drc-types" property)
2981      * are pre-initialized here. additional child connectors (such as
2982      * connectors for a PHBs PCI slots) are added as needed during their
2983      * parent's realization.
2984      */
2985     if (smc->dr_phb_enabled) {
2986         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2987             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2988         }
2989     }
2990 
2991     /* Set up PCI */
2992     spapr_pci_rtas_init();
2993 
2994     phb = spapr_create_default_phb();
2995 
2996     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
2997         spapr_vlan_create(spapr->vio_bus, nd);
2998     }
2999 
3000     pci_init_nic_devices(phb->bus, NULL);
3001 
3002     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3003         spapr_vscsi_create(spapr->vio_bus);
3004     }
3005 
3006     /* Graphics */
3007     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3008     if (has_vga) {
3009         spapr->want_stdout_path = !machine->enable_graphics;
3010         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3011     } else {
3012         spapr->want_stdout_path = true;
3013     }
3014 
3015     if (machine->usb) {
3016         pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3017 
3018         if (has_vga) {
3019             USBBus *usb_bus;
3020 
3021             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3022                                                               &error_abort));
3023             usb_create_simple(usb_bus, "usb-kbd");
3024             usb_create_simple(usb_bus, "usb-mouse");
3025         }
3026     }
3027 
3028     if (kernel_filename) {
3029         uint64_t loaded_addr = 0;
3030 
3031         spapr->kernel_size = load_elf(kernel_filename, NULL,
3032                                       translate_kernel_address, spapr,
3033                                       NULL, &loaded_addr, NULL, NULL,
3034                                       ELFDATA2MSB, PPC_ELF_MACHINE, 0, 0);
3035         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3036             spapr->kernel_size = load_elf(kernel_filename, NULL,
3037                                           translate_kernel_address, spapr,
3038                                           NULL, &loaded_addr, NULL, NULL,
3039                                           ELFDATA2LSB, PPC_ELF_MACHINE, 0, 0);
3040             spapr->kernel_le = spapr->kernel_size > 0;
3041         }
3042         if (spapr->kernel_size < 0) {
3043             error_report("error loading %s: %s", kernel_filename,
3044                          load_elf_strerror(spapr->kernel_size));
3045             exit(1);
3046         }
3047 
3048         if (spapr->kernel_addr != loaded_addr) {
3049             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3050                         " to 0x%"PRIx64,
3051                         spapr->kernel_addr, loaded_addr);
3052             spapr->kernel_addr = loaded_addr;
3053         }
3054 
3055         /* load initrd */
3056         if (initrd_filename) {
3057             /* Try to locate the initrd in the gap between the kernel
3058              * and the firmware. Add a bit of space just in case
3059              */
3060             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3061                                   + 0x1ffff) & ~0xffff;
3062             spapr->initrd_size = load_image_targphys(initrd_filename,
3063                                                      spapr->initrd_base,
3064                                                      load_limit
3065                                                      - spapr->initrd_base);
3066             if (spapr->initrd_size < 0) {
3067                 error_report("could not load initial ram disk '%s'",
3068                              initrd_filename);
3069                 exit(1);
3070             }
3071         }
3072     }
3073 
3074     /* FIXME: Should register things through the MachineState's qdev
3075      * interface, this is a legacy from the sPAPREnvironment structure
3076      * which predated MachineState but had a similar function */
3077     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3078     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3079                          &savevm_htab_handlers, spapr);
3080 
3081     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3082 
3083     qemu_register_boot_set(spapr_boot_set, spapr);
3084 
3085     /*
3086      * Nothing needs to be done to resume a suspended guest because
3087      * suspending does not change the machine state, so no need for
3088      * a ->wakeup method.
3089      */
3090     qemu_register_wakeup_support();
3091 
3092     if (kvm_enabled()) {
3093         /* to stop and start vmclock */
3094         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3095                                          &spapr->tb);
3096 
3097         kvmppc_spapr_enable_inkernel_multitce();
3098     }
3099 
3100     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3101     if (spapr->vof) {
3102         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3103         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3104     }
3105 
3106     spapr_watchdog_init(spapr);
3107 }
3108 
3109 #define DEFAULT_KVM_TYPE "auto"
3110 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3111 {
3112     /*
3113      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3114      * accommodate the 'HV' and 'PV' formats that exists in the
3115      * wild. The 'auto' mode is being introduced already as
3116      * lower-case, thus we don't need to bother checking for
3117      * "AUTO".
3118      */
3119     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3120         return 0;
3121     }
3122 
3123     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3124         return 1;
3125     }
3126 
3127     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3128         return 2;
3129     }
3130 
3131     error_report("Unknown kvm-type specified '%s'", vm_type);
3132     return -1;
3133 }
3134 
3135 /*
3136  * Implementation of an interface to adjust firmware path
3137  * for the bootindex property handling.
3138  */
3139 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3140                                    DeviceState *dev)
3141 {
3142 #define CAST(type, obj, name) \
3143     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3144     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3145     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3146     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3147     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3148 
3149     if (d && bus) {
3150         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3151         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3152         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3153 
3154         if (spapr) {
3155             /*
3156              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3157              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3158              * 0x8000 | (target << 8) | (bus << 5) | lun
3159              * (see the "Logical unit addressing format" table in SAM5)
3160              */
3161             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3162             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3163                                    (uint64_t)id << 48);
3164         } else if (virtio) {
3165             /*
3166              * We use SRP luns of the form 01000000 | (target << 8) | lun
3167              * in the top 32 bits of the 64-bit LUN
3168              * Note: the quote above is from SLOF and it is wrong,
3169              * the actual binding is:
3170              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3171              */
3172             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3173             if (d->lun >= 256) {
3174                 /* Use the LUN "flat space addressing method" */
3175                 id |= 0x4000;
3176             }
3177             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3178                                    (uint64_t)id << 32);
3179         } else if (usb) {
3180             /*
3181              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3182              * in the top 32 bits of the 64-bit LUN
3183              */
3184             unsigned usb_port = atoi(usb->port->path);
3185             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3186             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3187                                    (uint64_t)id << 32);
3188         }
3189     }
3190 
3191     /*
3192      * SLOF probes the USB devices, and if it recognizes that the device is a
3193      * storage device, it changes its name to "storage" instead of "usb-host",
3194      * and additionally adds a child node for the SCSI LUN, so the correct
3195      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3196      */
3197     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3198         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3199         if (usb_device_is_scsi_storage(usbdev)) {
3200             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3201         }
3202     }
3203 
3204     if (phb) {
3205         /* Replace "pci" with "pci@800000020000000" */
3206         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3207     }
3208 
3209     if (vsc) {
3210         /* Same logic as virtio above */
3211         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3212         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3213     }
3214 
3215     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3216         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3217         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3218         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3219     }
3220 
3221     if (pcidev) {
3222         return spapr_pci_fw_dev_name(pcidev);
3223     }
3224 
3225     return NULL;
3226 }
3227 
3228 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3229 {
3230     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3231 
3232     return g_strdup(spapr->kvm_type);
3233 }
3234 
3235 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3236 {
3237     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3238 
3239     g_free(spapr->kvm_type);
3240     spapr->kvm_type = g_strdup(value);
3241 }
3242 
3243 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3244 {
3245     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3246 
3247     return spapr->use_hotplug_event_source;
3248 }
3249 
3250 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3251                                             Error **errp)
3252 {
3253     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3254 
3255     spapr->use_hotplug_event_source = value;
3256 }
3257 
3258 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3259 {
3260     return true;
3261 }
3262 
3263 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3264 {
3265     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3266 
3267     switch (spapr->resize_hpt) {
3268     case SPAPR_RESIZE_HPT_DEFAULT:
3269         return g_strdup("default");
3270     case SPAPR_RESIZE_HPT_DISABLED:
3271         return g_strdup("disabled");
3272     case SPAPR_RESIZE_HPT_ENABLED:
3273         return g_strdup("enabled");
3274     case SPAPR_RESIZE_HPT_REQUIRED:
3275         return g_strdup("required");
3276     }
3277     g_assert_not_reached();
3278 }
3279 
3280 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     if (strcmp(value, "default") == 0) {
3285         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3286     } else if (strcmp(value, "disabled") == 0) {
3287         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3288     } else if (strcmp(value, "enabled") == 0) {
3289         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3290     } else if (strcmp(value, "required") == 0) {
3291         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3292     } else {
3293         error_setg(errp, "Bad value for \"resize-hpt\" property");
3294     }
3295 }
3296 
3297 static bool spapr_get_vof(Object *obj, Error **errp)
3298 {
3299     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3300 
3301     return spapr->vof != NULL;
3302 }
3303 
3304 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3305 {
3306     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3307 
3308     if (spapr->vof) {
3309         vof_cleanup(spapr->vof);
3310         g_free(spapr->vof);
3311         spapr->vof = NULL;
3312     }
3313     if (!value) {
3314         return;
3315     }
3316     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3317 }
3318 
3319 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3320 {
3321     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3322 
3323     if (spapr->irq == &spapr_irq_xics_legacy) {
3324         return g_strdup("legacy");
3325     } else if (spapr->irq == &spapr_irq_xics) {
3326         return g_strdup("xics");
3327     } else if (spapr->irq == &spapr_irq_xive) {
3328         return g_strdup("xive");
3329     } else if (spapr->irq == &spapr_irq_dual) {
3330         return g_strdup("dual");
3331     }
3332     g_assert_not_reached();
3333 }
3334 
3335 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3336 {
3337     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3338 
3339     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3340         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3341         return;
3342     }
3343 
3344     /* The legacy IRQ backend can not be set */
3345     if (strcmp(value, "xics") == 0) {
3346         spapr->irq = &spapr_irq_xics;
3347     } else if (strcmp(value, "xive") == 0) {
3348         spapr->irq = &spapr_irq_xive;
3349     } else if (strcmp(value, "dual") == 0) {
3350         spapr->irq = &spapr_irq_dual;
3351     } else {
3352         error_setg(errp, "Bad value for \"ic-mode\" property");
3353     }
3354 }
3355 
3356 static char *spapr_get_host_model(Object *obj, Error **errp)
3357 {
3358     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3359 
3360     return g_strdup(spapr->host_model);
3361 }
3362 
3363 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3364 {
3365     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3366 
3367     g_free(spapr->host_model);
3368     spapr->host_model = g_strdup(value);
3369 }
3370 
3371 static char *spapr_get_host_serial(Object *obj, Error **errp)
3372 {
3373     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3374 
3375     return g_strdup(spapr->host_serial);
3376 }
3377 
3378 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3379 {
3380     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3381 
3382     g_free(spapr->host_serial);
3383     spapr->host_serial = g_strdup(value);
3384 }
3385 
3386 static void spapr_instance_init(Object *obj)
3387 {
3388     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3389     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3390     MachineState *ms = MACHINE(spapr);
3391     MachineClass *mc = MACHINE_GET_CLASS(ms);
3392 
3393     /*
3394      * NVDIMM support went live in 5.1 without considering that, in
3395      * other archs, the user needs to enable NVDIMM support with the
3396      * 'nvdimm' machine option and the default behavior is NVDIMM
3397      * support disabled. It is too late to roll back to the standard
3398      * behavior without breaking 5.1 guests.
3399      */
3400     if (mc->nvdimm_supported) {
3401         ms->nvdimms_state->is_enabled = true;
3402     }
3403 
3404     spapr->htab_fd = -1;
3405     spapr->use_hotplug_event_source = true;
3406     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3407     object_property_add_str(obj, "kvm-type",
3408                             spapr_get_kvm_type, spapr_set_kvm_type);
3409     object_property_set_description(obj, "kvm-type",
3410                                     "Specifies the KVM virtualization mode (auto,"
3411                                     " hv, pr). Defaults to 'auto'. This mode will use"
3412                                     " any available KVM module loaded in the host,"
3413                                     " where kvm_hv takes precedence if both kvm_hv and"
3414                                     " kvm_pr are loaded.");
3415     object_property_add_bool(obj, "modern-hotplug-events",
3416                             spapr_get_modern_hotplug_events,
3417                             spapr_set_modern_hotplug_events);
3418     object_property_set_description(obj, "modern-hotplug-events",
3419                                     "Use dedicated hotplug event mechanism in"
3420                                     " place of standard EPOW events when possible"
3421                                     " (required for memory hot-unplug support)");
3422     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3423                             "Maximum permitted CPU compatibility mode");
3424 
3425     object_property_add_str(obj, "resize-hpt",
3426                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3427     object_property_set_description(obj, "resize-hpt",
3428                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3429     object_property_add_uint32_ptr(obj, "vsmt",
3430                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3431     object_property_set_description(obj, "vsmt",
3432                                     "Virtual SMT: KVM behaves as if this were"
3433                                     " the host's SMT mode");
3434 
3435     object_property_add_bool(obj, "vfio-no-msix-emulation",
3436                              spapr_get_msix_emulation, NULL);
3437 
3438     object_property_add_uint64_ptr(obj, "kernel-addr",
3439                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3440     object_property_set_description(obj, "kernel-addr",
3441                                     stringify(KERNEL_LOAD_ADDR)
3442                                     " for -kernel is the default");
3443     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3444 
3445     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3446     object_property_set_description(obj, "x-vof",
3447                                     "Enable Virtual Open Firmware (experimental)");
3448 
3449     /* The machine class defines the default interrupt controller mode */
3450     spapr->irq = smc->irq;
3451     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3452                             spapr_set_ic_mode);
3453     object_property_set_description(obj, "ic-mode",
3454                  "Specifies the interrupt controller mode (xics, xive, dual)");
3455 
3456     object_property_add_str(obj, "host-model",
3457         spapr_get_host_model, spapr_set_host_model);
3458     object_property_set_description(obj, "host-model",
3459         "Host model to advertise in guest device tree");
3460     object_property_add_str(obj, "host-serial",
3461         spapr_get_host_serial, spapr_set_host_serial);
3462     object_property_set_description(obj, "host-serial",
3463         "Host serial number to advertise in guest device tree");
3464 }
3465 
3466 static void spapr_machine_finalizefn(Object *obj)
3467 {
3468     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3469 
3470     g_free(spapr->kvm_type);
3471 }
3472 
3473 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3474 {
3475     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3476     CPUPPCState *env = cpu_env(cs);
3477 
3478     cpu_synchronize_state(cs);
3479     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3480     if (spapr->fwnmi_system_reset_addr != -1) {
3481         uint64_t rtas_addr, addr;
3482 
3483         /* get rtas addr from fdt */
3484         rtas_addr = spapr_get_rtas_addr();
3485         if (!rtas_addr) {
3486             qemu_system_guest_panicked(NULL);
3487             return;
3488         }
3489 
3490         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3491         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3492         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3493         env->gpr[3] = addr;
3494     }
3495     ppc_cpu_do_system_reset(cs);
3496     if (spapr->fwnmi_system_reset_addr != -1) {
3497         env->nip = spapr->fwnmi_system_reset_addr;
3498     }
3499 }
3500 
3501 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3502 {
3503     CPUState *cs;
3504 
3505     CPU_FOREACH(cs) {
3506         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3507     }
3508 }
3509 
3510 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3511                           void *fdt, int *fdt_start_offset, Error **errp)
3512 {
3513     uint64_t addr;
3514     uint32_t node;
3515 
3516     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3517     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3518                                     &error_abort);
3519     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3520                                              SPAPR_MEMORY_BLOCK_SIZE);
3521     return 0;
3522 }
3523 
3524 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3525                            bool dedicated_hp_event_source)
3526 {
3527     SpaprDrc *drc;
3528     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3529     int i;
3530     uint64_t addr = addr_start;
3531     bool hotplugged = spapr_drc_hotplugged(dev);
3532 
3533     for (i = 0; i < nr_lmbs; i++) {
3534         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3535                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3536         g_assert(drc);
3537 
3538         /*
3539          * memory_device_get_free_addr() provided a range of free addresses
3540          * that doesn't overlap with any existing mapping at pre-plug. The
3541          * corresponding LMB DRCs are thus assumed to be all attachable.
3542          */
3543         spapr_drc_attach(drc, dev);
3544         if (!hotplugged) {
3545             spapr_drc_reset(drc);
3546         }
3547         addr += SPAPR_MEMORY_BLOCK_SIZE;
3548     }
3549     /* send hotplug notification to the
3550      * guest only in case of hotplugged memory
3551      */
3552     if (hotplugged) {
3553         if (dedicated_hp_event_source) {
3554             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3555                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3556             g_assert(drc);
3557             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3558                                                    nr_lmbs,
3559                                                    spapr_drc_index(drc));
3560         } else {
3561             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3562                                            nr_lmbs);
3563         }
3564     }
3565 }
3566 
3567 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3568 {
3569     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3570     PCDIMMDevice *dimm = PC_DIMM(dev);
3571     uint64_t size, addr;
3572     int64_t slot;
3573     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3574 
3575     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3576 
3577     pc_dimm_plug(dimm, MACHINE(ms));
3578 
3579     if (!is_nvdimm) {
3580         addr = object_property_get_uint(OBJECT(dimm),
3581                                         PC_DIMM_ADDR_PROP, &error_abort);
3582         spapr_add_lmbs(dev, addr, size,
3583                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3584     } else {
3585         slot = object_property_get_int(OBJECT(dimm),
3586                                        PC_DIMM_SLOT_PROP, &error_abort);
3587         /* We should have valid slot number at this point */
3588         g_assert(slot >= 0);
3589         spapr_add_nvdimm(dev, slot);
3590     }
3591 }
3592 
3593 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3594                                   Error **errp)
3595 {
3596     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3597     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3598     PCDIMMDevice *dimm = PC_DIMM(dev);
3599     Error *local_err = NULL;
3600     uint64_t size;
3601     Object *memdev;
3602     hwaddr pagesize;
3603 
3604     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3605     if (local_err) {
3606         error_propagate(errp, local_err);
3607         return;
3608     }
3609 
3610     if (is_nvdimm) {
3611         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3612             return;
3613         }
3614     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3615         error_setg(errp, "Hotplugged memory size must be a multiple of "
3616                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3617         return;
3618     }
3619 
3620     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3621                                       &error_abort);
3622     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3623     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3624         return;
3625     }
3626 
3627     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), errp);
3628 }
3629 
3630 struct SpaprDimmState {
3631     PCDIMMDevice *dimm;
3632     uint32_t nr_lmbs;
3633     QTAILQ_ENTRY(SpaprDimmState) next;
3634 };
3635 
3636 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3637                                                        PCDIMMDevice *dimm)
3638 {
3639     SpaprDimmState *dimm_state = NULL;
3640 
3641     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3642         if (dimm_state->dimm == dimm) {
3643             break;
3644         }
3645     }
3646     return dimm_state;
3647 }
3648 
3649 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3650                                                       uint32_t nr_lmbs,
3651                                                       PCDIMMDevice *dimm)
3652 {
3653     SpaprDimmState *ds = NULL;
3654 
3655     /*
3656      * If this request is for a DIMM whose removal had failed earlier
3657      * (due to guest's refusal to remove the LMBs), we would have this
3658      * dimm already in the pending_dimm_unplugs list. In that
3659      * case don't add again.
3660      */
3661     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3662     if (!ds) {
3663         ds = g_new0(SpaprDimmState, 1);
3664         ds->nr_lmbs = nr_lmbs;
3665         ds->dimm = dimm;
3666         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3667     }
3668     return ds;
3669 }
3670 
3671 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3672                                               SpaprDimmState *dimm_state)
3673 {
3674     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3675     g_free(dimm_state);
3676 }
3677 
3678 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3679                                                         PCDIMMDevice *dimm)
3680 {
3681     SpaprDrc *drc;
3682     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3683                                                   &error_abort);
3684     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3685     uint32_t avail_lmbs = 0;
3686     uint64_t addr_start, addr;
3687     int i;
3688 
3689     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3690                                           &error_abort);
3691 
3692     addr = addr_start;
3693     for (i = 0; i < nr_lmbs; i++) {
3694         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3695                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3696         g_assert(drc);
3697         if (drc->dev) {
3698             avail_lmbs++;
3699         }
3700         addr += SPAPR_MEMORY_BLOCK_SIZE;
3701     }
3702 
3703     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3704 }
3705 
3706 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3707 {
3708     SpaprDimmState *ds;
3709     PCDIMMDevice *dimm;
3710     SpaprDrc *drc;
3711     uint32_t nr_lmbs;
3712     uint64_t size, addr_start, addr;
3713     int i;
3714 
3715     if (!dev) {
3716         return;
3717     }
3718 
3719     dimm = PC_DIMM(dev);
3720     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3721 
3722     /*
3723      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3724      * unplug state, but one of its DRC is marked as unplug_requested.
3725      * This is bad and weird enough to g_assert() out.
3726      */
3727     g_assert(ds);
3728 
3729     spapr_pending_dimm_unplugs_remove(spapr, ds);
3730 
3731     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3732     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3733 
3734     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3735                                           &error_abort);
3736 
3737     addr = addr_start;
3738     for (i = 0; i < nr_lmbs; i++) {
3739         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3740                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3741         g_assert(drc);
3742 
3743         drc->unplug_requested = false;
3744         addr += SPAPR_MEMORY_BLOCK_SIZE;
3745     }
3746 
3747     /*
3748      * Tell QAPI that something happened and the memory
3749      * hotunplug wasn't successful.
3750      */
3751     qapi_event_send_device_unplug_guest_error(dev->id,
3752                                               dev->canonical_path);
3753 }
3754 
3755 /* Callback to be called during DRC release. */
3756 void spapr_lmb_release(DeviceState *dev)
3757 {
3758     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3759     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3760     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3761 
3762     /* This information will get lost if a migration occurs
3763      * during the unplug process. In this case recover it. */
3764     if (ds == NULL) {
3765         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3766         g_assert(ds);
3767         /* The DRC being examined by the caller at least must be counted */
3768         g_assert(ds->nr_lmbs);
3769     }
3770 
3771     if (--ds->nr_lmbs) {
3772         return;
3773     }
3774 
3775     /*
3776      * Now that all the LMBs have been removed by the guest, call the
3777      * unplug handler chain. This can never fail.
3778      */
3779     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3780     object_unparent(OBJECT(dev));
3781 }
3782 
3783 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3784 {
3785     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3786     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3787 
3788     /* We really shouldn't get this far without anything to unplug */
3789     g_assert(ds);
3790 
3791     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3792     qdev_unrealize(dev);
3793     spapr_pending_dimm_unplugs_remove(spapr, ds);
3794 }
3795 
3796 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3797                                         DeviceState *dev, Error **errp)
3798 {
3799     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3800     PCDIMMDevice *dimm = PC_DIMM(dev);
3801     uint32_t nr_lmbs;
3802     uint64_t size, addr_start, addr;
3803     int i;
3804     SpaprDrc *drc;
3805 
3806     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3807         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3808         return;
3809     }
3810 
3811     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3812     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3813 
3814     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3815                                           &error_abort);
3816 
3817     /*
3818      * An existing pending dimm state for this DIMM means that there is an
3819      * unplug operation in progress, waiting for the spapr_lmb_release
3820      * callback to complete the job (BQL can't cover that far). In this case,
3821      * bail out to avoid detaching DRCs that were already released.
3822      */
3823     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3824         error_setg(errp, "Memory unplug already in progress for device %s",
3825                    dev->id);
3826         return;
3827     }
3828 
3829     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3830 
3831     addr = addr_start;
3832     for (i = 0; i < nr_lmbs; i++) {
3833         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3834                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3835         g_assert(drc);
3836 
3837         spapr_drc_unplug_request(drc);
3838         addr += SPAPR_MEMORY_BLOCK_SIZE;
3839     }
3840 
3841     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3842                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3843     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3844                                               nr_lmbs, spapr_drc_index(drc));
3845 }
3846 
3847 /* Callback to be called during DRC release. */
3848 void spapr_core_release(DeviceState *dev)
3849 {
3850     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3851 
3852     /* Call the unplug handler chain. This can never fail. */
3853     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3854     object_unparent(OBJECT(dev));
3855 }
3856 
3857 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3858 {
3859     MachineState *ms = MACHINE(hotplug_dev);
3860     CPUCore *cc = CPU_CORE(dev);
3861     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3862 
3863     assert(core_slot);
3864     core_slot->cpu = NULL;
3865     qdev_unrealize(dev);
3866 }
3867 
3868 static
3869 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3870                                Error **errp)
3871 {
3872     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3873     int index;
3874     SpaprDrc *drc;
3875     CPUCore *cc = CPU_CORE(dev);
3876 
3877     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3878         error_setg(errp, "Unable to find CPU core with core-id: %d",
3879                    cc->core_id);
3880         return;
3881     }
3882     if (index == 0) {
3883         error_setg(errp, "Boot CPU core may not be unplugged");
3884         return;
3885     }
3886 
3887     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3888                           spapr_vcpu_id(spapr, cc->core_id));
3889     g_assert(drc);
3890 
3891     if (!spapr_drc_unplug_requested(drc)) {
3892         spapr_drc_unplug_request(drc);
3893     }
3894 
3895     /*
3896      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3897      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3898      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3899      * attempt (e.g. the kernel will refuse to remove the last online
3900      * CPU), we will never attempt it again because unplug_requested
3901      * will still be 'true' in that case.
3902      */
3903     spapr_hotplug_req_remove_by_index(drc);
3904 }
3905 
3906 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3907                            void *fdt, int *fdt_start_offset, Error **errp)
3908 {
3909     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3910     CPUState *cs = CPU(core->threads[0]);
3911     PowerPCCPU *cpu = POWERPC_CPU(cs);
3912     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3913     int id = spapr_get_vcpu_id(cpu);
3914     g_autofree char *nodename = NULL;
3915     int offset;
3916 
3917     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3918     offset = fdt_add_subnode(fdt, 0, nodename);
3919 
3920     spapr_dt_cpu(cs, fdt, offset, spapr);
3921 
3922     /*
3923      * spapr_dt_cpu() does not fill the 'name' property in the
3924      * CPU node. The function is called during boot process, before
3925      * and after CAS, and overwriting the 'name' property written
3926      * by SLOF is not allowed.
3927      *
3928      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3929      * CPUs more compatible with the coldplugged ones, which have
3930      * the 'name' property. Linux Kernel also relies on this
3931      * property to identify CPU nodes.
3932      */
3933     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3934 
3935     *fdt_start_offset = offset;
3936     return 0;
3937 }
3938 
3939 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3940 {
3941     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3942     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3943     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3944     CPUCore *cc = CPU_CORE(dev);
3945     SpaprDrc *drc;
3946     CPUArchId *core_slot;
3947     int index;
3948     bool hotplugged = spapr_drc_hotplugged(dev);
3949     int i;
3950 
3951     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3952     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3953 
3954     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3955                           spapr_vcpu_id(spapr, cc->core_id));
3956 
3957     g_assert(drc || !mc->has_hotpluggable_cpus);
3958 
3959     if (drc) {
3960         /*
3961          * spapr_core_pre_plug() already buys us this is a brand new
3962          * core being plugged into a free slot. Nothing should already
3963          * be attached to the corresponding DRC.
3964          */
3965         spapr_drc_attach(drc, dev);
3966 
3967         if (hotplugged) {
3968             /*
3969              * Send hotplug notification interrupt to the guest only
3970              * in case of hotplugged CPUs.
3971              */
3972             spapr_hotplug_req_add_by_index(drc);
3973         } else {
3974             spapr_drc_reset(drc);
3975         }
3976     }
3977 
3978     core_slot->cpu = CPU(dev);
3979 
3980     /*
3981      * Set compatibility mode to match the boot CPU, which was either set
3982      * by the machine reset code or by CAS. This really shouldn't fail at
3983      * this point.
3984      */
3985     if (hotplugged) {
3986         for (i = 0; i < cc->nr_threads; i++) {
3987             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3988                            &error_abort);
3989         }
3990     }
3991 
3992 }
3993 
3994 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3995                                 Error **errp)
3996 {
3997     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3998     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3999     CPUCore *cc = CPU_CORE(dev);
4000     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4001     const char *type = object_get_typename(OBJECT(dev));
4002     CPUArchId *core_slot;
4003     int index;
4004     unsigned int smp_threads = machine->smp.threads;
4005 
4006     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4007         error_setg(errp, "CPU hotplug not supported for this machine");
4008         return;
4009     }
4010 
4011     if (strcmp(base_core_type, type)) {
4012         error_setg(errp, "CPU core type should be %s", base_core_type);
4013         return;
4014     }
4015 
4016     if (cc->core_id % smp_threads) {
4017         error_setg(errp, "invalid core id %d", cc->core_id);
4018         return;
4019     }
4020 
4021     /*
4022      * In general we should have homogeneous threads-per-core, but old
4023      * (pre hotplug support) machine types allow the last core to have
4024      * reduced threads as a compatibility hack for when we allowed
4025      * total vcpus not a multiple of threads-per-core.
4026      */
4027     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4028         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4029                    smp_threads);
4030         return;
4031     }
4032 
4033     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4034     if (!core_slot) {
4035         error_setg(errp, "core id %d out of range", cc->core_id);
4036         return;
4037     }
4038 
4039     if (core_slot->cpu) {
4040         error_setg(errp, "core %d already populated", cc->core_id);
4041         return;
4042     }
4043 
4044     numa_cpu_pre_plug(core_slot, dev, errp);
4045 }
4046 
4047 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4048                           void *fdt, int *fdt_start_offset, Error **errp)
4049 {
4050     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4051     int intc_phandle;
4052 
4053     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4054     if (intc_phandle <= 0) {
4055         return -1;
4056     }
4057 
4058     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4059         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4060         return -1;
4061     }
4062 
4063     /* generally SLOF creates these, for hotplug it's up to QEMU */
4064     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4065 
4066     return 0;
4067 }
4068 
4069 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4070                                Error **errp)
4071 {
4072     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4073     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4074     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4075     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4076     SpaprDrc *drc;
4077 
4078     if (dev->hotplugged && !smc->dr_phb_enabled) {
4079         error_setg(errp, "PHB hotplug not supported for this machine");
4080         return false;
4081     }
4082 
4083     if (sphb->index == (uint32_t)-1) {
4084         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4085         return false;
4086     }
4087 
4088     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4089     if (drc && drc->dev) {
4090         error_setg(errp, "PHB %d already attached", sphb->index);
4091         return false;
4092     }
4093 
4094     /*
4095      * This will check that sphb->index doesn't exceed the maximum number of
4096      * PHBs for the current machine type.
4097      */
4098     return
4099         smc->phb_placement(spapr, sphb->index,
4100                            &sphb->buid, &sphb->io_win_addr,
4101                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4102                            windows_supported, sphb->dma_liobn,
4103                            errp);
4104 }
4105 
4106 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4107 {
4108     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4109     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4110     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4111     SpaprDrc *drc;
4112     bool hotplugged = spapr_drc_hotplugged(dev);
4113 
4114     if (!smc->dr_phb_enabled) {
4115         return;
4116     }
4117 
4118     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4119     /* hotplug hooks should check it's enabled before getting this far */
4120     assert(drc);
4121 
4122     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4123     spapr_drc_attach(drc, dev);
4124 
4125     if (hotplugged) {
4126         spapr_hotplug_req_add_by_index(drc);
4127     } else {
4128         spapr_drc_reset(drc);
4129     }
4130 }
4131 
4132 void spapr_phb_release(DeviceState *dev)
4133 {
4134     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4135 
4136     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4137     object_unparent(OBJECT(dev));
4138 }
4139 
4140 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4141 {
4142     qdev_unrealize(dev);
4143 }
4144 
4145 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4146                                      DeviceState *dev, Error **errp)
4147 {
4148     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4149     SpaprDrc *drc;
4150 
4151     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4152     assert(drc);
4153 
4154     if (!spapr_drc_unplug_requested(drc)) {
4155         spapr_drc_unplug_request(drc);
4156         spapr_hotplug_req_remove_by_index(drc);
4157     } else {
4158         error_setg(errp,
4159                    "PCI Host Bridge unplug already in progress for device %s",
4160                    dev->id);
4161     }
4162 }
4163 
4164 static
4165 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4166                               Error **errp)
4167 {
4168     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4169 
4170     if (spapr->tpm_proxy != NULL) {
4171         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4172         return false;
4173     }
4174 
4175     return true;
4176 }
4177 
4178 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4179 {
4180     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4181     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4182 
4183     /* Already checked in spapr_tpm_proxy_pre_plug() */
4184     g_assert(spapr->tpm_proxy == NULL);
4185 
4186     spapr->tpm_proxy = tpm_proxy;
4187 }
4188 
4189 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4190 {
4191     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4192 
4193     qdev_unrealize(dev);
4194     object_unparent(OBJECT(dev));
4195     spapr->tpm_proxy = NULL;
4196 }
4197 
4198 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4199                                       DeviceState *dev, Error **errp)
4200 {
4201     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4202         spapr_memory_plug(hotplug_dev, dev);
4203     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4204         spapr_core_plug(hotplug_dev, dev);
4205     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4206         spapr_phb_plug(hotplug_dev, dev);
4207     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4208         spapr_tpm_proxy_plug(hotplug_dev, dev);
4209     }
4210 }
4211 
4212 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4213                                         DeviceState *dev, Error **errp)
4214 {
4215     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4216         spapr_memory_unplug(hotplug_dev, dev);
4217     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4218         spapr_core_unplug(hotplug_dev, dev);
4219     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4220         spapr_phb_unplug(hotplug_dev, dev);
4221     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4222         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4223     }
4224 }
4225 
4226 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4227 {
4228     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4229         /*
4230          * CAS will process all pending unplug requests.
4231          *
4232          * HACK: a guest could theoretically have cleared all bits in OV5,
4233          * but none of the guests we care for do.
4234          */
4235         spapr_ovec_empty(spapr->ov5_cas);
4236 }
4237 
4238 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4239                                                 DeviceState *dev, Error **errp)
4240 {
4241     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4242     MachineClass *mc = MACHINE_GET_CLASS(sms);
4243     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4244 
4245     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4246         if (spapr_memory_hot_unplug_supported(sms)) {
4247             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4248         } else {
4249             error_setg(errp, "Memory hot unplug not supported for this guest");
4250         }
4251     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4252         if (!mc->has_hotpluggable_cpus) {
4253             error_setg(errp, "CPU hot unplug not supported on this machine");
4254             return;
4255         }
4256         spapr_core_unplug_request(hotplug_dev, dev, errp);
4257     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4258         if (!smc->dr_phb_enabled) {
4259             error_setg(errp, "PHB hot unplug not supported on this machine");
4260             return;
4261         }
4262         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4263     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4264         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4265     }
4266 }
4267 
4268 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4269                                           DeviceState *dev, Error **errp)
4270 {
4271     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4272         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4273     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4274         spapr_core_pre_plug(hotplug_dev, dev, errp);
4275     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4276         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4277     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4278         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4279     }
4280 }
4281 
4282 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4283                                                  DeviceState *dev)
4284 {
4285     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4286         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4287         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4288         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4289         return HOTPLUG_HANDLER(machine);
4290     }
4291     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4292         PCIDevice *pcidev = PCI_DEVICE(dev);
4293         PCIBus *root = pci_device_root_bus(pcidev);
4294         SpaprPhbState *phb =
4295             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4296                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4297 
4298         if (phb) {
4299             return HOTPLUG_HANDLER(phb);
4300         }
4301     }
4302     return NULL;
4303 }
4304 
4305 static CpuInstanceProperties
4306 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4307 {
4308     CPUArchId *core_slot;
4309     MachineClass *mc = MACHINE_GET_CLASS(machine);
4310 
4311     /* make sure possible_cpu are initialized */
4312     mc->possible_cpu_arch_ids(machine);
4313     /* get CPU core slot containing thread that matches cpu_index */
4314     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4315     assert(core_slot);
4316     return core_slot->props;
4317 }
4318 
4319 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4320 {
4321     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4322 }
4323 
4324 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4325 {
4326     int i;
4327     unsigned int smp_threads = machine->smp.threads;
4328     unsigned int smp_cpus = machine->smp.cpus;
4329     const char *core_type;
4330     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4331     MachineClass *mc = MACHINE_GET_CLASS(machine);
4332 
4333     if (!mc->has_hotpluggable_cpus) {
4334         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4335     }
4336     if (machine->possible_cpus) {
4337         assert(machine->possible_cpus->len == spapr_max_cores);
4338         return machine->possible_cpus;
4339     }
4340 
4341     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4342     if (!core_type) {
4343         error_report("Unable to find sPAPR CPU Core definition");
4344         exit(1);
4345     }
4346 
4347     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4348                              sizeof(CPUArchId) * spapr_max_cores);
4349     machine->possible_cpus->len = spapr_max_cores;
4350     for (i = 0; i < machine->possible_cpus->len; i++) {
4351         int core_id = i * smp_threads;
4352 
4353         machine->possible_cpus->cpus[i].type = core_type;
4354         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4355         machine->possible_cpus->cpus[i].arch_id = core_id;
4356         machine->possible_cpus->cpus[i].props.has_core_id = true;
4357         machine->possible_cpus->cpus[i].props.core_id = core_id;
4358     }
4359     return machine->possible_cpus;
4360 }
4361 
4362 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4363                                 uint64_t *buid, hwaddr *pio,
4364                                 hwaddr *mmio32, hwaddr *mmio64,
4365                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4366 {
4367     /*
4368      * New-style PHB window placement.
4369      *
4370      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4371      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4372      * windows.
4373      *
4374      * Some guest kernels can't work with MMIO windows above 1<<46
4375      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4376      *
4377      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4378      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4379      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4380      * 1TiB 64-bit MMIO windows for each PHB.
4381      */
4382     const uint64_t base_buid = 0x800000020000000ULL;
4383     int i;
4384 
4385     /* Sanity check natural alignments */
4386     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4387     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4388     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4389     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4390     /* Sanity check bounds */
4391     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4392                       SPAPR_PCI_MEM32_WIN_SIZE);
4393     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4394                       SPAPR_PCI_MEM64_WIN_SIZE);
4395 
4396     if (index >= SPAPR_MAX_PHBS) {
4397         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4398                    SPAPR_MAX_PHBS - 1);
4399         return false;
4400     }
4401 
4402     *buid = base_buid + index;
4403     for (i = 0; i < n_dma; ++i) {
4404         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4405     }
4406 
4407     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4408     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4409     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4410     return true;
4411 }
4412 
4413 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4414 {
4415     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4416 
4417     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4418 }
4419 
4420 static void spapr_ics_resend(XICSFabric *dev)
4421 {
4422     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4423 
4424     ics_resend(spapr->ics);
4425 }
4426 
4427 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4428 {
4429     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4430 
4431     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4432 }
4433 
4434 static void spapr_pic_print_info(InterruptStatsProvider *obj, GString *buf)
4435 {
4436     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4437 
4438     spapr_irq_print_info(spapr, buf);
4439     g_string_append_printf(buf, "irqchip: %s\n",
4440                            kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4441 }
4442 
4443 /*
4444  * This is a XIVE only operation
4445  */
4446 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4447                            uint8_t nvt_blk, uint32_t nvt_idx,
4448                            bool crowd, bool cam_ignore, uint8_t priority,
4449                            uint32_t logic_serv, XiveTCTXMatch *match)
4450 {
4451     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4452     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4453     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4454     int count;
4455 
4456     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, crowd, cam_ignore,
4457                            priority, logic_serv, match);
4458     if (count < 0) {
4459         return count;
4460     }
4461 
4462     /*
4463      * When we implement the save and restore of the thread interrupt
4464      * contexts in the enter/exit CPU handlers of the machine and the
4465      * escalations in QEMU, we should be able to handle non dispatched
4466      * vCPUs.
4467      *
4468      * Until this is done, the sPAPR machine should find at least one
4469      * matching context always.
4470      */
4471     if (count == 0) {
4472         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4473                       nvt_blk, nvt_idx);
4474     }
4475 
4476     return count;
4477 }
4478 
4479 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4480 {
4481     return cpu->vcpu_id;
4482 }
4483 
4484 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4485 {
4486     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4487     MachineState *ms = MACHINE(spapr);
4488     int vcpu_id;
4489 
4490     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4491 
4492     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4493         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4494         error_append_hint(errp, "Adjust the number of cpus to %d "
4495                           "or try to raise the number of threads per core\n",
4496                           vcpu_id * ms->smp.threads / spapr->vsmt);
4497         return false;
4498     }
4499 
4500     cpu->vcpu_id = vcpu_id;
4501     return true;
4502 }
4503 
4504 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4505 {
4506     CPUState *cs;
4507 
4508     CPU_FOREACH(cs) {
4509         PowerPCCPU *cpu = POWERPC_CPU(cs);
4510 
4511         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4512             return cpu;
4513         }
4514     }
4515 
4516     return NULL;
4517 }
4518 
4519 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4520 {
4521     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4522 
4523     return spapr_cpu->in_nested;
4524 }
4525 
4526 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4527 {
4528     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4529 
4530     /* These are only called by TCG, KVM maintains dispatch state */
4531 
4532     spapr_cpu->prod = false;
4533     if (spapr_cpu->vpa_addr) {
4534         CPUState *cs = CPU(cpu);
4535         uint32_t dispatch;
4536 
4537         dispatch = ldl_be_phys(cs->as,
4538                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4539         dispatch++;
4540         if ((dispatch & 1) != 0) {
4541             qemu_log_mask(LOG_GUEST_ERROR,
4542                           "VPA: incorrect dispatch counter value for "
4543                           "dispatched partition %u, correcting.\n", dispatch);
4544             dispatch++;
4545         }
4546         stl_be_phys(cs->as,
4547                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4548     }
4549 }
4550 
4551 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4552 {
4553     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4554 
4555     if (spapr_cpu->vpa_addr) {
4556         CPUState *cs = CPU(cpu);
4557         uint32_t dispatch;
4558 
4559         dispatch = ldl_be_phys(cs->as,
4560                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4561         dispatch++;
4562         if ((dispatch & 1) != 1) {
4563             qemu_log_mask(LOG_GUEST_ERROR,
4564                           "VPA: incorrect dispatch counter value for "
4565                           "preempted partition %u, correcting.\n", dispatch);
4566             dispatch++;
4567         }
4568         stl_be_phys(cs->as,
4569                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4570     }
4571 }
4572 
4573 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4574 {
4575     MachineClass *mc = MACHINE_CLASS(oc);
4576     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4577     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4578     NMIClass *nc = NMI_CLASS(oc);
4579     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4580     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4581     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4582     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4583     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4584     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4585 
4586     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4587     mc->ignore_boot_device_suffixes = true;
4588 
4589     /*
4590      * We set up the default / latest behaviour here.  The class_init
4591      * functions for the specific versioned machine types can override
4592      * these details for backwards compatibility
4593      */
4594     mc->init = spapr_machine_init;
4595     mc->reset = spapr_machine_reset;
4596     mc->block_default_type = IF_SCSI;
4597 
4598     /*
4599      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4600      * In TCG the limit is restricted by the range of CPU IPIs available.
4601      */
4602     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4603 
4604     mc->no_parallel = 1;
4605     mc->default_boot_order = "";
4606     mc->default_ram_size = 512 * MiB;
4607     mc->default_ram_id = "ppc_spapr.ram";
4608     mc->default_display = "std";
4609     mc->kvm_type = spapr_kvm_type;
4610     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4611     mc->pci_allow_0_address = true;
4612     assert(!mc->get_hotplug_handler);
4613     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4614     hc->pre_plug = spapr_machine_device_pre_plug;
4615     hc->plug = spapr_machine_device_plug;
4616     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4617     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4618     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4619     hc->unplug_request = spapr_machine_device_unplug_request;
4620     hc->unplug = spapr_machine_device_unplug;
4621 
4622     smc->update_dt_enabled = true;
4623     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4624     mc->has_hotpluggable_cpus = true;
4625     mc->nvdimm_supported = true;
4626     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4627     fwc->get_dev_path = spapr_get_fw_dev_path;
4628     nc->nmi_monitor_handler = spapr_nmi;
4629     smc->phb_placement = spapr_phb_placement;
4630     vhc->cpu_in_nested = spapr_cpu_in_nested;
4631     vhc->deliver_hv_excp = spapr_exit_nested;
4632     vhc->hypercall = emulate_spapr_hypercall;
4633     vhc->hpt_mask = spapr_hpt_mask;
4634     vhc->map_hptes = spapr_map_hptes;
4635     vhc->unmap_hptes = spapr_unmap_hptes;
4636     vhc->hpte_set_c = spapr_hpte_set_c;
4637     vhc->hpte_set_r = spapr_hpte_set_r;
4638     vhc->get_pate = spapr_get_pate;
4639     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4640     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4641     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4642     xic->ics_get = spapr_ics_get;
4643     xic->ics_resend = spapr_ics_resend;
4644     xic->icp_get = spapr_icp_get;
4645     ispc->print_info = spapr_pic_print_info;
4646     /* Force NUMA node memory size to be a multiple of
4647      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4648      * in which LMBs are represented and hot-added
4649      */
4650     mc->numa_mem_align_shift = 28;
4651     mc->auto_enable_numa = true;
4652 
4653     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4654     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4655     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4656     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4657     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4658     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4659     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4660     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4661     smc->default_caps.caps[SPAPR_CAP_NESTED_PAPR] = SPAPR_CAP_OFF;
4662     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4663     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4664     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4665     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4666 
4667     /*
4668      * This cap specifies whether the AIL 3 mode for
4669      * H_SET_RESOURCE is supported. The default is modified
4670      * by default_caps_with_cpu().
4671      */
4672     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4673     spapr_caps_add_properties(smc);
4674     smc->irq = &spapr_irq_dual;
4675     smc->dr_phb_enabled = true;
4676     smc->linux_pci_probe = true;
4677     smc->smp_threads_vsmt = true;
4678     smc->nr_xirqs = SPAPR_NR_XIRQS;
4679     xfc->match_nvt = spapr_match_nvt;
4680     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4681     vmc->quiesce = spapr_vof_quiesce;
4682     vmc->setprop = spapr_vof_setprop;
4683 }
4684 
4685 static const TypeInfo spapr_machine_info = {
4686     .name          = TYPE_SPAPR_MACHINE,
4687     .parent        = TYPE_MACHINE,
4688     .abstract      = true,
4689     .instance_size = sizeof(SpaprMachineState),
4690     .instance_init = spapr_instance_init,
4691     .instance_finalize = spapr_machine_finalizefn,
4692     .class_size    = sizeof(SpaprMachineClass),
4693     .class_init    = spapr_machine_class_init,
4694     .interfaces = (InterfaceInfo[]) {
4695         { TYPE_FW_PATH_PROVIDER },
4696         { TYPE_NMI },
4697         { TYPE_HOTPLUG_HANDLER },
4698         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4699         { TYPE_XICS_FABRIC },
4700         { TYPE_INTERRUPT_STATS_PROVIDER },
4701         { TYPE_XIVE_FABRIC },
4702         { TYPE_VOF_MACHINE_IF },
4703         { }
4704     },
4705 };
4706 
4707 static void spapr_machine_latest_class_options(MachineClass *mc)
4708 {
4709     mc->alias = "pseries";
4710     mc->is_default = true;
4711 }
4712 
4713 #define DEFINE_SPAPR_MACHINE_IMPL(latest, ...)                       \
4714     static void MACHINE_VER_SYM(class_init, spapr, __VA_ARGS__)(     \
4715         ObjectClass *oc,                                             \
4716         void *data)                                                  \
4717     {                                                                \
4718         MachineClass *mc = MACHINE_CLASS(oc);                        \
4719         MACHINE_VER_SYM(class_options, spapr, __VA_ARGS__)(mc);      \
4720         MACHINE_VER_DEPRECATION(__VA_ARGS__);                        \
4721         if (latest) {                                                \
4722             spapr_machine_latest_class_options(mc);                  \
4723         }                                                            \
4724     }                                                                \
4725     static const TypeInfo MACHINE_VER_SYM(info, spapr, __VA_ARGS__) = \
4726     {                                                                \
4727         .name = MACHINE_VER_TYPE_NAME("pseries", __VA_ARGS__),       \
4728         .parent = TYPE_SPAPR_MACHINE,                                \
4729         .class_init = MACHINE_VER_SYM(class_init, spapr, __VA_ARGS__), \
4730     };                                                               \
4731     static void MACHINE_VER_SYM(register, spapr, __VA_ARGS__)(void)  \
4732     {                                                                \
4733         MACHINE_VER_DELETION(__VA_ARGS__);                           \
4734         type_register_static(&MACHINE_VER_SYM(info, spapr, __VA_ARGS__));   \
4735     }                                                                \
4736     type_init(MACHINE_VER_SYM(register, spapr, __VA_ARGS__))
4737 
4738 #define DEFINE_SPAPR_MACHINE_AS_LATEST(major, minor) \
4739     DEFINE_SPAPR_MACHINE_IMPL(true, major, minor)
4740 #define DEFINE_SPAPR_MACHINE(major, minor) \
4741     DEFINE_SPAPR_MACHINE_IMPL(false, major, minor)
4742 
4743 /*
4744  * pseries-10.0
4745  */
4746 static void spapr_machine_10_0_class_options(MachineClass *mc)
4747 {
4748     /* Defaults for the latest behaviour inherited from the base class */
4749 }
4750 
4751 DEFINE_SPAPR_MACHINE_AS_LATEST(10, 0);
4752 
4753 /*
4754  * pseries-9.2
4755  */
4756 static void spapr_machine_9_2_class_options(MachineClass *mc)
4757 {
4758     spapr_machine_10_0_class_options(mc);
4759     compat_props_add(mc->compat_props, hw_compat_9_2, hw_compat_9_2_len);
4760 }
4761 
4762 DEFINE_SPAPR_MACHINE(9, 2);
4763 
4764 /*
4765  * pseries-9.1
4766  */
4767 static void spapr_machine_9_1_class_options(MachineClass *mc)
4768 {
4769     spapr_machine_9_2_class_options(mc);
4770     compat_props_add(mc->compat_props, hw_compat_9_1, hw_compat_9_1_len);
4771 }
4772 
4773 DEFINE_SPAPR_MACHINE(9, 1);
4774 
4775 /*
4776  * pseries-9.0
4777  */
4778 static void spapr_machine_9_0_class_options(MachineClass *mc)
4779 {
4780     spapr_machine_9_1_class_options(mc);
4781     compat_props_add(mc->compat_props, hw_compat_9_0, hw_compat_9_0_len);
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(9, 0);
4785 
4786 /*
4787  * pseries-8.2
4788  */
4789 static void spapr_machine_8_2_class_options(MachineClass *mc)
4790 {
4791     spapr_machine_9_0_class_options(mc);
4792     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4793 }
4794 
4795 DEFINE_SPAPR_MACHINE(8, 2);
4796 
4797 /*
4798  * pseries-8.1
4799  */
4800 static void spapr_machine_8_1_class_options(MachineClass *mc)
4801 {
4802     spapr_machine_8_2_class_options(mc);
4803     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4804 }
4805 
4806 DEFINE_SPAPR_MACHINE(8, 1);
4807 
4808 /*
4809  * pseries-8.0
4810  */
4811 static void spapr_machine_8_0_class_options(MachineClass *mc)
4812 {
4813     spapr_machine_8_1_class_options(mc);
4814     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4815 }
4816 
4817 DEFINE_SPAPR_MACHINE(8, 0);
4818 
4819 /*
4820  * pseries-7.2
4821  */
4822 static void spapr_machine_7_2_class_options(MachineClass *mc)
4823 {
4824     spapr_machine_8_0_class_options(mc);
4825     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4826 }
4827 
4828 DEFINE_SPAPR_MACHINE(7, 2);
4829 
4830 /*
4831  * pseries-7.1
4832  */
4833 static void spapr_machine_7_1_class_options(MachineClass *mc)
4834 {
4835     spapr_machine_7_2_class_options(mc);
4836     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4837 }
4838 
4839 DEFINE_SPAPR_MACHINE(7, 1);
4840 
4841 /*
4842  * pseries-7.0
4843  */
4844 static void spapr_machine_7_0_class_options(MachineClass *mc)
4845 {
4846     spapr_machine_7_1_class_options(mc);
4847     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4848 }
4849 
4850 DEFINE_SPAPR_MACHINE(7, 0);
4851 
4852 /*
4853  * pseries-6.2
4854  */
4855 static void spapr_machine_6_2_class_options(MachineClass *mc)
4856 {
4857     spapr_machine_7_0_class_options(mc);
4858     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4859 }
4860 
4861 DEFINE_SPAPR_MACHINE(6, 2);
4862 
4863 /*
4864  * pseries-6.1
4865  */
4866 static void spapr_machine_6_1_class_options(MachineClass *mc)
4867 {
4868     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4869 
4870     spapr_machine_6_2_class_options(mc);
4871     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4872     smc->pre_6_2_numa_affinity = true;
4873     mc->smp_props.prefer_sockets = true;
4874 }
4875 
4876 DEFINE_SPAPR_MACHINE(6, 1);
4877 
4878 /*
4879  * pseries-6.0
4880  */
4881 static void spapr_machine_6_0_class_options(MachineClass *mc)
4882 {
4883     spapr_machine_6_1_class_options(mc);
4884     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4885 }
4886 
4887 DEFINE_SPAPR_MACHINE(6, 0);
4888 
4889 /*
4890  * pseries-5.2
4891  */
4892 static void spapr_machine_5_2_class_options(MachineClass *mc)
4893 {
4894     spapr_machine_6_0_class_options(mc);
4895     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4896 }
4897 
4898 DEFINE_SPAPR_MACHINE(5, 2);
4899 
4900 /*
4901  * pseries-5.1
4902  */
4903 static void spapr_machine_5_1_class_options(MachineClass *mc)
4904 {
4905     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4906 
4907     spapr_machine_5_2_class_options(mc);
4908     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4909     smc->pre_5_2_numa_associativity = true;
4910 }
4911 
4912 DEFINE_SPAPR_MACHINE(5, 1);
4913 
4914 /*
4915  * pseries-5.0
4916  */
4917 static void spapr_machine_5_0_class_options(MachineClass *mc)
4918 {
4919     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4920     static GlobalProperty compat[] = {
4921         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4922     };
4923 
4924     spapr_machine_5_1_class_options(mc);
4925     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4926     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4927     mc->numa_mem_supported = true;
4928     smc->pre_5_1_assoc_refpoints = true;
4929 }
4930 
4931 DEFINE_SPAPR_MACHINE(5, 0);
4932 
4933 /*
4934  * pseries-4.2
4935  */
4936 static void spapr_machine_4_2_class_options(MachineClass *mc)
4937 {
4938     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4939 
4940     spapr_machine_5_0_class_options(mc);
4941     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4942     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4943     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4944     smc->rma_limit = 16 * GiB;
4945     mc->nvdimm_supported = false;
4946 }
4947 
4948 DEFINE_SPAPR_MACHINE(4, 2);
4949 
4950 /*
4951  * pseries-4.1
4952  */
4953 static void spapr_machine_4_1_class_options(MachineClass *mc)
4954 {
4955     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4956     static GlobalProperty compat[] = {
4957         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4958         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4959     };
4960 
4961     spapr_machine_4_2_class_options(mc);
4962     smc->linux_pci_probe = false;
4963     smc->smp_threads_vsmt = false;
4964     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4965     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4966 }
4967 
4968 DEFINE_SPAPR_MACHINE(4, 1);
4969 
4970 /*
4971  * pseries-4.0
4972  */
4973 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4974                               uint64_t *buid, hwaddr *pio,
4975                               hwaddr *mmio32, hwaddr *mmio64,
4976                               unsigned n_dma, uint32_t *liobns, Error **errp)
4977 {
4978     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4979                              liobns, errp)) {
4980         return false;
4981     }
4982     return true;
4983 }
4984 static void spapr_machine_4_0_class_options(MachineClass *mc)
4985 {
4986     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4987 
4988     spapr_machine_4_1_class_options(mc);
4989     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4990     smc->phb_placement = phb_placement_4_0;
4991     smc->irq = &spapr_irq_xics;
4992     smc->pre_4_1_migration = true;
4993 }
4994 
4995 DEFINE_SPAPR_MACHINE(4, 0);
4996 
4997 /*
4998  * pseries-3.1
4999  */
5000 static void spapr_machine_3_1_class_options(MachineClass *mc)
5001 {
5002     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5003 
5004     spapr_machine_4_0_class_options(mc);
5005     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5006 
5007     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5008     smc->update_dt_enabled = false;
5009     smc->dr_phb_enabled = false;
5010     smc->broken_host_serial_model = true;
5011     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5012     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5013     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5014     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5015 }
5016 
5017 DEFINE_SPAPR_MACHINE(3, 1);
5018 
5019 /*
5020  * pseries-3.0
5021  */
5022 
5023 static void spapr_machine_3_0_class_options(MachineClass *mc)
5024 {
5025     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5026 
5027     spapr_machine_3_1_class_options(mc);
5028     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5029 
5030     smc->legacy_irq_allocation = true;
5031     smc->nr_xirqs = 0x400;
5032     smc->irq = &spapr_irq_xics_legacy;
5033 }
5034 
5035 DEFINE_SPAPR_MACHINE(3, 0);
5036 
5037 static void spapr_machine_register_types(void)
5038 {
5039     type_register_static(&spapr_machine_info);
5040 }
5041 
5042 type_init(spapr_machine_register_types)
5043