xref: /openbmc/qemu/hw/ppc/spapr.c (revision 3df4c288)
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  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu/datadir.h"
29 #include "qemu/memalign.h"
30 #include "qemu/guest-random.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-events-machine.h"
33 #include "qapi/qapi-events-qdev.h"
34 #include "qapi/visitor.h"
35 #include "sysemu/sysemu.h"
36 #include "sysemu/hostmem.h"
37 #include "sysemu/numa.h"
38 #include "sysemu/tcg.h"
39 #include "sysemu/qtest.h"
40 #include "sysemu/reset.h"
41 #include "sysemu/runstate.h"
42 #include "qemu/log.h"
43 #include "hw/fw-path-provider.h"
44 #include "elf.h"
45 #include "net/net.h"
46 #include "sysemu/device_tree.h"
47 #include "sysemu/cpus.h"
48 #include "sysemu/hw_accel.h"
49 #include "kvm_ppc.h"
50 #include "migration/misc.h"
51 #include "migration/qemu-file-types.h"
52 #include "migration/global_state.h"
53 #include "migration/register.h"
54 #include "migration/blocker.h"
55 #include "mmu-hash64.h"
56 #include "mmu-book3s-v3.h"
57 #include "cpu-models.h"
58 #include "hw/core/cpu.h"
59 
60 #include "hw/ppc/ppc.h"
61 #include "hw/loader.h"
62 
63 #include "hw/ppc/fdt.h"
64 #include "hw/ppc/spapr.h"
65 #include "hw/ppc/spapr_nested.h"
66 #include "hw/ppc/spapr_vio.h"
67 #include "hw/ppc/vof.h"
68 #include "hw/qdev-properties.h"
69 #include "hw/pci-host/spapr.h"
70 #include "hw/pci/msi.h"
71 
72 #include "hw/pci/pci.h"
73 #include "hw/scsi/scsi.h"
74 #include "hw/virtio/virtio-scsi.h"
75 #include "hw/virtio/vhost-scsi-common.h"
76 
77 #include "exec/ram_addr.h"
78 #include "hw/usb.h"
79 #include "qemu/config-file.h"
80 #include "qemu/error-report.h"
81 #include "trace.h"
82 #include "hw/nmi.h"
83 #include "hw/intc/intc.h"
84 
85 #include "hw/ppc/spapr_cpu_core.h"
86 #include "hw/mem/memory-device.h"
87 #include "hw/ppc/spapr_tpm_proxy.h"
88 #include "hw/ppc/spapr_nvdimm.h"
89 #include "hw/ppc/spapr_numa.h"
90 #include "hw/ppc/pef.h"
91 
92 #include "monitor/monitor.h"
93 
94 #include <libfdt.h>
95 
96 /* SLOF memory layout:
97  *
98  * SLOF raw image loaded at 0, copies its romfs right below the flat
99  * device-tree, then position SLOF itself 31M below that
100  *
101  * So we set FW_OVERHEAD to 40MB which should account for all of that
102  * and more
103  *
104  * We load our kernel at 4M, leaving space for SLOF initial image
105  */
106 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
107 #define FW_MAX_SIZE             0x400000
108 #define FW_FILE_NAME            "slof.bin"
109 #define FW_FILE_NAME_VOF        "vof.bin"
110 #define FW_OVERHEAD             0x2800000
111 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
112 
113 #define MIN_RMA_SLOF            (128 * MiB)
114 
115 #define PHANDLE_INTC            0x00001111
116 
117 /* These two functions implement the VCPU id numbering: one to compute them
118  * all and one to identify thread 0 of a VCORE. Any change to the first one
119  * is likely to have an impact on the second one, so let's keep them close.
120  */
121 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
122 {
123     MachineState *ms = MACHINE(spapr);
124     unsigned int smp_threads = ms->smp.threads;
125 
126     assert(spapr->vsmt);
127     return
128         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
129 }
130 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
131                                       PowerPCCPU *cpu)
132 {
133     assert(spapr->vsmt);
134     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
135 }
136 
137 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
138 {
139     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
140      * and newer QEMUs don't even have them. In both cases, we don't want
141      * to send anything on the wire.
142      */
143     return false;
144 }
145 
146 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
147     /*
148      * Hack ahead.  We can't have two devices with the same name and
149      * instance id.  So I rename this to pass make check.
150      * Real help from people who knows the hardware is needed.
151      */
152     .name = "icp/server",
153     .version_id = 1,
154     .minimum_version_id = 1,
155     .needed = pre_2_10_vmstate_dummy_icp_needed,
156     .fields = (const VMStateField[]) {
157         VMSTATE_UNUSED(4), /* uint32_t xirr */
158         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
159         VMSTATE_UNUSED(1), /* uint8_t mfrr */
160         VMSTATE_END_OF_LIST()
161     },
162 };
163 
164 /*
165  * See comment in hw/intc/xics.c:icp_realize()
166  *
167  * You have to remove vmstate_replace_hack_for_ppc() when you remove
168  * the machine types that need the following function.
169  */
170 static void pre_2_10_vmstate_register_dummy_icp(int i)
171 {
172     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
173                      (void *)(uintptr_t) i);
174 }
175 
176 /*
177  * See comment in hw/intc/xics.c:icp_realize()
178  *
179  * You have to remove vmstate_replace_hack_for_ppc() when you remove
180  * the machine types that need the following function.
181  */
182 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
183 {
184     /*
185      * This used to be:
186      *
187      *    vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
188      *                      (void *)(uintptr_t) i);
189      */
190 }
191 
192 int spapr_max_server_number(SpaprMachineState *spapr)
193 {
194     MachineState *ms = MACHINE(spapr);
195 
196     assert(spapr->vsmt);
197     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
198 }
199 
200 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
201                                   int smt_threads)
202 {
203     int i, ret = 0;
204     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
205     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
206     int index = spapr_get_vcpu_id(cpu);
207 
208     if (cpu->compat_pvr) {
209         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
210         if (ret < 0) {
211             return ret;
212         }
213     }
214 
215     /* Build interrupt servers and gservers properties */
216     for (i = 0; i < smt_threads; i++) {
217         servers_prop[i] = cpu_to_be32(index + i);
218         /* Hack, direct the group queues back to cpu 0 */
219         gservers_prop[i*2] = cpu_to_be32(index + i);
220         gservers_prop[i*2 + 1] = 0;
221     }
222     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
223                       servers_prop, sizeof(*servers_prop) * smt_threads);
224     if (ret < 0) {
225         return ret;
226     }
227     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
228                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
229 
230     return ret;
231 }
232 
233 static void spapr_dt_pa_features(SpaprMachineState *spapr,
234                                  PowerPCCPU *cpu,
235                                  void *fdt, int offset)
236 {
237     /*
238      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
239      * but not MTTCG, so disable it. To advertise it, a cap would have
240      * to be added, or support implemented for MTTCG.
241      *
242      * Copy/paste is not supported by TCG, so it is not advertised. KVM
243      * can execute them but it has no accelerator drivers which are usable,
244      * so there isn't much need for it anyway.
245      */
246 
247     /* These should be kept in sync with pnv */
248     uint8_t pa_features_206[] = { 6, 0,
249         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
250     uint8_t pa_features_207[] = { 24, 0,
251         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
252         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
253         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
255     uint8_t pa_features_300[] = { 66, 0,
256         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
257         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
258         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
259         /* 6: DS207 */
260         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
261         /* 16: Vector */
262         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
263         /* 18: Vec. Scalar, 20: Vec. XOR */
264         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
265         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
266         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
267         /* 32: LE atomic, 34: EBB + ext EBB */
268         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
269         /* 40: Radix MMU */
270         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
271         /* 42: PM, 44: PC RA, 46: SC vec'd */
272         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
273         /* 48: SIMD, 50: QP BFP, 52: String */
274         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
275         /* 54: DecFP, 56: DecI, 58: SHA */
276         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
277         /* 60: NM atomic, 62: RNG */
278         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
279     };
280     /* 3.1 removes SAO, HTM support */
281     uint8_t pa_features_31[] = { 74, 0,
282         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
283         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
284         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
285         /* 6: DS207 */
286         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
287         /* 16: Vector */
288         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
289         /* 18: Vec. Scalar, 20: Vec. XOR */
290         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
291         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
292         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
293         /* 32: LE atomic, 34: EBB + ext EBB */
294         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
295         /* 40: Radix MMU */
296         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
297         /* 42: PM, 44: PC RA, 46: SC vec'd */
298         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
299         /* 48: SIMD, 50: QP BFP, 52: String */
300         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
301         /* 54: DecFP, 56: DecI, 58: SHA */
302         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
303         /* 60: NM atomic, 62: RNG */
304         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
305         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
306         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
307         /* 72: [P]HASHST/[P]HASHCHK */
308         0x80, 0x00,                         /* 72 - 73 */
309     };
310     uint8_t *pa_features = NULL;
311     size_t pa_size;
312 
313     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
314         pa_features = pa_features_206;
315         pa_size = sizeof(pa_features_206);
316     }
317     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
318         pa_features = pa_features_207;
319         pa_size = sizeof(pa_features_207);
320     }
321     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
322         pa_features = pa_features_300;
323         pa_size = sizeof(pa_features_300);
324     }
325     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
326         pa_features = pa_features_31;
327         pa_size = sizeof(pa_features_31);
328     }
329     if (!pa_features) {
330         return;
331     }
332 
333     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
334         /*
335          * Note: we keep CI large pages off by default because a 64K capable
336          * guest provisioned with large pages might otherwise try to map a qemu
337          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
338          * even if that qemu runs on a 4k host.
339          * We dd this bit back here if we are confident this is not an issue
340          */
341         pa_features[3] |= 0x20;
342     }
343     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
344         pa_features[24] |= 0x80;    /* Transactional memory support */
345     }
346     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
347         /* Workaround for broken kernels that attempt (guest) radix
348          * mode when they can't handle it, if they see the radix bit set
349          * in pa-features. So hide it from them. */
350         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
351     }
352 
353     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
354 }
355 
356 static hwaddr spapr_node0_size(MachineState *machine)
357 {
358     if (machine->numa_state->num_nodes) {
359         int i;
360         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
361             if (machine->numa_state->nodes[i].node_mem) {
362                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
363                            machine->ram_size);
364             }
365         }
366     }
367     return machine->ram_size;
368 }
369 
370 static void add_str(GString *s, const gchar *s1)
371 {
372     g_string_append_len(s, s1, strlen(s1) + 1);
373 }
374 
375 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
376                                 hwaddr start, hwaddr size)
377 {
378     char mem_name[32];
379     uint64_t mem_reg_property[2];
380     int off;
381 
382     mem_reg_property[0] = cpu_to_be64(start);
383     mem_reg_property[1] = cpu_to_be64(size);
384 
385     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
386     off = fdt_add_subnode(fdt, 0, mem_name);
387     _FDT(off);
388     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
389     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
390                       sizeof(mem_reg_property))));
391     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
392     return off;
393 }
394 
395 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
396 {
397     MemoryDeviceInfoList *info;
398 
399     for (info = list; info; info = info->next) {
400         MemoryDeviceInfo *value = info->value;
401 
402         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
403             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
404 
405             if (addr >= pcdimm_info->addr &&
406                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
407                 return pcdimm_info->node;
408             }
409         }
410     }
411 
412     return -1;
413 }
414 
415 struct sPAPRDrconfCellV2 {
416      uint32_t seq_lmbs;
417      uint64_t base_addr;
418      uint32_t drc_index;
419      uint32_t aa_index;
420      uint32_t flags;
421 } QEMU_PACKED;
422 
423 typedef struct DrconfCellQueue {
424     struct sPAPRDrconfCellV2 cell;
425     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
426 } DrconfCellQueue;
427 
428 static DrconfCellQueue *
429 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
430                       uint32_t drc_index, uint32_t aa_index,
431                       uint32_t flags)
432 {
433     DrconfCellQueue *elem;
434 
435     elem = g_malloc0(sizeof(*elem));
436     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
437     elem->cell.base_addr = cpu_to_be64(base_addr);
438     elem->cell.drc_index = cpu_to_be32(drc_index);
439     elem->cell.aa_index = cpu_to_be32(aa_index);
440     elem->cell.flags = cpu_to_be32(flags);
441 
442     return elem;
443 }
444 
445 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
446                                       int offset, MemoryDeviceInfoList *dimms)
447 {
448     MachineState *machine = MACHINE(spapr);
449     uint8_t *int_buf, *cur_index;
450     int ret;
451     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
452     uint64_t addr, cur_addr, size;
453     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
454     uint64_t mem_end = machine->device_memory->base +
455                        memory_region_size(&machine->device_memory->mr);
456     uint32_t node, buf_len, nr_entries = 0;
457     SpaprDrc *drc;
458     DrconfCellQueue *elem, *next;
459     MemoryDeviceInfoList *info;
460     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
461         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
462 
463     /* Entry to cover RAM and the gap area */
464     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
465                                  SPAPR_LMB_FLAGS_RESERVED |
466                                  SPAPR_LMB_FLAGS_DRC_INVALID);
467     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
468     nr_entries++;
469 
470     cur_addr = machine->device_memory->base;
471     for (info = dimms; info; info = info->next) {
472         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
473 
474         addr = di->addr;
475         size = di->size;
476         node = di->node;
477 
478         /*
479          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
480          * area is marked hotpluggable in the next iteration for the bigger
481          * chunk including the NVDIMM occupied area.
482          */
483         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
484             continue;
485 
486         /* Entry for hot-pluggable area */
487         if (cur_addr < addr) {
488             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
489             g_assert(drc);
490             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
491                                          cur_addr, spapr_drc_index(drc), -1, 0);
492             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
493             nr_entries++;
494         }
495 
496         /* Entry for DIMM */
497         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
498         g_assert(drc);
499         elem = spapr_get_drconf_cell(size / lmb_size, addr,
500                                      spapr_drc_index(drc), node,
501                                      (SPAPR_LMB_FLAGS_ASSIGNED |
502                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
503         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
504         nr_entries++;
505         cur_addr = addr + size;
506     }
507 
508     /* Entry for remaining hotpluggable area */
509     if (cur_addr < mem_end) {
510         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
511         g_assert(drc);
512         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
513                                      cur_addr, spapr_drc_index(drc), -1, 0);
514         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
515         nr_entries++;
516     }
517 
518     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
519     int_buf = cur_index = g_malloc0(buf_len);
520     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
521     cur_index += sizeof(nr_entries);
522 
523     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
524         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
525         cur_index += sizeof(elem->cell);
526         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
527         g_free(elem);
528     }
529 
530     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
531     g_free(int_buf);
532     if (ret < 0) {
533         return -1;
534     }
535     return 0;
536 }
537 
538 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
539                                    int offset, MemoryDeviceInfoList *dimms)
540 {
541     MachineState *machine = MACHINE(spapr);
542     int i, ret;
543     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
544     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
545     uint32_t nr_lmbs = (machine->device_memory->base +
546                        memory_region_size(&machine->device_memory->mr)) /
547                        lmb_size;
548     uint32_t *int_buf, *cur_index, buf_len;
549 
550     /*
551      * Allocate enough buffer size to fit in ibm,dynamic-memory
552      */
553     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
554     cur_index = int_buf = g_malloc0(buf_len);
555     int_buf[0] = cpu_to_be32(nr_lmbs);
556     cur_index++;
557     for (i = 0; i < nr_lmbs; i++) {
558         uint64_t addr = i * lmb_size;
559         uint32_t *dynamic_memory = cur_index;
560 
561         if (i >= device_lmb_start) {
562             SpaprDrc *drc;
563 
564             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
565             g_assert(drc);
566 
567             dynamic_memory[0] = cpu_to_be32(addr >> 32);
568             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
569             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
570             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
571             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
572             if (memory_region_present(get_system_memory(), addr)) {
573                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
574             } else {
575                 dynamic_memory[5] = cpu_to_be32(0);
576             }
577         } else {
578             /*
579              * LMB information for RMA, boot time RAM and gap b/n RAM and
580              * device memory region -- all these are marked as reserved
581              * and as having no valid DRC.
582              */
583             dynamic_memory[0] = cpu_to_be32(addr >> 32);
584             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
585             dynamic_memory[2] = cpu_to_be32(0);
586             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
587             dynamic_memory[4] = cpu_to_be32(-1);
588             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
589                                             SPAPR_LMB_FLAGS_DRC_INVALID);
590         }
591 
592         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
593     }
594     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
595     g_free(int_buf);
596     if (ret < 0) {
597         return -1;
598     }
599     return 0;
600 }
601 
602 /*
603  * Adds ibm,dynamic-reconfiguration-memory node.
604  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
605  * of this device tree node.
606  */
607 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
608                                                    void *fdt)
609 {
610     MachineState *machine = MACHINE(spapr);
611     int ret, offset;
612     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
613     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
614                                 cpu_to_be32(lmb_size & 0xffffffff)};
615     MemoryDeviceInfoList *dimms = NULL;
616 
617     /* Don't create the node if there is no device memory. */
618     if (!machine->device_memory) {
619         return 0;
620     }
621 
622     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
623 
624     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
625                     sizeof(prop_lmb_size));
626     if (ret < 0) {
627         return ret;
628     }
629 
630     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
631     if (ret < 0) {
632         return ret;
633     }
634 
635     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
636     if (ret < 0) {
637         return ret;
638     }
639 
640     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
641     dimms = qmp_memory_device_list();
642     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
643         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
644     } else {
645         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
646     }
647     qapi_free_MemoryDeviceInfoList(dimms);
648 
649     if (ret < 0) {
650         return ret;
651     }
652 
653     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
654 
655     return ret;
656 }
657 
658 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
659 {
660     MachineState *machine = MACHINE(spapr);
661     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
662     hwaddr mem_start, node_size;
663     int i, nb_nodes = machine->numa_state->num_nodes;
664     NodeInfo *nodes = machine->numa_state->nodes;
665 
666     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
667         if (!nodes[i].node_mem) {
668             continue;
669         }
670         if (mem_start >= machine->ram_size) {
671             node_size = 0;
672         } else {
673             node_size = nodes[i].node_mem;
674             if (node_size > machine->ram_size - mem_start) {
675                 node_size = machine->ram_size - mem_start;
676             }
677         }
678         if (!mem_start) {
679             /* spapr_machine_init() checks for rma_size <= node0_size
680              * already */
681             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
682             mem_start += spapr->rma_size;
683             node_size -= spapr->rma_size;
684         }
685         for ( ; node_size; ) {
686             hwaddr sizetmp = pow2floor(node_size);
687 
688             /* mem_start != 0 here */
689             if (ctzl(mem_start) < ctzl(sizetmp)) {
690                 sizetmp = 1ULL << ctzl(mem_start);
691             }
692 
693             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
694             node_size -= sizetmp;
695             mem_start += sizetmp;
696         }
697     }
698 
699     /* Generate ibm,dynamic-reconfiguration-memory node if required */
700     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
701         int ret;
702 
703         g_assert(smc->dr_lmb_enabled);
704         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
705         if (ret) {
706             return ret;
707         }
708     }
709 
710     return 0;
711 }
712 
713 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
714                          SpaprMachineState *spapr)
715 {
716     MachineState *ms = MACHINE(spapr);
717     PowerPCCPU *cpu = POWERPC_CPU(cs);
718     CPUPPCState *env = &cpu->env;
719     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
720     int index = spapr_get_vcpu_id(cpu);
721     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
722                        0xffffffff, 0xffffffff};
723     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
724         : SPAPR_TIMEBASE_FREQ;
725     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
726     uint32_t page_sizes_prop[64];
727     size_t page_sizes_prop_size;
728     unsigned int smp_threads = ms->smp.threads;
729     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
730     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
731     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
732     SpaprDrc *drc;
733     int drc_index;
734     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
735     int i;
736 
737     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
738     if (drc) {
739         drc_index = spapr_drc_index(drc);
740         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
741     }
742 
743     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
744     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
745 
746     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
747     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
748                            env->dcache_line_size)));
749     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
750                            env->dcache_line_size)));
751     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
752                            env->icache_line_size)));
753     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
754                            env->icache_line_size)));
755 
756     if (pcc->l1_dcache_size) {
757         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
758                                pcc->l1_dcache_size)));
759     } else {
760         warn_report("Unknown L1 dcache size for cpu");
761     }
762     if (pcc->l1_icache_size) {
763         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
764                                pcc->l1_icache_size)));
765     } else {
766         warn_report("Unknown L1 icache size for cpu");
767     }
768 
769     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
770     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
771     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
772     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
773     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
774     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
775 
776     if (ppc_has_spr(cpu, SPR_PURR)) {
777         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
778     }
779     if (ppc_has_spr(cpu, SPR_PURR)) {
780         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
781     }
782 
783     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
784         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
785                           segs, sizeof(segs))));
786     }
787 
788     /* Advertise VSX (vector extensions) if available
789      *   1               == VMX / Altivec available
790      *   2               == VSX available
791      *
792      * Only CPUs for which we create core types in spapr_cpu_core.c
793      * are possible, and all of those have VMX */
794     if (env->insns_flags & PPC_ALTIVEC) {
795         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
796             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
797         } else {
798             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
799         }
800     }
801 
802     /* Advertise DFP (Decimal Floating Point) if available
803      *   0 / no property == no DFP
804      *   1               == DFP available */
805     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
806         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
807     }
808 
809     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
810                                                       sizeof(page_sizes_prop));
811     if (page_sizes_prop_size) {
812         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
813                           page_sizes_prop, page_sizes_prop_size)));
814     }
815 
816     spapr_dt_pa_features(spapr, cpu, fdt, offset);
817 
818     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
819                            cs->cpu_index / vcpus_per_socket)));
820 
821     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
822                       pft_size_prop, sizeof(pft_size_prop))));
823 
824     if (ms->numa_state->num_nodes > 1) {
825         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
826     }
827 
828     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
829 
830     if (pcc->radix_page_info) {
831         for (i = 0; i < pcc->radix_page_info->count; i++) {
832             radix_AP_encodings[i] =
833                 cpu_to_be32(pcc->radix_page_info->entries[i]);
834         }
835         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
836                           radix_AP_encodings,
837                           pcc->radix_page_info->count *
838                           sizeof(radix_AP_encodings[0]))));
839     }
840 
841     /*
842      * We set this property to let the guest know that it can use the large
843      * decrementer and its width in bits.
844      */
845     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
846         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
847                               pcc->lrg_decr_bits)));
848 }
849 
850 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
851                              int cpus_offset)
852 {
853     PowerPCCPU *cpu = POWERPC_CPU(cs);
854     int index = spapr_get_vcpu_id(cpu);
855     DeviceClass *dc = DEVICE_GET_CLASS(cs);
856     g_autofree char *nodename = NULL;
857     int offset;
858 
859     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
860         return;
861     }
862 
863     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
864     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
865     _FDT(offset);
866     spapr_dt_cpu(cs, fdt, offset, spapr);
867 }
868 
869 
870 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
871 {
872     CPUState **rev;
873     CPUState *cs;
874     int n_cpus;
875     int cpus_offset;
876     int i;
877 
878     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
879     _FDT(cpus_offset);
880     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
881     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
882 
883     /*
884      * We walk the CPUs in reverse order to ensure that CPU DT nodes
885      * created by fdt_add_subnode() end up in the right order in FDT
886      * for the guest kernel the enumerate the CPUs correctly.
887      *
888      * The CPU list cannot be traversed in reverse order, so we need
889      * to do extra work.
890      */
891     n_cpus = 0;
892     rev = NULL;
893     CPU_FOREACH(cs) {
894         rev = g_renew(CPUState *, rev, n_cpus + 1);
895         rev[n_cpus++] = cs;
896     }
897 
898     for (i = n_cpus - 1; i >= 0; i--) {
899         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
900     }
901 
902     g_free(rev);
903 }
904 
905 static int spapr_dt_rng(void *fdt)
906 {
907     int node;
908     int ret;
909 
910     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
911     if (node <= 0) {
912         return -1;
913     }
914     ret = fdt_setprop_string(fdt, node, "device_type",
915                              "ibm,platform-facilities");
916     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
917     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
918 
919     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
920     if (node <= 0) {
921         return -1;
922     }
923     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
924 
925     return ret ? -1 : 0;
926 }
927 
928 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
929 {
930     MachineState *ms = MACHINE(spapr);
931     int rtas;
932     GString *hypertas = g_string_sized_new(256);
933     GString *qemu_hypertas = g_string_sized_new(256);
934     uint32_t lrdr_capacity[] = {
935         0,
936         0,
937         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
938         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
939         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
940     };
941 
942     /* Do we have device memory? */
943     if (MACHINE(spapr)->device_memory) {
944         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
945             memory_region_size(&MACHINE(spapr)->device_memory->mr);
946 
947         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
948         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
949     }
950 
951     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
952 
953     /* hypertas */
954     add_str(hypertas, "hcall-pft");
955     add_str(hypertas, "hcall-term");
956     add_str(hypertas, "hcall-dabr");
957     add_str(hypertas, "hcall-interrupt");
958     add_str(hypertas, "hcall-tce");
959     add_str(hypertas, "hcall-vio");
960     add_str(hypertas, "hcall-splpar");
961     add_str(hypertas, "hcall-join");
962     add_str(hypertas, "hcall-bulk");
963     add_str(hypertas, "hcall-set-mode");
964     add_str(hypertas, "hcall-sprg0");
965     add_str(hypertas, "hcall-copy");
966     add_str(hypertas, "hcall-debug");
967     add_str(hypertas, "hcall-vphn");
968     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
969         add_str(hypertas, "hcall-rpt-invalidate");
970     }
971 
972     add_str(qemu_hypertas, "hcall-memop1");
973 
974     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
975         add_str(hypertas, "hcall-multi-tce");
976     }
977 
978     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
979         add_str(hypertas, "hcall-hpt-resize");
980     }
981 
982     add_str(hypertas, "hcall-watchdog");
983 
984     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
985                      hypertas->str, hypertas->len));
986     g_string_free(hypertas, TRUE);
987     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
988                      qemu_hypertas->str, qemu_hypertas->len));
989     g_string_free(qemu_hypertas, TRUE);
990 
991     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
992 
993     /*
994      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
995      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
996      *
997      * The system reset requirements are driven by existing Linux and PowerVM
998      * implementation which (contrary to PAPR) saves r3 in the error log
999      * structure like machine check, so Linux expects to find the saved r3
1000      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
1001      * does not look at the error value).
1002      *
1003      * System reset interrupts are not subject to interlock like machine
1004      * check, so this memory area could be corrupted if the sreset is
1005      * interrupted by a machine check (or vice versa) if it was shared. To
1006      * prevent this, system reset uses per-CPU areas for the sreset save
1007      * area. A system reset that interrupts a system reset handler could
1008      * still overwrite this area, but Linux doesn't try to recover in that
1009      * case anyway.
1010      *
1011      * The extra 8 bytes is required because Linux's FWNMI error log check
1012      * is off-by-one.
1013      *
1014      * RTAS_MIN_SIZE is required for the RTAS blob itself.
1015      */
1016     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
1017                           RTAS_ERROR_LOG_MAX +
1018                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
1019                           sizeof(uint64_t)));
1020     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1021                           RTAS_ERROR_LOG_MAX));
1022     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1023                           RTAS_EVENT_SCAN_RATE));
1024 
1025     g_assert(msi_nonbroken);
1026     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1027 
1028     /*
1029      * According to PAPR, rtas ibm,os-term does not guarantee a return
1030      * back to the guest cpu.
1031      *
1032      * While an additional ibm,extended-os-term property indicates
1033      * that rtas call return will always occur. Set this property.
1034      */
1035     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1036 
1037     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1038                      lrdr_capacity, sizeof(lrdr_capacity)));
1039 
1040     spapr_dt_rtas_tokens(fdt, rtas);
1041 }
1042 
1043 /*
1044  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1045  * and the XIVE features that the guest may request and thus the valid
1046  * values for bytes 23..26 of option vector 5:
1047  */
1048 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1049                                           int chosen)
1050 {
1051     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1052 
1053     char val[2 * 4] = {
1054         23, 0x00, /* XICS / XIVE mode */
1055         24, 0x00, /* Hash/Radix, filled in below. */
1056         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1057         26, 0x40, /* Radix options: GTSE == yes. */
1058     };
1059 
1060     if (spapr->irq->xics && spapr->irq->xive) {
1061         val[1] = SPAPR_OV5_XIVE_BOTH;
1062     } else if (spapr->irq->xive) {
1063         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1064     } else {
1065         assert(spapr->irq->xics);
1066         val[1] = SPAPR_OV5_XIVE_LEGACY;
1067     }
1068 
1069     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1070                           first_ppc_cpu->compat_pvr)) {
1071         /*
1072          * If we're in a pre POWER9 compat mode then the guest should
1073          * do hash and use the legacy interrupt mode
1074          */
1075         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1076         val[3] = 0x00; /* Hash */
1077         spapr_check_mmu_mode(false);
1078     } else if (kvm_enabled()) {
1079         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1080             val[3] = 0x80; /* OV5_MMU_BOTH */
1081         } else if (kvmppc_has_cap_mmu_radix()) {
1082             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1083         } else {
1084             val[3] = 0x00; /* Hash */
1085         }
1086     } else {
1087         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1088         val[3] = 0xC0;
1089     }
1090     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1091                      val, sizeof(val)));
1092 }
1093 
1094 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1095 {
1096     MachineState *machine = MACHINE(spapr);
1097     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1098     int chosen;
1099 
1100     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1101 
1102     if (reset) {
1103         const char *boot_device = spapr->boot_device;
1104         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1105         size_t cb = 0;
1106         g_autofree char *bootlist = get_boot_devices_list(&cb);
1107 
1108         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1109             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1110                                     machine->kernel_cmdline));
1111         }
1112 
1113         if (spapr->initrd_size) {
1114             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1115                                   spapr->initrd_base));
1116             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1117                                   spapr->initrd_base + spapr->initrd_size));
1118         }
1119 
1120         if (spapr->kernel_size) {
1121             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1122                                   cpu_to_be64(spapr->kernel_size) };
1123 
1124             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1125                          &kprop, sizeof(kprop)));
1126             if (spapr->kernel_le) {
1127                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1128             }
1129         }
1130         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1131             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1132         }
1133         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1134         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1135         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1136 
1137         if (cb && bootlist) {
1138             int i;
1139 
1140             for (i = 0; i < cb; i++) {
1141                 if (bootlist[i] == '\n') {
1142                     bootlist[i] = ' ';
1143                 }
1144             }
1145             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1146         }
1147 
1148         if (boot_device && strlen(boot_device)) {
1149             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1150         }
1151 
1152         if (spapr->want_stdout_path && stdout_path) {
1153             /*
1154              * "linux,stdout-path" and "stdout" properties are
1155              * deprecated by linux kernel. New platforms should only
1156              * use the "stdout-path" property. Set the new property
1157              * and continue using older property to remain compatible
1158              * with the existing firmware.
1159              */
1160             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1161             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1162         }
1163 
1164         /*
1165          * We can deal with BAR reallocation just fine, advertise it
1166          * to the guest
1167          */
1168         if (smc->linux_pci_probe) {
1169             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1170         }
1171 
1172         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1173     }
1174 
1175     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1176 
1177     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1178 }
1179 
1180 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1181 {
1182     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1183      * KVM to work under pHyp with some guest co-operation */
1184     int hypervisor;
1185     uint8_t hypercall[16];
1186 
1187     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1188     /* indicate KVM hypercall interface */
1189     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1190     if (kvmppc_has_cap_fixup_hcalls()) {
1191         /*
1192          * Older KVM versions with older guest kernels were broken
1193          * with the magic page, don't allow the guest to map it.
1194          */
1195         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1196                                   sizeof(hypercall))) {
1197             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1198                              hypercall, sizeof(hypercall)));
1199         }
1200     }
1201 }
1202 
1203 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1204 {
1205     MachineState *machine = MACHINE(spapr);
1206     MachineClass *mc = MACHINE_GET_CLASS(machine);
1207     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1208     uint32_t root_drc_type_mask = 0;
1209     int ret;
1210     void *fdt;
1211     SpaprPhbState *phb;
1212     char *buf;
1213 
1214     fdt = g_malloc0(space);
1215     _FDT((fdt_create_empty_tree(fdt, space)));
1216 
1217     /* Root node */
1218     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1219     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1220     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1221 
1222     /* Guest UUID & Name*/
1223     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1224     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1225     if (qemu_uuid_set) {
1226         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1227     }
1228     g_free(buf);
1229 
1230     if (qemu_get_vm_name()) {
1231         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1232                                 qemu_get_vm_name()));
1233     }
1234 
1235     /* Host Model & Serial Number */
1236     if (spapr->host_model) {
1237         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1238     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1239         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1240         g_free(buf);
1241     }
1242 
1243     if (spapr->host_serial) {
1244         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1245     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1246         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1247         g_free(buf);
1248     }
1249 
1250     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1251     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1252 
1253     /* /interrupt controller */
1254     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1255 
1256     ret = spapr_dt_memory(spapr, fdt);
1257     if (ret < 0) {
1258         error_report("couldn't setup memory nodes in fdt");
1259         exit(1);
1260     }
1261 
1262     /* /vdevice */
1263     spapr_dt_vdevice(spapr->vio_bus, fdt);
1264 
1265     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1266         ret = spapr_dt_rng(fdt);
1267         if (ret < 0) {
1268             error_report("could not set up rng device in the fdt");
1269             exit(1);
1270         }
1271     }
1272 
1273     QLIST_FOREACH(phb, &spapr->phbs, list) {
1274         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1275         if (ret < 0) {
1276             error_report("couldn't setup PCI devices in fdt");
1277             exit(1);
1278         }
1279     }
1280 
1281     spapr_dt_cpus(fdt, spapr);
1282 
1283     /* ibm,drc-indexes and friends */
1284     if (smc->dr_lmb_enabled) {
1285         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1286     }
1287     if (smc->dr_phb_enabled) {
1288         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1289     }
1290     if (mc->nvdimm_supported) {
1291         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1292     }
1293     if (root_drc_type_mask) {
1294         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1295     }
1296 
1297     if (mc->has_hotpluggable_cpus) {
1298         int offset = fdt_path_offset(fdt, "/cpus");
1299         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1300         if (ret < 0) {
1301             error_report("Couldn't set up CPU DR device tree properties");
1302             exit(1);
1303         }
1304     }
1305 
1306     /* /event-sources */
1307     spapr_dt_events(spapr, fdt);
1308 
1309     /* /rtas */
1310     spapr_dt_rtas(spapr, fdt);
1311 
1312     /* /chosen */
1313     spapr_dt_chosen(spapr, fdt, reset);
1314 
1315     /* /hypervisor */
1316     if (kvm_enabled()) {
1317         spapr_dt_hypervisor(spapr, fdt);
1318     }
1319 
1320     /* Build memory reserve map */
1321     if (reset) {
1322         if (spapr->kernel_size) {
1323             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1324                                   spapr->kernel_size)));
1325         }
1326         if (spapr->initrd_size) {
1327             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1328                                   spapr->initrd_size)));
1329         }
1330     }
1331 
1332     /* NVDIMM devices */
1333     if (mc->nvdimm_supported) {
1334         spapr_dt_persistent_memory(spapr, fdt);
1335     }
1336 
1337     return fdt;
1338 }
1339 
1340 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1341 {
1342     SpaprMachineState *spapr = opaque;
1343 
1344     return (addr & 0x0fffffff) + spapr->kernel_addr;
1345 }
1346 
1347 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1348                                     PowerPCCPU *cpu)
1349 {
1350     CPUPPCState *env = &cpu->env;
1351 
1352     /* The TCG path should also be holding the BQL at this point */
1353     g_assert(bql_locked());
1354 
1355     g_assert(!vhyp_cpu_in_nested(cpu));
1356 
1357     if (FIELD_EX64(env->msr, MSR, PR)) {
1358         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1359         env->gpr[3] = H_PRIVILEGE;
1360     } else {
1361         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1362     }
1363 }
1364 
1365 struct LPCRSyncState {
1366     target_ulong value;
1367     target_ulong mask;
1368 };
1369 
1370 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1371 {
1372     struct LPCRSyncState *s = arg.host_ptr;
1373     PowerPCCPU *cpu = POWERPC_CPU(cs);
1374     CPUPPCState *env = &cpu->env;
1375     target_ulong lpcr;
1376 
1377     cpu_synchronize_state(cs);
1378     lpcr = env->spr[SPR_LPCR];
1379     lpcr &= ~s->mask;
1380     lpcr |= s->value;
1381     ppc_store_lpcr(cpu, lpcr);
1382 }
1383 
1384 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1385 {
1386     CPUState *cs;
1387     struct LPCRSyncState s = {
1388         .value = value,
1389         .mask = mask
1390     };
1391     CPU_FOREACH(cs) {
1392         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1393     }
1394 }
1395 
1396 /* May be used when the machine is not running */
1397 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1398 {
1399     CPUState *cs;
1400     CPU_FOREACH(cs) {
1401         PowerPCCPU *cpu = POWERPC_CPU(cs);
1402         CPUPPCState *env = &cpu->env;
1403         target_ulong lpcr;
1404 
1405         lpcr = env->spr[SPR_LPCR];
1406         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1407         ppc_store_lpcr(cpu, lpcr);
1408     }
1409 }
1410 
1411 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1412                            target_ulong lpid, ppc_v3_pate_t *entry)
1413 {
1414     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1415     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1416 
1417     if (!spapr_cpu->in_nested) {
1418         assert(lpid == 0);
1419 
1420         /* Copy PATE1:GR into PATE0:HR */
1421         entry->dw0 = spapr->patb_entry & PATE0_HR;
1422         entry->dw1 = spapr->patb_entry;
1423         return true;
1424     } else {
1425         if (spapr_nested_api(spapr) == NESTED_API_KVM_HV) {
1426             return spapr_get_pate_nested_hv(spapr, cpu, lpid, entry);
1427         } else if (spapr_nested_api(spapr) == NESTED_API_PAPR) {
1428             return spapr_get_pate_nested_papr(spapr, cpu, lpid, entry);
1429         } else {
1430             g_assert_not_reached();
1431         }
1432     }
1433 }
1434 
1435 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1436 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1437 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1438 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1439 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1440 
1441 /*
1442  * Get the fd to access the kernel htab, re-opening it if necessary
1443  */
1444 static int get_htab_fd(SpaprMachineState *spapr)
1445 {
1446     Error *local_err = NULL;
1447 
1448     if (spapr->htab_fd >= 0) {
1449         return spapr->htab_fd;
1450     }
1451 
1452     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1453     if (spapr->htab_fd < 0) {
1454         error_report_err(local_err);
1455     }
1456 
1457     return spapr->htab_fd;
1458 }
1459 
1460 void close_htab_fd(SpaprMachineState *spapr)
1461 {
1462     if (spapr->htab_fd >= 0) {
1463         close(spapr->htab_fd);
1464     }
1465     spapr->htab_fd = -1;
1466 }
1467 
1468 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1469 {
1470     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1471 
1472     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1473 }
1474 
1475 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1476 {
1477     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1478 
1479     assert(kvm_enabled());
1480 
1481     if (!spapr->htab) {
1482         return 0;
1483     }
1484 
1485     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1486 }
1487 
1488 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1489                                                 hwaddr ptex, int n)
1490 {
1491     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1492     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1493 
1494     if (!spapr->htab) {
1495         /*
1496          * HTAB is controlled by KVM. Fetch into temporary buffer
1497          */
1498         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1499         kvmppc_read_hptes(hptes, ptex, n);
1500         return hptes;
1501     }
1502 
1503     /*
1504      * HTAB is controlled by QEMU. Just point to the internally
1505      * accessible PTEG.
1506      */
1507     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1508 }
1509 
1510 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1511                               const ppc_hash_pte64_t *hptes,
1512                               hwaddr ptex, int n)
1513 {
1514     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1515 
1516     if (!spapr->htab) {
1517         g_free((void *)hptes);
1518     }
1519 
1520     /* Nothing to do for qemu managed HPT */
1521 }
1522 
1523 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1524                       uint64_t pte0, uint64_t pte1)
1525 {
1526     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1527     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1528 
1529     if (!spapr->htab) {
1530         kvmppc_write_hpte(ptex, pte0, pte1);
1531     } else {
1532         if (pte0 & HPTE64_V_VALID) {
1533             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1534             /*
1535              * When setting valid, we write PTE1 first. This ensures
1536              * proper synchronization with the reading code in
1537              * ppc_hash64_pteg_search()
1538              */
1539             smp_wmb();
1540             stq_p(spapr->htab + offset, pte0);
1541         } else {
1542             stq_p(spapr->htab + offset, pte0);
1543             /*
1544              * When clearing it we set PTE0 first. This ensures proper
1545              * synchronization with the reading code in
1546              * ppc_hash64_pteg_search()
1547              */
1548             smp_wmb();
1549             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1550         }
1551     }
1552 }
1553 
1554 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1555                              uint64_t pte1)
1556 {
1557     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1558     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1559 
1560     if (!spapr->htab) {
1561         /* There should always be a hash table when this is called */
1562         error_report("spapr_hpte_set_c called with no hash table !");
1563         return;
1564     }
1565 
1566     /* The HW performs a non-atomic byte update */
1567     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1568 }
1569 
1570 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1571                              uint64_t pte1)
1572 {
1573     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1574     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1575 
1576     if (!spapr->htab) {
1577         /* There should always be a hash table when this is called */
1578         error_report("spapr_hpte_set_r called with no hash table !");
1579         return;
1580     }
1581 
1582     /* The HW performs a non-atomic byte update */
1583     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1584 }
1585 
1586 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1587 {
1588     int shift;
1589 
1590     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1591      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1592      * that's much more than is needed for Linux guests */
1593     shift = ctz64(pow2ceil(ramsize)) - 7;
1594     shift = MAX(shift, 18); /* Minimum architected size */
1595     shift = MIN(shift, 46); /* Maximum architected size */
1596     return shift;
1597 }
1598 
1599 void spapr_free_hpt(SpaprMachineState *spapr)
1600 {
1601     qemu_vfree(spapr->htab);
1602     spapr->htab = NULL;
1603     spapr->htab_shift = 0;
1604     close_htab_fd(spapr);
1605 }
1606 
1607 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1608 {
1609     ERRP_GUARD();
1610     long rc;
1611 
1612     /* Clean up any HPT info from a previous boot */
1613     spapr_free_hpt(spapr);
1614 
1615     rc = kvmppc_reset_htab(shift);
1616 
1617     if (rc == -EOPNOTSUPP) {
1618         error_setg(errp, "HPT not supported in nested guests");
1619         return -EOPNOTSUPP;
1620     }
1621 
1622     if (rc < 0) {
1623         /* kernel-side HPT needed, but couldn't allocate one */
1624         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1625                          shift);
1626         error_append_hint(errp, "Try smaller maxmem?\n");
1627         return -errno;
1628     } else if (rc > 0) {
1629         /* kernel-side HPT allocated */
1630         if (rc != shift) {
1631             error_setg(errp,
1632                        "Requested order %d HPT, but kernel allocated order %ld",
1633                        shift, rc);
1634             error_append_hint(errp, "Try smaller maxmem?\n");
1635             return -ENOSPC;
1636         }
1637 
1638         spapr->htab_shift = shift;
1639         spapr->htab = NULL;
1640     } else {
1641         /* kernel-side HPT not needed, allocate in userspace instead */
1642         size_t size = 1ULL << shift;
1643         int i;
1644 
1645         spapr->htab = qemu_memalign(size, size);
1646         memset(spapr->htab, 0, size);
1647         spapr->htab_shift = shift;
1648 
1649         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1650             DIRTY_HPTE(HPTE(spapr->htab, i));
1651         }
1652     }
1653     /* We're setting up a hash table, so that means we're not radix */
1654     spapr->patb_entry = 0;
1655     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1656     return 0;
1657 }
1658 
1659 void spapr_setup_hpt(SpaprMachineState *spapr)
1660 {
1661     int hpt_shift;
1662 
1663     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1664         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1665     } else {
1666         uint64_t current_ram_size;
1667 
1668         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1669         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1670     }
1671     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1672 
1673     if (kvm_enabled()) {
1674         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1675 
1676         /* Check our RMA fits in the possible VRMA */
1677         if (vrma_limit < spapr->rma_size) {
1678             error_report("Unable to create %" HWADDR_PRIu
1679                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1680                          spapr->rma_size / MiB, vrma_limit / MiB);
1681             exit(EXIT_FAILURE);
1682         }
1683     }
1684 }
1685 
1686 void spapr_check_mmu_mode(bool guest_radix)
1687 {
1688     if (guest_radix) {
1689         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1690             error_report("Guest requested unavailable MMU mode (radix).");
1691             exit(EXIT_FAILURE);
1692         }
1693     } else {
1694         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1695             && !kvmppc_has_cap_mmu_hash_v3()) {
1696             error_report("Guest requested unavailable MMU mode (hash).");
1697             exit(EXIT_FAILURE);
1698         }
1699     }
1700 }
1701 
1702 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1703 {
1704     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1705     PowerPCCPU *first_ppc_cpu;
1706     hwaddr fdt_addr;
1707     void *fdt;
1708     int rc;
1709 
1710     if (reason != SHUTDOWN_CAUSE_SNAPSHOT_LOAD) {
1711         /*
1712          * Record-replay snapshot load must not consume random, this was
1713          * already replayed from initial machine reset.
1714          */
1715         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1716     }
1717 
1718     pef_kvm_reset(machine->cgs, &error_fatal);
1719     spapr_caps_apply(spapr);
1720     spapr_nested_reset(spapr);
1721 
1722     first_ppc_cpu = POWERPC_CPU(first_cpu);
1723     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1724         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1725                               spapr->max_compat_pvr)) {
1726         /*
1727          * If using KVM with radix mode available, VCPUs can be started
1728          * without a HPT because KVM will start them in radix mode.
1729          * Set the GR bit in PATE so that we know there is no HPT.
1730          */
1731         spapr->patb_entry = PATE1_GR;
1732         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1733     } else {
1734         spapr_setup_hpt(spapr);
1735     }
1736 
1737     qemu_devices_reset(reason);
1738 
1739     spapr_ovec_cleanup(spapr->ov5_cas);
1740     spapr->ov5_cas = spapr_ovec_new();
1741 
1742     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1743 
1744     /*
1745      * This is fixing some of the default configuration of the XIVE
1746      * devices. To be called after the reset of the machine devices.
1747      */
1748     spapr_irq_reset(spapr, &error_fatal);
1749 
1750     /*
1751      * There is no CAS under qtest. Simulate one to please the code that
1752      * depends on spapr->ov5_cas. This is especially needed to test device
1753      * unplug, so we do that before resetting the DRCs.
1754      */
1755     if (qtest_enabled()) {
1756         spapr_ovec_cleanup(spapr->ov5_cas);
1757         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1758     }
1759 
1760     spapr_nvdimm_finish_flushes();
1761 
1762     /* DRC reset may cause a device to be unplugged. This will cause troubles
1763      * if this device is used by another device (eg, a running vhost backend
1764      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1765      * situations, we reset DRCs after all devices have been reset.
1766      */
1767     spapr_drc_reset_all(spapr);
1768 
1769     spapr_clear_pending_events(spapr);
1770 
1771     /*
1772      * We place the device tree just below either the top of the RMA,
1773      * or just below 2GB, whichever is lower, so that it can be
1774      * processed with 32-bit real mode code if necessary
1775      */
1776     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1777 
1778     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1779     if (spapr->vof) {
1780         spapr_vof_reset(spapr, fdt, &error_fatal);
1781         /*
1782          * Do not pack the FDT as the client may change properties.
1783          * VOF client does not expect the FDT so we do not load it to the VM.
1784          */
1785     } else {
1786         rc = fdt_pack(fdt);
1787         /* Should only fail if we've built a corrupted tree */
1788         assert(rc == 0);
1789 
1790         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1791                                   0, fdt_addr, 0);
1792         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1793     }
1794     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1795 
1796     g_free(spapr->fdt_blob);
1797     spapr->fdt_size = fdt_totalsize(fdt);
1798     spapr->fdt_initial_size = spapr->fdt_size;
1799     spapr->fdt_blob = fdt;
1800 
1801     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1802     machine->fdt = fdt;
1803 
1804     /* Set up the entry state */
1805     first_ppc_cpu->env.gpr[5] = 0;
1806 
1807     spapr->fwnmi_system_reset_addr = -1;
1808     spapr->fwnmi_machine_check_addr = -1;
1809     spapr->fwnmi_machine_check_interlock = -1;
1810 
1811     /* Signal all vCPUs waiting on this condition */
1812     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1813 
1814     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1815 }
1816 
1817 static void spapr_create_nvram(SpaprMachineState *spapr)
1818 {
1819     DeviceState *dev = qdev_new("spapr-nvram");
1820     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1821 
1822     if (dinfo) {
1823         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1824                                 &error_fatal);
1825     }
1826 
1827     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1828 
1829     spapr->nvram = (struct SpaprNvram *)dev;
1830 }
1831 
1832 static void spapr_rtc_create(SpaprMachineState *spapr)
1833 {
1834     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1835                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1836                                        &error_fatal, NULL);
1837     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1838     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1839                               "date");
1840 }
1841 
1842 /* Returns whether we want to use VGA or not */
1843 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1844 {
1845     vga_interface_created = true;
1846     switch (vga_interface_type) {
1847     case VGA_NONE:
1848         return false;
1849     case VGA_DEVICE:
1850         return true;
1851     case VGA_STD:
1852     case VGA_VIRTIO:
1853     case VGA_CIRRUS:
1854         return pci_vga_init(pci_bus) != NULL;
1855     default:
1856         error_setg(errp,
1857                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1858         return false;
1859     }
1860 }
1861 
1862 static int spapr_pre_load(void *opaque)
1863 {
1864     int rc;
1865 
1866     rc = spapr_caps_pre_load(opaque);
1867     if (rc) {
1868         return rc;
1869     }
1870 
1871     return 0;
1872 }
1873 
1874 static int spapr_post_load(void *opaque, int version_id)
1875 {
1876     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1877     int err = 0;
1878 
1879     err = spapr_caps_post_migration(spapr);
1880     if (err) {
1881         return err;
1882     }
1883 
1884     /*
1885      * In earlier versions, there was no separate qdev for the PAPR
1886      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1887      * So when migrating from those versions, poke the incoming offset
1888      * value into the RTC device
1889      */
1890     if (version_id < 3) {
1891         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1892         if (err) {
1893             return err;
1894         }
1895     }
1896 
1897     if (kvm_enabled() && spapr->patb_entry) {
1898         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1899         bool radix = !!(spapr->patb_entry & PATE1_GR);
1900         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1901 
1902         /*
1903          * Update LPCR:HR and UPRT as they may not be set properly in
1904          * the stream
1905          */
1906         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1907                             LPCR_HR | LPCR_UPRT);
1908 
1909         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1910         if (err) {
1911             error_report("Process table config unsupported by the host");
1912             return -EINVAL;
1913         }
1914     }
1915 
1916     err = spapr_irq_post_load(spapr, version_id);
1917     if (err) {
1918         return err;
1919     }
1920 
1921     return err;
1922 }
1923 
1924 static int spapr_pre_save(void *opaque)
1925 {
1926     int rc;
1927 
1928     rc = spapr_caps_pre_save(opaque);
1929     if (rc) {
1930         return rc;
1931     }
1932 
1933     return 0;
1934 }
1935 
1936 static bool version_before_3(void *opaque, int version_id)
1937 {
1938     return version_id < 3;
1939 }
1940 
1941 static bool spapr_pending_events_needed(void *opaque)
1942 {
1943     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1944     return !QTAILQ_EMPTY(&spapr->pending_events);
1945 }
1946 
1947 static const VMStateDescription vmstate_spapr_event_entry = {
1948     .name = "spapr_event_log_entry",
1949     .version_id = 1,
1950     .minimum_version_id = 1,
1951     .fields = (const VMStateField[]) {
1952         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1953         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1954         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1955                                      NULL, extended_length),
1956         VMSTATE_END_OF_LIST()
1957     },
1958 };
1959 
1960 static const VMStateDescription vmstate_spapr_pending_events = {
1961     .name = "spapr_pending_events",
1962     .version_id = 1,
1963     .minimum_version_id = 1,
1964     .needed = spapr_pending_events_needed,
1965     .fields = (const VMStateField[]) {
1966         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1967                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1968         VMSTATE_END_OF_LIST()
1969     },
1970 };
1971 
1972 static bool spapr_ov5_cas_needed(void *opaque)
1973 {
1974     SpaprMachineState *spapr = opaque;
1975     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1976     bool cas_needed;
1977 
1978     /* Prior to the introduction of SpaprOptionVector, we had two option
1979      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1980      * Both of these options encode machine topology into the device-tree
1981      * in such a way that the now-booted OS should still be able to interact
1982      * appropriately with QEMU regardless of what options were actually
1983      * negotiatied on the source side.
1984      *
1985      * As such, we can avoid migrating the CAS-negotiated options if these
1986      * are the only options available on the current machine/platform.
1987      * Since these are the only options available for pseries-2.7 and
1988      * earlier, this allows us to maintain old->new/new->old migration
1989      * compatibility.
1990      *
1991      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1992      * via default pseries-2.8 machines and explicit command-line parameters.
1993      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1994      * of the actual CAS-negotiated values to continue working properly. For
1995      * example, availability of memory unplug depends on knowing whether
1996      * OV5_HP_EVT was negotiated via CAS.
1997      *
1998      * Thus, for any cases where the set of available CAS-negotiatable
1999      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2000      * include the CAS-negotiated options in the migration stream, unless
2001      * if they affect boot time behaviour only.
2002      */
2003     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2004     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2005     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2006 
2007     /* We need extra information if we have any bits outside the mask
2008      * defined above */
2009     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
2010 
2011     spapr_ovec_cleanup(ov5_mask);
2012 
2013     return cas_needed;
2014 }
2015 
2016 static const VMStateDescription vmstate_spapr_ov5_cas = {
2017     .name = "spapr_option_vector_ov5_cas",
2018     .version_id = 1,
2019     .minimum_version_id = 1,
2020     .needed = spapr_ov5_cas_needed,
2021     .fields = (const VMStateField[]) {
2022         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2023                                  vmstate_spapr_ovec, SpaprOptionVector),
2024         VMSTATE_END_OF_LIST()
2025     },
2026 };
2027 
2028 static bool spapr_patb_entry_needed(void *opaque)
2029 {
2030     SpaprMachineState *spapr = opaque;
2031 
2032     return !!spapr->patb_entry;
2033 }
2034 
2035 static const VMStateDescription vmstate_spapr_patb_entry = {
2036     .name = "spapr_patb_entry",
2037     .version_id = 1,
2038     .minimum_version_id = 1,
2039     .needed = spapr_patb_entry_needed,
2040     .fields = (const VMStateField[]) {
2041         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2042         VMSTATE_END_OF_LIST()
2043     },
2044 };
2045 
2046 static bool spapr_irq_map_needed(void *opaque)
2047 {
2048     SpaprMachineState *spapr = opaque;
2049 
2050     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2051 }
2052 
2053 static const VMStateDescription vmstate_spapr_irq_map = {
2054     .name = "spapr_irq_map",
2055     .version_id = 1,
2056     .minimum_version_id = 1,
2057     .needed = spapr_irq_map_needed,
2058     .fields = (const VMStateField[]) {
2059         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2060         VMSTATE_END_OF_LIST()
2061     },
2062 };
2063 
2064 static bool spapr_dtb_needed(void *opaque)
2065 {
2066     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2067 
2068     return smc->update_dt_enabled;
2069 }
2070 
2071 static int spapr_dtb_pre_load(void *opaque)
2072 {
2073     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2074 
2075     g_free(spapr->fdt_blob);
2076     spapr->fdt_blob = NULL;
2077     spapr->fdt_size = 0;
2078 
2079     return 0;
2080 }
2081 
2082 static const VMStateDescription vmstate_spapr_dtb = {
2083     .name = "spapr_dtb",
2084     .version_id = 1,
2085     .minimum_version_id = 1,
2086     .needed = spapr_dtb_needed,
2087     .pre_load = spapr_dtb_pre_load,
2088     .fields = (const VMStateField[]) {
2089         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2090         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2091         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2092                                      fdt_size),
2093         VMSTATE_END_OF_LIST()
2094     },
2095 };
2096 
2097 static bool spapr_fwnmi_needed(void *opaque)
2098 {
2099     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2100 
2101     return spapr->fwnmi_machine_check_addr != -1;
2102 }
2103 
2104 static int spapr_fwnmi_pre_save(void *opaque)
2105 {
2106     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2107 
2108     /*
2109      * Check if machine check handling is in progress and print a
2110      * warning message.
2111      */
2112     if (spapr->fwnmi_machine_check_interlock != -1) {
2113         warn_report("A machine check is being handled during migration. The"
2114                 "handler may run and log hardware error on the destination");
2115     }
2116 
2117     return 0;
2118 }
2119 
2120 static const VMStateDescription vmstate_spapr_fwnmi = {
2121     .name = "spapr_fwnmi",
2122     .version_id = 1,
2123     .minimum_version_id = 1,
2124     .needed = spapr_fwnmi_needed,
2125     .pre_save = spapr_fwnmi_pre_save,
2126     .fields = (const VMStateField[]) {
2127         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2128         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2129         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2130         VMSTATE_END_OF_LIST()
2131     },
2132 };
2133 
2134 static const VMStateDescription vmstate_spapr = {
2135     .name = "spapr",
2136     .version_id = 3,
2137     .minimum_version_id = 1,
2138     .pre_load = spapr_pre_load,
2139     .post_load = spapr_post_load,
2140     .pre_save = spapr_pre_save,
2141     .fields = (const VMStateField[]) {
2142         /* used to be @next_irq */
2143         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2144 
2145         /* RTC offset */
2146         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2147 
2148         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2149         VMSTATE_END_OF_LIST()
2150     },
2151     .subsections = (const VMStateDescription * const []) {
2152         &vmstate_spapr_ov5_cas,
2153         &vmstate_spapr_patb_entry,
2154         &vmstate_spapr_pending_events,
2155         &vmstate_spapr_cap_htm,
2156         &vmstate_spapr_cap_vsx,
2157         &vmstate_spapr_cap_dfp,
2158         &vmstate_spapr_cap_cfpc,
2159         &vmstate_spapr_cap_sbbc,
2160         &vmstate_spapr_cap_ibs,
2161         &vmstate_spapr_cap_hpt_maxpagesize,
2162         &vmstate_spapr_irq_map,
2163         &vmstate_spapr_cap_nested_kvm_hv,
2164         &vmstate_spapr_dtb,
2165         &vmstate_spapr_cap_large_decr,
2166         &vmstate_spapr_cap_ccf_assist,
2167         &vmstate_spapr_cap_fwnmi,
2168         &vmstate_spapr_fwnmi,
2169         &vmstate_spapr_cap_rpt_invalidate,
2170         &vmstate_spapr_cap_nested_papr,
2171         NULL
2172     }
2173 };
2174 
2175 static int htab_save_setup(QEMUFile *f, void *opaque)
2176 {
2177     SpaprMachineState *spapr = opaque;
2178 
2179     /* "Iteration" header */
2180     if (!spapr->htab_shift) {
2181         qemu_put_be32(f, -1);
2182     } else {
2183         qemu_put_be32(f, spapr->htab_shift);
2184     }
2185 
2186     if (spapr->htab) {
2187         spapr->htab_save_index = 0;
2188         spapr->htab_first_pass = true;
2189     } else {
2190         if (spapr->htab_shift) {
2191             assert(kvm_enabled());
2192         }
2193     }
2194 
2195 
2196     return 0;
2197 }
2198 
2199 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2200                             int chunkstart, int n_valid, int n_invalid)
2201 {
2202     qemu_put_be32(f, chunkstart);
2203     qemu_put_be16(f, n_valid);
2204     qemu_put_be16(f, n_invalid);
2205     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2206                     HASH_PTE_SIZE_64 * n_valid);
2207 }
2208 
2209 static void htab_save_end_marker(QEMUFile *f)
2210 {
2211     qemu_put_be32(f, 0);
2212     qemu_put_be16(f, 0);
2213     qemu_put_be16(f, 0);
2214 }
2215 
2216 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2217                                  int64_t max_ns)
2218 {
2219     bool has_timeout = max_ns != -1;
2220     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2221     int index = spapr->htab_save_index;
2222     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2223 
2224     assert(spapr->htab_first_pass);
2225 
2226     do {
2227         int chunkstart;
2228 
2229         /* Consume invalid HPTEs */
2230         while ((index < htabslots)
2231                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2232             CLEAN_HPTE(HPTE(spapr->htab, index));
2233             index++;
2234         }
2235 
2236         /* Consume valid HPTEs */
2237         chunkstart = index;
2238         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2239                && HPTE_VALID(HPTE(spapr->htab, index))) {
2240             CLEAN_HPTE(HPTE(spapr->htab, index));
2241             index++;
2242         }
2243 
2244         if (index > chunkstart) {
2245             int n_valid = index - chunkstart;
2246 
2247             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2248 
2249             if (has_timeout &&
2250                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2251                 break;
2252             }
2253         }
2254     } while ((index < htabslots) && !migration_rate_exceeded(f));
2255 
2256     if (index >= htabslots) {
2257         assert(index == htabslots);
2258         index = 0;
2259         spapr->htab_first_pass = false;
2260     }
2261     spapr->htab_save_index = index;
2262 }
2263 
2264 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2265                                 int64_t max_ns)
2266 {
2267     bool final = max_ns < 0;
2268     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2269     int examined = 0, sent = 0;
2270     int index = spapr->htab_save_index;
2271     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2272 
2273     assert(!spapr->htab_first_pass);
2274 
2275     do {
2276         int chunkstart, invalidstart;
2277 
2278         /* Consume non-dirty HPTEs */
2279         while ((index < htabslots)
2280                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2281             index++;
2282             examined++;
2283         }
2284 
2285         chunkstart = index;
2286         /* Consume valid dirty HPTEs */
2287         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2288                && HPTE_DIRTY(HPTE(spapr->htab, index))
2289                && HPTE_VALID(HPTE(spapr->htab, index))) {
2290             CLEAN_HPTE(HPTE(spapr->htab, index));
2291             index++;
2292             examined++;
2293         }
2294 
2295         invalidstart = index;
2296         /* Consume invalid dirty HPTEs */
2297         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2298                && HPTE_DIRTY(HPTE(spapr->htab, index))
2299                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2300             CLEAN_HPTE(HPTE(spapr->htab, index));
2301             index++;
2302             examined++;
2303         }
2304 
2305         if (index > chunkstart) {
2306             int n_valid = invalidstart - chunkstart;
2307             int n_invalid = index - invalidstart;
2308 
2309             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2310             sent += index - chunkstart;
2311 
2312             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2313                 break;
2314             }
2315         }
2316 
2317         if (examined >= htabslots) {
2318             break;
2319         }
2320 
2321         if (index >= htabslots) {
2322             assert(index == htabslots);
2323             index = 0;
2324         }
2325     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2326 
2327     if (index >= htabslots) {
2328         assert(index == htabslots);
2329         index = 0;
2330     }
2331 
2332     spapr->htab_save_index = index;
2333 
2334     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2335 }
2336 
2337 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2338 #define MAX_KVM_BUF_SIZE    2048
2339 
2340 static int htab_save_iterate(QEMUFile *f, void *opaque)
2341 {
2342     SpaprMachineState *spapr = opaque;
2343     int fd;
2344     int rc = 0;
2345 
2346     /* Iteration header */
2347     if (!spapr->htab_shift) {
2348         qemu_put_be32(f, -1);
2349         return 1;
2350     } else {
2351         qemu_put_be32(f, 0);
2352     }
2353 
2354     if (!spapr->htab) {
2355         assert(kvm_enabled());
2356 
2357         fd = get_htab_fd(spapr);
2358         if (fd < 0) {
2359             return fd;
2360         }
2361 
2362         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2363         if (rc < 0) {
2364             return rc;
2365         }
2366     } else  if (spapr->htab_first_pass) {
2367         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2368     } else {
2369         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2370     }
2371 
2372     htab_save_end_marker(f);
2373 
2374     return rc;
2375 }
2376 
2377 static int htab_save_complete(QEMUFile *f, void *opaque)
2378 {
2379     SpaprMachineState *spapr = opaque;
2380     int fd;
2381 
2382     /* Iteration header */
2383     if (!spapr->htab_shift) {
2384         qemu_put_be32(f, -1);
2385         return 0;
2386     } else {
2387         qemu_put_be32(f, 0);
2388     }
2389 
2390     if (!spapr->htab) {
2391         int rc;
2392 
2393         assert(kvm_enabled());
2394 
2395         fd = get_htab_fd(spapr);
2396         if (fd < 0) {
2397             return fd;
2398         }
2399 
2400         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2401         if (rc < 0) {
2402             return rc;
2403         }
2404     } else {
2405         if (spapr->htab_first_pass) {
2406             htab_save_first_pass(f, spapr, -1);
2407         }
2408         htab_save_later_pass(f, spapr, -1);
2409     }
2410 
2411     /* End marker */
2412     htab_save_end_marker(f);
2413 
2414     return 0;
2415 }
2416 
2417 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2418 {
2419     SpaprMachineState *spapr = opaque;
2420     uint32_t section_hdr;
2421     int fd = -1;
2422     Error *local_err = NULL;
2423 
2424     if (version_id < 1 || version_id > 1) {
2425         error_report("htab_load() bad version");
2426         return -EINVAL;
2427     }
2428 
2429     section_hdr = qemu_get_be32(f);
2430 
2431     if (section_hdr == -1) {
2432         spapr_free_hpt(spapr);
2433         return 0;
2434     }
2435 
2436     if (section_hdr) {
2437         int ret;
2438 
2439         /* First section gives the htab size */
2440         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2441         if (ret < 0) {
2442             error_report_err(local_err);
2443             return ret;
2444         }
2445         return 0;
2446     }
2447 
2448     if (!spapr->htab) {
2449         assert(kvm_enabled());
2450 
2451         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2452         if (fd < 0) {
2453             error_report_err(local_err);
2454             return fd;
2455         }
2456     }
2457 
2458     while (true) {
2459         uint32_t index;
2460         uint16_t n_valid, n_invalid;
2461 
2462         index = qemu_get_be32(f);
2463         n_valid = qemu_get_be16(f);
2464         n_invalid = qemu_get_be16(f);
2465 
2466         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2467             /* End of Stream */
2468             break;
2469         }
2470 
2471         if ((index + n_valid + n_invalid) >
2472             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2473             /* Bad index in stream */
2474             error_report(
2475                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2476                 index, n_valid, n_invalid, spapr->htab_shift);
2477             return -EINVAL;
2478         }
2479 
2480         if (spapr->htab) {
2481             if (n_valid) {
2482                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2483                                 HASH_PTE_SIZE_64 * n_valid);
2484             }
2485             if (n_invalid) {
2486                 memset(HPTE(spapr->htab, index + n_valid), 0,
2487                        HASH_PTE_SIZE_64 * n_invalid);
2488             }
2489         } else {
2490             int rc;
2491 
2492             assert(fd >= 0);
2493 
2494             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2495                                         &local_err);
2496             if (rc < 0) {
2497                 error_report_err(local_err);
2498                 return rc;
2499             }
2500         }
2501     }
2502 
2503     if (!spapr->htab) {
2504         assert(fd >= 0);
2505         close(fd);
2506     }
2507 
2508     return 0;
2509 }
2510 
2511 static void htab_save_cleanup(void *opaque)
2512 {
2513     SpaprMachineState *spapr = opaque;
2514 
2515     close_htab_fd(spapr);
2516 }
2517 
2518 static SaveVMHandlers savevm_htab_handlers = {
2519     .save_setup = htab_save_setup,
2520     .save_live_iterate = htab_save_iterate,
2521     .save_live_complete_precopy = htab_save_complete,
2522     .save_cleanup = htab_save_cleanup,
2523     .load_state = htab_load,
2524 };
2525 
2526 static void spapr_boot_set(void *opaque, const char *boot_device,
2527                            Error **errp)
2528 {
2529     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2530 
2531     g_free(spapr->boot_device);
2532     spapr->boot_device = g_strdup(boot_device);
2533 }
2534 
2535 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2536 {
2537     MachineState *machine = MACHINE(spapr);
2538     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2539     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2540     int i;
2541 
2542     g_assert(!nr_lmbs || machine->device_memory);
2543     for (i = 0; i < nr_lmbs; i++) {
2544         uint64_t addr;
2545 
2546         addr = i * lmb_size + machine->device_memory->base;
2547         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2548                                addr / lmb_size);
2549     }
2550 }
2551 
2552 /*
2553  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2554  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2555  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2556  */
2557 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2558 {
2559     int i;
2560 
2561     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2562         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2563                    " is not aligned to %" PRIu64 " MiB",
2564                    machine->ram_size,
2565                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2566         return;
2567     }
2568 
2569     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2570         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2571                    " is not aligned to %" PRIu64 " MiB",
2572                    machine->ram_size,
2573                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2574         return;
2575     }
2576 
2577     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2578         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2579             error_setg(errp,
2580                        "Node %d memory size 0x%" PRIx64
2581                        " is not aligned to %" PRIu64 " MiB",
2582                        i, machine->numa_state->nodes[i].node_mem,
2583                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2584             return;
2585         }
2586     }
2587 }
2588 
2589 /* find cpu slot in machine->possible_cpus by core_id */
2590 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2591 {
2592     int index = id / ms->smp.threads;
2593 
2594     if (index >= ms->possible_cpus->len) {
2595         return NULL;
2596     }
2597     if (idx) {
2598         *idx = index;
2599     }
2600     return &ms->possible_cpus->cpus[index];
2601 }
2602 
2603 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2604 {
2605     MachineState *ms = MACHINE(spapr);
2606     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2607     Error *local_err = NULL;
2608     bool vsmt_user = !!spapr->vsmt;
2609     int kvm_smt = kvmppc_smt_threads();
2610     int ret;
2611     unsigned int smp_threads = ms->smp.threads;
2612 
2613     if (tcg_enabled()) {
2614         if (smp_threads > 1 &&
2615             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2616                                    spapr->max_compat_pvr)) {
2617             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2618             return;
2619         }
2620 
2621         if (smp_threads > 8) {
2622             error_setg(errp, "TCG cannot support more than 8 threads/core "
2623                        "on a pseries machine");
2624             return;
2625         }
2626     }
2627     if (!is_power_of_2(smp_threads)) {
2628         error_setg(errp, "Cannot support %d threads/core on a pseries "
2629                    "machine because it must be a power of 2", smp_threads);
2630         return;
2631     }
2632 
2633     /* Determine the VSMT mode to use: */
2634     if (vsmt_user) {
2635         if (spapr->vsmt < smp_threads) {
2636             error_setg(errp, "Cannot support VSMT mode %d"
2637                        " because it must be >= threads/core (%d)",
2638                        spapr->vsmt, smp_threads);
2639             return;
2640         }
2641         /* In this case, spapr->vsmt has been set by the command line */
2642     } else if (!smc->smp_threads_vsmt) {
2643         /*
2644          * Default VSMT value is tricky, because we need it to be as
2645          * consistent as possible (for migration), but this requires
2646          * changing it for at least some existing cases.  We pick 8 as
2647          * the value that we'd get with KVM on POWER8, the
2648          * overwhelmingly common case in production systems.
2649          */
2650         spapr->vsmt = MAX(8, smp_threads);
2651     } else {
2652         spapr->vsmt = smp_threads;
2653     }
2654 
2655     /* KVM: If necessary, set the SMT mode: */
2656     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2657         ret = kvmppc_set_smt_threads(spapr->vsmt);
2658         if (ret) {
2659             /* Looks like KVM isn't able to change VSMT mode */
2660             error_setg(&local_err,
2661                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2662                        spapr->vsmt, ret);
2663             /* We can live with that if the default one is big enough
2664              * for the number of threads, and a submultiple of the one
2665              * we want.  In this case we'll waste some vcpu ids, but
2666              * behaviour will be correct */
2667             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2668                 warn_report_err(local_err);
2669             } else {
2670                 if (!vsmt_user) {
2671                     error_append_hint(&local_err,
2672                                       "On PPC, a VM with %d threads/core"
2673                                       " on a host with %d threads/core"
2674                                       " requires the use of VSMT mode %d.\n",
2675                                       smp_threads, kvm_smt, spapr->vsmt);
2676                 }
2677                 kvmppc_error_append_smt_possible_hint(&local_err);
2678                 error_propagate(errp, local_err);
2679             }
2680         }
2681     }
2682     /* else TCG: nothing to do currently */
2683 }
2684 
2685 static void spapr_init_cpus(SpaprMachineState *spapr)
2686 {
2687     MachineState *machine = MACHINE(spapr);
2688     MachineClass *mc = MACHINE_GET_CLASS(machine);
2689     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2690     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2691     const CPUArchIdList *possible_cpus;
2692     unsigned int smp_cpus = machine->smp.cpus;
2693     unsigned int smp_threads = machine->smp.threads;
2694     unsigned int max_cpus = machine->smp.max_cpus;
2695     int boot_cores_nr = smp_cpus / smp_threads;
2696     int i;
2697 
2698     possible_cpus = mc->possible_cpu_arch_ids(machine);
2699     if (mc->has_hotpluggable_cpus) {
2700         if (smp_cpus % smp_threads) {
2701             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2702                          smp_cpus, smp_threads);
2703             exit(1);
2704         }
2705         if (max_cpus % smp_threads) {
2706             error_report("max_cpus (%u) must be multiple of threads (%u)",
2707                          max_cpus, smp_threads);
2708             exit(1);
2709         }
2710     } else {
2711         if (max_cpus != smp_cpus) {
2712             error_report("This machine version does not support CPU hotplug");
2713             exit(1);
2714         }
2715         boot_cores_nr = possible_cpus->len;
2716     }
2717 
2718     if (smc->pre_2_10_has_unused_icps) {
2719         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2720             /* Dummy entries get deregistered when real ICPState objects
2721              * are registered during CPU core hotplug.
2722              */
2723             pre_2_10_vmstate_register_dummy_icp(i);
2724         }
2725     }
2726 
2727     for (i = 0; i < possible_cpus->len; i++) {
2728         int core_id = i * smp_threads;
2729 
2730         if (mc->has_hotpluggable_cpus) {
2731             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2732                                    spapr_vcpu_id(spapr, core_id));
2733         }
2734 
2735         if (i < boot_cores_nr) {
2736             Object *core  = object_new(type);
2737             int nr_threads = smp_threads;
2738 
2739             /* Handle the partially filled core for older machine types */
2740             if ((i + 1) * smp_threads >= smp_cpus) {
2741                 nr_threads = smp_cpus - i * smp_threads;
2742             }
2743 
2744             object_property_set_int(core, "nr-threads", nr_threads,
2745                                     &error_fatal);
2746             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2747                                     &error_fatal);
2748             qdev_realize(DEVICE(core), NULL, &error_fatal);
2749 
2750             object_unref(core);
2751         }
2752     }
2753 }
2754 
2755 static PCIHostState *spapr_create_default_phb(void)
2756 {
2757     DeviceState *dev;
2758 
2759     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2760     qdev_prop_set_uint32(dev, "index", 0);
2761     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2762 
2763     return PCI_HOST_BRIDGE(dev);
2764 }
2765 
2766 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2767 {
2768     MachineState *machine = MACHINE(spapr);
2769     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2770     hwaddr rma_size = machine->ram_size;
2771     hwaddr node0_size = spapr_node0_size(machine);
2772 
2773     /* RMA has to fit in the first NUMA node */
2774     rma_size = MIN(rma_size, node0_size);
2775 
2776     /*
2777      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2778      * never exceed that
2779      */
2780     rma_size = MIN(rma_size, 1 * TiB);
2781 
2782     /*
2783      * Clamp the RMA size based on machine type.  This is for
2784      * migration compatibility with older qemu versions, which limited
2785      * the RMA size for complicated and mostly bad reasons.
2786      */
2787     if (smc->rma_limit) {
2788         rma_size = MIN(rma_size, smc->rma_limit);
2789     }
2790 
2791     if (rma_size < MIN_RMA_SLOF) {
2792         error_setg(errp,
2793                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2794                    "ldMiB guest RMA (Real Mode Area memory)",
2795                    MIN_RMA_SLOF / MiB);
2796         return 0;
2797     }
2798 
2799     return rma_size;
2800 }
2801 
2802 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2803 {
2804     MachineState *machine = MACHINE(spapr);
2805     int i;
2806 
2807     for (i = 0; i < machine->ram_slots; i++) {
2808         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2809     }
2810 }
2811 
2812 /* pSeries LPAR / sPAPR hardware init */
2813 static void spapr_machine_init(MachineState *machine)
2814 {
2815     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2816     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2817     MachineClass *mc = MACHINE_GET_CLASS(machine);
2818     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2819     const char *bios_name = machine->firmware ?: bios_default;
2820     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2821     const char *kernel_filename = machine->kernel_filename;
2822     const char *initrd_filename = machine->initrd_filename;
2823     PCIHostState *phb;
2824     bool has_vga;
2825     int i;
2826     MemoryRegion *sysmem = get_system_memory();
2827     long load_limit, fw_size;
2828     Error *resize_hpt_err = NULL;
2829     NICInfo *nd;
2830 
2831     if (!filename) {
2832         error_report("Could not find LPAR firmware '%s'", bios_name);
2833         exit(1);
2834     }
2835     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2836     if (fw_size <= 0) {
2837         error_report("Could not load LPAR firmware '%s'", filename);
2838         exit(1);
2839     }
2840 
2841     /*
2842      * if Secure VM (PEF) support is configured, then initialize it
2843      */
2844     pef_kvm_init(machine->cgs, &error_fatal);
2845 
2846     msi_nonbroken = true;
2847 
2848     QLIST_INIT(&spapr->phbs);
2849     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2850 
2851     /* Determine capabilities to run with */
2852     spapr_caps_init(spapr);
2853 
2854     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2855     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2856         /*
2857          * If the user explicitly requested a mode we should either
2858          * supply it, or fail completely (which we do below).  But if
2859          * it's not set explicitly, we reset our mode to something
2860          * that works
2861          */
2862         if (resize_hpt_err) {
2863             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2864             error_free(resize_hpt_err);
2865             resize_hpt_err = NULL;
2866         } else {
2867             spapr->resize_hpt = smc->resize_hpt_default;
2868         }
2869     }
2870 
2871     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2872 
2873     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2874         /*
2875          * User requested HPT resize, but this host can't supply it.  Bail out
2876          */
2877         error_report_err(resize_hpt_err);
2878         exit(1);
2879     }
2880     error_free(resize_hpt_err);
2881 
2882     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2883 
2884     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2885     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2886 
2887     /*
2888      * VSMT must be set in order to be able to compute VCPU ids, ie to
2889      * call spapr_max_server_number() or spapr_vcpu_id().
2890      */
2891     spapr_set_vsmt_mode(spapr, &error_fatal);
2892 
2893     /* Set up Interrupt Controller before we create the VCPUs */
2894     spapr_irq_init(spapr, &error_fatal);
2895 
2896     /* Set up containers for ibm,client-architecture-support negotiated options
2897      */
2898     spapr->ov5 = spapr_ovec_new();
2899     spapr->ov5_cas = spapr_ovec_new();
2900 
2901     if (smc->dr_lmb_enabled) {
2902         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2903         spapr_validate_node_memory(machine, &error_fatal);
2904     }
2905 
2906     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2907 
2908     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2909     if (!smc->pre_6_2_numa_affinity) {
2910         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2911     }
2912 
2913     /* advertise support for dedicated HP event source to guests */
2914     if (spapr->use_hotplug_event_source) {
2915         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2916     }
2917 
2918     /* advertise support for HPT resizing */
2919     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2920         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2921     }
2922 
2923     /* advertise support for ibm,dyamic-memory-v2 */
2924     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2925 
2926     /* advertise XIVE on POWER9 machines */
2927     if (spapr->irq->xive) {
2928         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2929     }
2930 
2931     /* init CPUs */
2932     spapr_init_cpus(spapr);
2933 
2934     /* Init numa_assoc_array */
2935     spapr_numa_associativity_init(spapr, machine);
2936 
2937     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2938         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2939                               spapr->max_compat_pvr)) {
2940         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2941         /* KVM and TCG always allow GTSE with radix... */
2942         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2943     }
2944     /* ... but not with hash (currently). */
2945 
2946     if (kvm_enabled()) {
2947         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2948         kvmppc_enable_logical_ci_hcalls();
2949         kvmppc_enable_set_mode_hcall();
2950 
2951         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2952         kvmppc_enable_clear_ref_mod_hcalls();
2953 
2954         /* Enable H_PAGE_INIT */
2955         kvmppc_enable_h_page_init();
2956     }
2957 
2958     /* map RAM */
2959     memory_region_add_subregion(sysmem, 0, machine->ram);
2960 
2961     /* initialize hotplug memory address space */
2962     if (machine->ram_size < machine->maxram_size) {
2963         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2964         hwaddr device_mem_base;
2965 
2966         /*
2967          * Limit the number of hotpluggable memory slots to half the number
2968          * slots that KVM supports, leaving the other half for PCI and other
2969          * devices. However ensure that number of slots doesn't drop below 32.
2970          */
2971         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2972                            SPAPR_MAX_RAM_SLOTS;
2973 
2974         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2975             max_memslots = SPAPR_MAX_RAM_SLOTS;
2976         }
2977         if (machine->ram_slots > max_memslots) {
2978             error_report("Specified number of memory slots %"
2979                          PRIu64" exceeds max supported %d",
2980                          machine->ram_slots, max_memslots);
2981             exit(1);
2982         }
2983 
2984         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
2985         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
2986     }
2987 
2988     if (smc->dr_lmb_enabled) {
2989         spapr_create_lmb_dr_connectors(spapr);
2990     }
2991 
2992     if (mc->nvdimm_supported) {
2993         spapr_create_nvdimm_dr_connectors(spapr);
2994     }
2995 
2996     /* Set up RTAS event infrastructure */
2997     spapr_events_init(spapr);
2998 
2999     /* Set up the RTC RTAS interfaces */
3000     spapr_rtc_create(spapr);
3001 
3002     /* Set up VIO bus */
3003     spapr->vio_bus = spapr_vio_bus_init();
3004 
3005     for (i = 0; serial_hd(i); i++) {
3006         spapr_vty_create(spapr->vio_bus, serial_hd(i));
3007     }
3008 
3009     /* We always have at least the nvram device on VIO */
3010     spapr_create_nvram(spapr);
3011 
3012     /*
3013      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3014      * connectors (described in root DT node's "ibm,drc-types" property)
3015      * are pre-initialized here. additional child connectors (such as
3016      * connectors for a PHBs PCI slots) are added as needed during their
3017      * parent's realization.
3018      */
3019     if (smc->dr_phb_enabled) {
3020         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3021             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3022         }
3023     }
3024 
3025     /* Set up PCI */
3026     spapr_pci_rtas_init();
3027 
3028     phb = spapr_create_default_phb();
3029 
3030     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3031         spapr_vlan_create(spapr->vio_bus, nd);
3032     }
3033 
3034     pci_init_nic_devices(phb->bus, NULL);
3035 
3036     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3037         spapr_vscsi_create(spapr->vio_bus);
3038     }
3039 
3040     /* Graphics */
3041     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3042     if (has_vga) {
3043         spapr->want_stdout_path = !machine->enable_graphics;
3044         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3045     } else {
3046         spapr->want_stdout_path = true;
3047     }
3048 
3049     if (machine->usb) {
3050         if (smc->use_ohci_by_default) {
3051             pci_create_simple(phb->bus, -1, "pci-ohci");
3052         } else {
3053             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3054         }
3055 
3056         if (has_vga) {
3057             USBBus *usb_bus;
3058 
3059             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3060                                                               &error_abort));
3061             usb_create_simple(usb_bus, "usb-kbd");
3062             usb_create_simple(usb_bus, "usb-mouse");
3063         }
3064     }
3065 
3066     if (kernel_filename) {
3067         uint64_t loaded_addr = 0;
3068 
3069         spapr->kernel_size = load_elf(kernel_filename, NULL,
3070                                       translate_kernel_address, spapr,
3071                                       NULL, &loaded_addr, NULL, NULL, 1,
3072                                       PPC_ELF_MACHINE, 0, 0);
3073         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3074             spapr->kernel_size = load_elf(kernel_filename, NULL,
3075                                           translate_kernel_address, spapr,
3076                                           NULL, &loaded_addr, NULL, NULL, 0,
3077                                           PPC_ELF_MACHINE, 0, 0);
3078             spapr->kernel_le = spapr->kernel_size > 0;
3079         }
3080         if (spapr->kernel_size < 0) {
3081             error_report("error loading %s: %s", kernel_filename,
3082                          load_elf_strerror(spapr->kernel_size));
3083             exit(1);
3084         }
3085 
3086         if (spapr->kernel_addr != loaded_addr) {
3087             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3088                         " to 0x%"PRIx64,
3089                         spapr->kernel_addr, loaded_addr);
3090             spapr->kernel_addr = loaded_addr;
3091         }
3092 
3093         /* load initrd */
3094         if (initrd_filename) {
3095             /* Try to locate the initrd in the gap between the kernel
3096              * and the firmware. Add a bit of space just in case
3097              */
3098             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3099                                   + 0x1ffff) & ~0xffff;
3100             spapr->initrd_size = load_image_targphys(initrd_filename,
3101                                                      spapr->initrd_base,
3102                                                      load_limit
3103                                                      - spapr->initrd_base);
3104             if (spapr->initrd_size < 0) {
3105                 error_report("could not load initial ram disk '%s'",
3106                              initrd_filename);
3107                 exit(1);
3108             }
3109         }
3110     }
3111 
3112     /* FIXME: Should register things through the MachineState's qdev
3113      * interface, this is a legacy from the sPAPREnvironment structure
3114      * which predated MachineState but had a similar function */
3115     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3116     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3117                          &savevm_htab_handlers, spapr);
3118 
3119     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3120 
3121     qemu_register_boot_set(spapr_boot_set, spapr);
3122 
3123     /*
3124      * Nothing needs to be done to resume a suspended guest because
3125      * suspending does not change the machine state, so no need for
3126      * a ->wakeup method.
3127      */
3128     qemu_register_wakeup_support();
3129 
3130     if (kvm_enabled()) {
3131         /* to stop and start vmclock */
3132         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3133                                          &spapr->tb);
3134 
3135         kvmppc_spapr_enable_inkernel_multitce();
3136     }
3137 
3138     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3139     if (spapr->vof) {
3140         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3141         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3142     }
3143 
3144     spapr_watchdog_init(spapr);
3145 }
3146 
3147 #define DEFAULT_KVM_TYPE "auto"
3148 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3149 {
3150     /*
3151      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3152      * accommodate the 'HV' and 'PV' formats that exists in the
3153      * wild. The 'auto' mode is being introduced already as
3154      * lower-case, thus we don't need to bother checking for
3155      * "AUTO".
3156      */
3157     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3158         return 0;
3159     }
3160 
3161     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3162         return 1;
3163     }
3164 
3165     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3166         return 2;
3167     }
3168 
3169     error_report("Unknown kvm-type specified '%s'", vm_type);
3170     return -1;
3171 }
3172 
3173 /*
3174  * Implementation of an interface to adjust firmware path
3175  * for the bootindex property handling.
3176  */
3177 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3178                                    DeviceState *dev)
3179 {
3180 #define CAST(type, obj, name) \
3181     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3182     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3183     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3184     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3185     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3186 
3187     if (d && bus) {
3188         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3189         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3190         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3191 
3192         if (spapr) {
3193             /*
3194              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3195              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3196              * 0x8000 | (target << 8) | (bus << 5) | lun
3197              * (see the "Logical unit addressing format" table in SAM5)
3198              */
3199             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3200             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3201                                    (uint64_t)id << 48);
3202         } else if (virtio) {
3203             /*
3204              * We use SRP luns of the form 01000000 | (target << 8) | lun
3205              * in the top 32 bits of the 64-bit LUN
3206              * Note: the quote above is from SLOF and it is wrong,
3207              * the actual binding is:
3208              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3209              */
3210             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3211             if (d->lun >= 256) {
3212                 /* Use the LUN "flat space addressing method" */
3213                 id |= 0x4000;
3214             }
3215             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3216                                    (uint64_t)id << 32);
3217         } else if (usb) {
3218             /*
3219              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3220              * in the top 32 bits of the 64-bit LUN
3221              */
3222             unsigned usb_port = atoi(usb->port->path);
3223             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3224             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3225                                    (uint64_t)id << 32);
3226         }
3227     }
3228 
3229     /*
3230      * SLOF probes the USB devices, and if it recognizes that the device is a
3231      * storage device, it changes its name to "storage" instead of "usb-host",
3232      * and additionally adds a child node for the SCSI LUN, so the correct
3233      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3234      */
3235     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3236         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3237         if (usb_device_is_scsi_storage(usbdev)) {
3238             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3239         }
3240     }
3241 
3242     if (phb) {
3243         /* Replace "pci" with "pci@800000020000000" */
3244         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3245     }
3246 
3247     if (vsc) {
3248         /* Same logic as virtio above */
3249         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3250         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3251     }
3252 
3253     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3254         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3255         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3256         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3257     }
3258 
3259     if (pcidev) {
3260         return spapr_pci_fw_dev_name(pcidev);
3261     }
3262 
3263     return NULL;
3264 }
3265 
3266 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3267 {
3268     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3269 
3270     return g_strdup(spapr->kvm_type);
3271 }
3272 
3273 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3274 {
3275     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3276 
3277     g_free(spapr->kvm_type);
3278     spapr->kvm_type = g_strdup(value);
3279 }
3280 
3281 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3282 {
3283     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3284 
3285     return spapr->use_hotplug_event_source;
3286 }
3287 
3288 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3289                                             Error **errp)
3290 {
3291     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3292 
3293     spapr->use_hotplug_event_source = value;
3294 }
3295 
3296 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3297 {
3298     return true;
3299 }
3300 
3301 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3302 {
3303     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3304 
3305     switch (spapr->resize_hpt) {
3306     case SPAPR_RESIZE_HPT_DEFAULT:
3307         return g_strdup("default");
3308     case SPAPR_RESIZE_HPT_DISABLED:
3309         return g_strdup("disabled");
3310     case SPAPR_RESIZE_HPT_ENABLED:
3311         return g_strdup("enabled");
3312     case SPAPR_RESIZE_HPT_REQUIRED:
3313         return g_strdup("required");
3314     }
3315     g_assert_not_reached();
3316 }
3317 
3318 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3319 {
3320     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3321 
3322     if (strcmp(value, "default") == 0) {
3323         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3324     } else if (strcmp(value, "disabled") == 0) {
3325         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3326     } else if (strcmp(value, "enabled") == 0) {
3327         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3328     } else if (strcmp(value, "required") == 0) {
3329         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3330     } else {
3331         error_setg(errp, "Bad value for \"resize-hpt\" property");
3332     }
3333 }
3334 
3335 static bool spapr_get_vof(Object *obj, Error **errp)
3336 {
3337     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3338 
3339     return spapr->vof != NULL;
3340 }
3341 
3342 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3343 {
3344     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3345 
3346     if (spapr->vof) {
3347         vof_cleanup(spapr->vof);
3348         g_free(spapr->vof);
3349         spapr->vof = NULL;
3350     }
3351     if (!value) {
3352         return;
3353     }
3354     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3355 }
3356 
3357 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3358 {
3359     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3360 
3361     if (spapr->irq == &spapr_irq_xics_legacy) {
3362         return g_strdup("legacy");
3363     } else if (spapr->irq == &spapr_irq_xics) {
3364         return g_strdup("xics");
3365     } else if (spapr->irq == &spapr_irq_xive) {
3366         return g_strdup("xive");
3367     } else if (spapr->irq == &spapr_irq_dual) {
3368         return g_strdup("dual");
3369     }
3370     g_assert_not_reached();
3371 }
3372 
3373 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3374 {
3375     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3376 
3377     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3378         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3379         return;
3380     }
3381 
3382     /* The legacy IRQ backend can not be set */
3383     if (strcmp(value, "xics") == 0) {
3384         spapr->irq = &spapr_irq_xics;
3385     } else if (strcmp(value, "xive") == 0) {
3386         spapr->irq = &spapr_irq_xive;
3387     } else if (strcmp(value, "dual") == 0) {
3388         spapr->irq = &spapr_irq_dual;
3389     } else {
3390         error_setg(errp, "Bad value for \"ic-mode\" property");
3391     }
3392 }
3393 
3394 static char *spapr_get_host_model(Object *obj, Error **errp)
3395 {
3396     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3397 
3398     return g_strdup(spapr->host_model);
3399 }
3400 
3401 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3402 {
3403     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3404 
3405     g_free(spapr->host_model);
3406     spapr->host_model = g_strdup(value);
3407 }
3408 
3409 static char *spapr_get_host_serial(Object *obj, Error **errp)
3410 {
3411     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3412 
3413     return g_strdup(spapr->host_serial);
3414 }
3415 
3416 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3417 {
3418     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3419 
3420     g_free(spapr->host_serial);
3421     spapr->host_serial = g_strdup(value);
3422 }
3423 
3424 static void spapr_instance_init(Object *obj)
3425 {
3426     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3427     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3428     MachineState *ms = MACHINE(spapr);
3429     MachineClass *mc = MACHINE_GET_CLASS(ms);
3430 
3431     /*
3432      * NVDIMM support went live in 5.1 without considering that, in
3433      * other archs, the user needs to enable NVDIMM support with the
3434      * 'nvdimm' machine option and the default behavior is NVDIMM
3435      * support disabled. It is too late to roll back to the standard
3436      * behavior without breaking 5.1 guests.
3437      */
3438     if (mc->nvdimm_supported) {
3439         ms->nvdimms_state->is_enabled = true;
3440     }
3441 
3442     spapr->htab_fd = -1;
3443     spapr->use_hotplug_event_source = true;
3444     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3445     object_property_add_str(obj, "kvm-type",
3446                             spapr_get_kvm_type, spapr_set_kvm_type);
3447     object_property_set_description(obj, "kvm-type",
3448                                     "Specifies the KVM virtualization mode (auto,"
3449                                     " hv, pr). Defaults to 'auto'. This mode will use"
3450                                     " any available KVM module loaded in the host,"
3451                                     " where kvm_hv takes precedence if both kvm_hv and"
3452                                     " kvm_pr are loaded.");
3453     object_property_add_bool(obj, "modern-hotplug-events",
3454                             spapr_get_modern_hotplug_events,
3455                             spapr_set_modern_hotplug_events);
3456     object_property_set_description(obj, "modern-hotplug-events",
3457                                     "Use dedicated hotplug event mechanism in"
3458                                     " place of standard EPOW events when possible"
3459                                     " (required for memory hot-unplug support)");
3460     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3461                             "Maximum permitted CPU compatibility mode");
3462 
3463     object_property_add_str(obj, "resize-hpt",
3464                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3465     object_property_set_description(obj, "resize-hpt",
3466                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3467     object_property_add_uint32_ptr(obj, "vsmt",
3468                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3469     object_property_set_description(obj, "vsmt",
3470                                     "Virtual SMT: KVM behaves as if this were"
3471                                     " the host's SMT mode");
3472 
3473     object_property_add_bool(obj, "vfio-no-msix-emulation",
3474                              spapr_get_msix_emulation, NULL);
3475 
3476     object_property_add_uint64_ptr(obj, "kernel-addr",
3477                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3478     object_property_set_description(obj, "kernel-addr",
3479                                     stringify(KERNEL_LOAD_ADDR)
3480                                     " for -kernel is the default");
3481     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3482 
3483     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3484     object_property_set_description(obj, "x-vof",
3485                                     "Enable Virtual Open Firmware (experimental)");
3486 
3487     /* The machine class defines the default interrupt controller mode */
3488     spapr->irq = smc->irq;
3489     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3490                             spapr_set_ic_mode);
3491     object_property_set_description(obj, "ic-mode",
3492                  "Specifies the interrupt controller mode (xics, xive, dual)");
3493 
3494     object_property_add_str(obj, "host-model",
3495         spapr_get_host_model, spapr_set_host_model);
3496     object_property_set_description(obj, "host-model",
3497         "Host model to advertise in guest device tree");
3498     object_property_add_str(obj, "host-serial",
3499         spapr_get_host_serial, spapr_set_host_serial);
3500     object_property_set_description(obj, "host-serial",
3501         "Host serial number to advertise in guest device tree");
3502 }
3503 
3504 static void spapr_machine_finalizefn(Object *obj)
3505 {
3506     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3507 
3508     g_free(spapr->kvm_type);
3509 }
3510 
3511 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3512 {
3513     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3514     CPUPPCState *env = cpu_env(cs);
3515 
3516     cpu_synchronize_state(cs);
3517     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3518     if (spapr->fwnmi_system_reset_addr != -1) {
3519         uint64_t rtas_addr, addr;
3520 
3521         /* get rtas addr from fdt */
3522         rtas_addr = spapr_get_rtas_addr();
3523         if (!rtas_addr) {
3524             qemu_system_guest_panicked(NULL);
3525             return;
3526         }
3527 
3528         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3529         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3530         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3531         env->gpr[3] = addr;
3532     }
3533     ppc_cpu_do_system_reset(cs);
3534     if (spapr->fwnmi_system_reset_addr != -1) {
3535         env->nip = spapr->fwnmi_system_reset_addr;
3536     }
3537 }
3538 
3539 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3540 {
3541     CPUState *cs;
3542 
3543     CPU_FOREACH(cs) {
3544         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3545     }
3546 }
3547 
3548 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3549                           void *fdt, int *fdt_start_offset, Error **errp)
3550 {
3551     uint64_t addr;
3552     uint32_t node;
3553 
3554     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3555     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3556                                     &error_abort);
3557     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3558                                              SPAPR_MEMORY_BLOCK_SIZE);
3559     return 0;
3560 }
3561 
3562 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3563                            bool dedicated_hp_event_source)
3564 {
3565     SpaprDrc *drc;
3566     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3567     int i;
3568     uint64_t addr = addr_start;
3569     bool hotplugged = spapr_drc_hotplugged(dev);
3570 
3571     for (i = 0; i < nr_lmbs; i++) {
3572         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3573                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3574         g_assert(drc);
3575 
3576         /*
3577          * memory_device_get_free_addr() provided a range of free addresses
3578          * that doesn't overlap with any existing mapping at pre-plug. The
3579          * corresponding LMB DRCs are thus assumed to be all attachable.
3580          */
3581         spapr_drc_attach(drc, dev);
3582         if (!hotplugged) {
3583             spapr_drc_reset(drc);
3584         }
3585         addr += SPAPR_MEMORY_BLOCK_SIZE;
3586     }
3587     /* send hotplug notification to the
3588      * guest only in case of hotplugged memory
3589      */
3590     if (hotplugged) {
3591         if (dedicated_hp_event_source) {
3592             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3593                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3594             g_assert(drc);
3595             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3596                                                    nr_lmbs,
3597                                                    spapr_drc_index(drc));
3598         } else {
3599             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3600                                            nr_lmbs);
3601         }
3602     }
3603 }
3604 
3605 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3606 {
3607     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3608     PCDIMMDevice *dimm = PC_DIMM(dev);
3609     uint64_t size, addr;
3610     int64_t slot;
3611     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3612 
3613     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3614 
3615     pc_dimm_plug(dimm, MACHINE(ms));
3616 
3617     if (!is_nvdimm) {
3618         addr = object_property_get_uint(OBJECT(dimm),
3619                                         PC_DIMM_ADDR_PROP, &error_abort);
3620         spapr_add_lmbs(dev, addr, size,
3621                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3622     } else {
3623         slot = object_property_get_int(OBJECT(dimm),
3624                                        PC_DIMM_SLOT_PROP, &error_abort);
3625         /* We should have valid slot number at this point */
3626         g_assert(slot >= 0);
3627         spapr_add_nvdimm(dev, slot);
3628     }
3629 }
3630 
3631 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3632                                   Error **errp)
3633 {
3634     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3635     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3636     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3637     PCDIMMDevice *dimm = PC_DIMM(dev);
3638     Error *local_err = NULL;
3639     uint64_t size;
3640     Object *memdev;
3641     hwaddr pagesize;
3642 
3643     if (!smc->dr_lmb_enabled) {
3644         error_setg(errp, "Memory hotplug not supported for this machine");
3645         return;
3646     }
3647 
3648     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3649     if (local_err) {
3650         error_propagate(errp, local_err);
3651         return;
3652     }
3653 
3654     if (is_nvdimm) {
3655         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3656             return;
3657         }
3658     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3659         error_setg(errp, "Hotplugged memory size must be a multiple of "
3660                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3661         return;
3662     }
3663 
3664     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3665                                       &error_abort);
3666     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3667     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3668         return;
3669     }
3670 
3671     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3672 }
3673 
3674 struct SpaprDimmState {
3675     PCDIMMDevice *dimm;
3676     uint32_t nr_lmbs;
3677     QTAILQ_ENTRY(SpaprDimmState) next;
3678 };
3679 
3680 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3681                                                        PCDIMMDevice *dimm)
3682 {
3683     SpaprDimmState *dimm_state = NULL;
3684 
3685     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3686         if (dimm_state->dimm == dimm) {
3687             break;
3688         }
3689     }
3690     return dimm_state;
3691 }
3692 
3693 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3694                                                       uint32_t nr_lmbs,
3695                                                       PCDIMMDevice *dimm)
3696 {
3697     SpaprDimmState *ds = NULL;
3698 
3699     /*
3700      * If this request is for a DIMM whose removal had failed earlier
3701      * (due to guest's refusal to remove the LMBs), we would have this
3702      * dimm already in the pending_dimm_unplugs list. In that
3703      * case don't add again.
3704      */
3705     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3706     if (!ds) {
3707         ds = g_new0(SpaprDimmState, 1);
3708         ds->nr_lmbs = nr_lmbs;
3709         ds->dimm = dimm;
3710         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3711     }
3712     return ds;
3713 }
3714 
3715 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3716                                               SpaprDimmState *dimm_state)
3717 {
3718     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3719     g_free(dimm_state);
3720 }
3721 
3722 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3723                                                         PCDIMMDevice *dimm)
3724 {
3725     SpaprDrc *drc;
3726     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3727                                                   &error_abort);
3728     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3729     uint32_t avail_lmbs = 0;
3730     uint64_t addr_start, addr;
3731     int i;
3732 
3733     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3734                                           &error_abort);
3735 
3736     addr = addr_start;
3737     for (i = 0; i < nr_lmbs; i++) {
3738         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3739                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3740         g_assert(drc);
3741         if (drc->dev) {
3742             avail_lmbs++;
3743         }
3744         addr += SPAPR_MEMORY_BLOCK_SIZE;
3745     }
3746 
3747     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3748 }
3749 
3750 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3751 {
3752     SpaprDimmState *ds;
3753     PCDIMMDevice *dimm;
3754     SpaprDrc *drc;
3755     uint32_t nr_lmbs;
3756     uint64_t size, addr_start, addr;
3757     g_autofree char *qapi_error = NULL;
3758     int i;
3759 
3760     if (!dev) {
3761         return;
3762     }
3763 
3764     dimm = PC_DIMM(dev);
3765     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3766 
3767     /*
3768      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3769      * unplug state, but one of its DRC is marked as unplug_requested.
3770      * This is bad and weird enough to g_assert() out.
3771      */
3772     g_assert(ds);
3773 
3774     spapr_pending_dimm_unplugs_remove(spapr, ds);
3775 
3776     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3777     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3778 
3779     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3780                                           &error_abort);
3781 
3782     addr = addr_start;
3783     for (i = 0; i < nr_lmbs; i++) {
3784         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3785                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3786         g_assert(drc);
3787 
3788         drc->unplug_requested = false;
3789         addr += SPAPR_MEMORY_BLOCK_SIZE;
3790     }
3791 
3792     /*
3793      * Tell QAPI that something happened and the memory
3794      * hotunplug wasn't successful. Keep sending
3795      * MEM_UNPLUG_ERROR even while sending
3796      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3797      * MEM_UNPLUG_ERROR is due.
3798      */
3799     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3800                                  "for device %s", dev->id);
3801 
3802     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3803 
3804     qapi_event_send_device_unplug_guest_error(dev->id,
3805                                               dev->canonical_path);
3806 }
3807 
3808 /* Callback to be called during DRC release. */
3809 void spapr_lmb_release(DeviceState *dev)
3810 {
3811     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3812     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3813     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3814 
3815     /* This information will get lost if a migration occurs
3816      * during the unplug process. In this case recover it. */
3817     if (ds == NULL) {
3818         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3819         g_assert(ds);
3820         /* The DRC being examined by the caller at least must be counted */
3821         g_assert(ds->nr_lmbs);
3822     }
3823 
3824     if (--ds->nr_lmbs) {
3825         return;
3826     }
3827 
3828     /*
3829      * Now that all the LMBs have been removed by the guest, call the
3830      * unplug handler chain. This can never fail.
3831      */
3832     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3833     object_unparent(OBJECT(dev));
3834 }
3835 
3836 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3837 {
3838     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3839     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3840 
3841     /* We really shouldn't get this far without anything to unplug */
3842     g_assert(ds);
3843 
3844     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3845     qdev_unrealize(dev);
3846     spapr_pending_dimm_unplugs_remove(spapr, ds);
3847 }
3848 
3849 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3850                                         DeviceState *dev, Error **errp)
3851 {
3852     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3853     PCDIMMDevice *dimm = PC_DIMM(dev);
3854     uint32_t nr_lmbs;
3855     uint64_t size, addr_start, addr;
3856     int i;
3857     SpaprDrc *drc;
3858 
3859     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3860         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3861         return;
3862     }
3863 
3864     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3865     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3866 
3867     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3868                                           &error_abort);
3869 
3870     /*
3871      * An existing pending dimm state for this DIMM means that there is an
3872      * unplug operation in progress, waiting for the spapr_lmb_release
3873      * callback to complete the job (BQL can't cover that far). In this case,
3874      * bail out to avoid detaching DRCs that were already released.
3875      */
3876     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3877         error_setg(errp, "Memory unplug already in progress for device %s",
3878                    dev->id);
3879         return;
3880     }
3881 
3882     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3883 
3884     addr = addr_start;
3885     for (i = 0; i < nr_lmbs; i++) {
3886         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3887                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3888         g_assert(drc);
3889 
3890         spapr_drc_unplug_request(drc);
3891         addr += SPAPR_MEMORY_BLOCK_SIZE;
3892     }
3893 
3894     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3895                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3896     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3897                                               nr_lmbs, spapr_drc_index(drc));
3898 }
3899 
3900 /* Callback to be called during DRC release. */
3901 void spapr_core_release(DeviceState *dev)
3902 {
3903     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3904 
3905     /* Call the unplug handler chain. This can never fail. */
3906     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3907     object_unparent(OBJECT(dev));
3908 }
3909 
3910 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3911 {
3912     MachineState *ms = MACHINE(hotplug_dev);
3913     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3914     CPUCore *cc = CPU_CORE(dev);
3915     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3916 
3917     if (smc->pre_2_10_has_unused_icps) {
3918         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3919         int i;
3920 
3921         for (i = 0; i < cc->nr_threads; i++) {
3922             CPUState *cs = CPU(sc->threads[i]);
3923 
3924             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3925         }
3926     }
3927 
3928     assert(core_slot);
3929     core_slot->cpu = NULL;
3930     qdev_unrealize(dev);
3931 }
3932 
3933 static
3934 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3935                                Error **errp)
3936 {
3937     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3938     int index;
3939     SpaprDrc *drc;
3940     CPUCore *cc = CPU_CORE(dev);
3941 
3942     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3943         error_setg(errp, "Unable to find CPU core with core-id: %d",
3944                    cc->core_id);
3945         return;
3946     }
3947     if (index == 0) {
3948         error_setg(errp, "Boot CPU core may not be unplugged");
3949         return;
3950     }
3951 
3952     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3953                           spapr_vcpu_id(spapr, cc->core_id));
3954     g_assert(drc);
3955 
3956     if (!spapr_drc_unplug_requested(drc)) {
3957         spapr_drc_unplug_request(drc);
3958     }
3959 
3960     /*
3961      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3962      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3963      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3964      * attempt (e.g. the kernel will refuse to remove the last online
3965      * CPU), we will never attempt it again because unplug_requested
3966      * will still be 'true' in that case.
3967      */
3968     spapr_hotplug_req_remove_by_index(drc);
3969 }
3970 
3971 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3972                            void *fdt, int *fdt_start_offset, Error **errp)
3973 {
3974     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3975     CPUState *cs = CPU(core->threads[0]);
3976     PowerPCCPU *cpu = POWERPC_CPU(cs);
3977     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3978     int id = spapr_get_vcpu_id(cpu);
3979     g_autofree char *nodename = NULL;
3980     int offset;
3981 
3982     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3983     offset = fdt_add_subnode(fdt, 0, nodename);
3984 
3985     spapr_dt_cpu(cs, fdt, offset, spapr);
3986 
3987     /*
3988      * spapr_dt_cpu() does not fill the 'name' property in the
3989      * CPU node. The function is called during boot process, before
3990      * and after CAS, and overwriting the 'name' property written
3991      * by SLOF is not allowed.
3992      *
3993      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3994      * CPUs more compatible with the coldplugged ones, which have
3995      * the 'name' property. Linux Kernel also relies on this
3996      * property to identify CPU nodes.
3997      */
3998     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3999 
4000     *fdt_start_offset = offset;
4001     return 0;
4002 }
4003 
4004 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4005 {
4006     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4007     MachineClass *mc = MACHINE_GET_CLASS(spapr);
4008     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4009     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
4010     CPUCore *cc = CPU_CORE(dev);
4011     SpaprDrc *drc;
4012     CPUArchId *core_slot;
4013     int index;
4014     bool hotplugged = spapr_drc_hotplugged(dev);
4015     int i;
4016 
4017     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4018     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4019 
4020     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4021                           spapr_vcpu_id(spapr, cc->core_id));
4022 
4023     g_assert(drc || !mc->has_hotpluggable_cpus);
4024 
4025     if (drc) {
4026         /*
4027          * spapr_core_pre_plug() already buys us this is a brand new
4028          * core being plugged into a free slot. Nothing should already
4029          * be attached to the corresponding DRC.
4030          */
4031         spapr_drc_attach(drc, dev);
4032 
4033         if (hotplugged) {
4034             /*
4035              * Send hotplug notification interrupt to the guest only
4036              * in case of hotplugged CPUs.
4037              */
4038             spapr_hotplug_req_add_by_index(drc);
4039         } else {
4040             spapr_drc_reset(drc);
4041         }
4042     }
4043 
4044     core_slot->cpu = CPU(dev);
4045 
4046     /*
4047      * Set compatibility mode to match the boot CPU, which was either set
4048      * by the machine reset code or by CAS. This really shouldn't fail at
4049      * this point.
4050      */
4051     if (hotplugged) {
4052         for (i = 0; i < cc->nr_threads; i++) {
4053             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4054                            &error_abort);
4055         }
4056     }
4057 
4058     if (smc->pre_2_10_has_unused_icps) {
4059         for (i = 0; i < cc->nr_threads; i++) {
4060             CPUState *cs = CPU(core->threads[i]);
4061             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4062         }
4063     }
4064 }
4065 
4066 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4067                                 Error **errp)
4068 {
4069     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4070     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4071     CPUCore *cc = CPU_CORE(dev);
4072     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4073     const char *type = object_get_typename(OBJECT(dev));
4074     CPUArchId *core_slot;
4075     int index;
4076     unsigned int smp_threads = machine->smp.threads;
4077 
4078     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4079         error_setg(errp, "CPU hotplug not supported for this machine");
4080         return;
4081     }
4082 
4083     if (strcmp(base_core_type, type)) {
4084         error_setg(errp, "CPU core type should be %s", base_core_type);
4085         return;
4086     }
4087 
4088     if (cc->core_id % smp_threads) {
4089         error_setg(errp, "invalid core id %d", cc->core_id);
4090         return;
4091     }
4092 
4093     /*
4094      * In general we should have homogeneous threads-per-core, but old
4095      * (pre hotplug support) machine types allow the last core to have
4096      * reduced threads as a compatibility hack for when we allowed
4097      * total vcpus not a multiple of threads-per-core.
4098      */
4099     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4100         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4101                    smp_threads);
4102         return;
4103     }
4104 
4105     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4106     if (!core_slot) {
4107         error_setg(errp, "core id %d out of range", cc->core_id);
4108         return;
4109     }
4110 
4111     if (core_slot->cpu) {
4112         error_setg(errp, "core %d already populated", cc->core_id);
4113         return;
4114     }
4115 
4116     numa_cpu_pre_plug(core_slot, dev, errp);
4117 }
4118 
4119 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4120                           void *fdt, int *fdt_start_offset, Error **errp)
4121 {
4122     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4123     int intc_phandle;
4124 
4125     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4126     if (intc_phandle <= 0) {
4127         return -1;
4128     }
4129 
4130     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4131         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4132         return -1;
4133     }
4134 
4135     /* generally SLOF creates these, for hotplug it's up to QEMU */
4136     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4137 
4138     return 0;
4139 }
4140 
4141 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4142                                Error **errp)
4143 {
4144     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4145     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4146     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4147     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4148     SpaprDrc *drc;
4149 
4150     if (dev->hotplugged && !smc->dr_phb_enabled) {
4151         error_setg(errp, "PHB hotplug not supported for this machine");
4152         return false;
4153     }
4154 
4155     if (sphb->index == (uint32_t)-1) {
4156         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4157         return false;
4158     }
4159 
4160     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4161     if (drc && drc->dev) {
4162         error_setg(errp, "PHB %d already attached", sphb->index);
4163         return false;
4164     }
4165 
4166     /*
4167      * This will check that sphb->index doesn't exceed the maximum number of
4168      * PHBs for the current machine type.
4169      */
4170     return
4171         smc->phb_placement(spapr, sphb->index,
4172                            &sphb->buid, &sphb->io_win_addr,
4173                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4174                            windows_supported, sphb->dma_liobn,
4175                            errp);
4176 }
4177 
4178 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4179 {
4180     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4181     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4182     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4183     SpaprDrc *drc;
4184     bool hotplugged = spapr_drc_hotplugged(dev);
4185 
4186     if (!smc->dr_phb_enabled) {
4187         return;
4188     }
4189 
4190     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4191     /* hotplug hooks should check it's enabled before getting this far */
4192     assert(drc);
4193 
4194     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4195     spapr_drc_attach(drc, dev);
4196 
4197     if (hotplugged) {
4198         spapr_hotplug_req_add_by_index(drc);
4199     } else {
4200         spapr_drc_reset(drc);
4201     }
4202 }
4203 
4204 void spapr_phb_release(DeviceState *dev)
4205 {
4206     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4207 
4208     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4209     object_unparent(OBJECT(dev));
4210 }
4211 
4212 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4213 {
4214     qdev_unrealize(dev);
4215 }
4216 
4217 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4218                                      DeviceState *dev, Error **errp)
4219 {
4220     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4221     SpaprDrc *drc;
4222 
4223     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4224     assert(drc);
4225 
4226     if (!spapr_drc_unplug_requested(drc)) {
4227         spapr_drc_unplug_request(drc);
4228         spapr_hotplug_req_remove_by_index(drc);
4229     } else {
4230         error_setg(errp,
4231                    "PCI Host Bridge unplug already in progress for device %s",
4232                    dev->id);
4233     }
4234 }
4235 
4236 static
4237 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4238                               Error **errp)
4239 {
4240     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4241 
4242     if (spapr->tpm_proxy != NULL) {
4243         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4244         return false;
4245     }
4246 
4247     return true;
4248 }
4249 
4250 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4251 {
4252     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4253     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4254 
4255     /* Already checked in spapr_tpm_proxy_pre_plug() */
4256     g_assert(spapr->tpm_proxy == NULL);
4257 
4258     spapr->tpm_proxy = tpm_proxy;
4259 }
4260 
4261 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4262 {
4263     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4264 
4265     qdev_unrealize(dev);
4266     object_unparent(OBJECT(dev));
4267     spapr->tpm_proxy = NULL;
4268 }
4269 
4270 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4271                                       DeviceState *dev, Error **errp)
4272 {
4273     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4274         spapr_memory_plug(hotplug_dev, dev);
4275     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4276         spapr_core_plug(hotplug_dev, dev);
4277     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4278         spapr_phb_plug(hotplug_dev, dev);
4279     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4280         spapr_tpm_proxy_plug(hotplug_dev, dev);
4281     }
4282 }
4283 
4284 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4285                                         DeviceState *dev, Error **errp)
4286 {
4287     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4288         spapr_memory_unplug(hotplug_dev, dev);
4289     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4290         spapr_core_unplug(hotplug_dev, dev);
4291     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4292         spapr_phb_unplug(hotplug_dev, dev);
4293     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4294         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4295     }
4296 }
4297 
4298 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4299 {
4300     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4301         /*
4302          * CAS will process all pending unplug requests.
4303          *
4304          * HACK: a guest could theoretically have cleared all bits in OV5,
4305          * but none of the guests we care for do.
4306          */
4307         spapr_ovec_empty(spapr->ov5_cas);
4308 }
4309 
4310 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4311                                                 DeviceState *dev, Error **errp)
4312 {
4313     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4314     MachineClass *mc = MACHINE_GET_CLASS(sms);
4315     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4316 
4317     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4318         if (spapr_memory_hot_unplug_supported(sms)) {
4319             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4320         } else {
4321             error_setg(errp, "Memory hot unplug not supported for this guest");
4322         }
4323     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4324         if (!mc->has_hotpluggable_cpus) {
4325             error_setg(errp, "CPU hot unplug not supported on this machine");
4326             return;
4327         }
4328         spapr_core_unplug_request(hotplug_dev, dev, errp);
4329     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4330         if (!smc->dr_phb_enabled) {
4331             error_setg(errp, "PHB hot unplug not supported on this machine");
4332             return;
4333         }
4334         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4335     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4336         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4337     }
4338 }
4339 
4340 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4341                                           DeviceState *dev, Error **errp)
4342 {
4343     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4344         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4345     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4346         spapr_core_pre_plug(hotplug_dev, dev, errp);
4347     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4348         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4349     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4350         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4351     }
4352 }
4353 
4354 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4355                                                  DeviceState *dev)
4356 {
4357     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4358         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4359         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4360         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4361         return HOTPLUG_HANDLER(machine);
4362     }
4363     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4364         PCIDevice *pcidev = PCI_DEVICE(dev);
4365         PCIBus *root = pci_device_root_bus(pcidev);
4366         SpaprPhbState *phb =
4367             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4368                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4369 
4370         if (phb) {
4371             return HOTPLUG_HANDLER(phb);
4372         }
4373     }
4374     return NULL;
4375 }
4376 
4377 static CpuInstanceProperties
4378 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4379 {
4380     CPUArchId *core_slot;
4381     MachineClass *mc = MACHINE_GET_CLASS(machine);
4382 
4383     /* make sure possible_cpu are initialized */
4384     mc->possible_cpu_arch_ids(machine);
4385     /* get CPU core slot containing thread that matches cpu_index */
4386     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4387     assert(core_slot);
4388     return core_slot->props;
4389 }
4390 
4391 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4392 {
4393     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4394 }
4395 
4396 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4397 {
4398     int i;
4399     unsigned int smp_threads = machine->smp.threads;
4400     unsigned int smp_cpus = machine->smp.cpus;
4401     const char *core_type;
4402     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4403     MachineClass *mc = MACHINE_GET_CLASS(machine);
4404 
4405     if (!mc->has_hotpluggable_cpus) {
4406         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4407     }
4408     if (machine->possible_cpus) {
4409         assert(machine->possible_cpus->len == spapr_max_cores);
4410         return machine->possible_cpus;
4411     }
4412 
4413     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4414     if (!core_type) {
4415         error_report("Unable to find sPAPR CPU Core definition");
4416         exit(1);
4417     }
4418 
4419     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4420                              sizeof(CPUArchId) * spapr_max_cores);
4421     machine->possible_cpus->len = spapr_max_cores;
4422     for (i = 0; i < machine->possible_cpus->len; i++) {
4423         int core_id = i * smp_threads;
4424 
4425         machine->possible_cpus->cpus[i].type = core_type;
4426         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4427         machine->possible_cpus->cpus[i].arch_id = core_id;
4428         machine->possible_cpus->cpus[i].props.has_core_id = true;
4429         machine->possible_cpus->cpus[i].props.core_id = core_id;
4430     }
4431     return machine->possible_cpus;
4432 }
4433 
4434 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4435                                 uint64_t *buid, hwaddr *pio,
4436                                 hwaddr *mmio32, hwaddr *mmio64,
4437                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4438 {
4439     /*
4440      * New-style PHB window placement.
4441      *
4442      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4443      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4444      * windows.
4445      *
4446      * Some guest kernels can't work with MMIO windows above 1<<46
4447      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4448      *
4449      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4450      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4451      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4452      * 1TiB 64-bit MMIO windows for each PHB.
4453      */
4454     const uint64_t base_buid = 0x800000020000000ULL;
4455     int i;
4456 
4457     /* Sanity check natural alignments */
4458     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4459     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4460     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4461     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4462     /* Sanity check bounds */
4463     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4464                       SPAPR_PCI_MEM32_WIN_SIZE);
4465     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4466                       SPAPR_PCI_MEM64_WIN_SIZE);
4467 
4468     if (index >= SPAPR_MAX_PHBS) {
4469         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4470                    SPAPR_MAX_PHBS - 1);
4471         return false;
4472     }
4473 
4474     *buid = base_buid + index;
4475     for (i = 0; i < n_dma; ++i) {
4476         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4477     }
4478 
4479     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4480     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4481     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4482     return true;
4483 }
4484 
4485 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4486 {
4487     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4488 
4489     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4490 }
4491 
4492 static void spapr_ics_resend(XICSFabric *dev)
4493 {
4494     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4495 
4496     ics_resend(spapr->ics);
4497 }
4498 
4499 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4500 {
4501     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4502 
4503     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4504 }
4505 
4506 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4507                                  Monitor *mon)
4508 {
4509     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4510 
4511     spapr_irq_print_info(spapr, mon);
4512     monitor_printf(mon, "irqchip: %s\n",
4513                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4514 }
4515 
4516 /*
4517  * This is a XIVE only operation
4518  */
4519 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4520                            uint8_t nvt_blk, uint32_t nvt_idx,
4521                            bool cam_ignore, uint8_t priority,
4522                            uint32_t logic_serv, XiveTCTXMatch *match)
4523 {
4524     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4525     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4526     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4527     int count;
4528 
4529     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4530                            priority, logic_serv, match);
4531     if (count < 0) {
4532         return count;
4533     }
4534 
4535     /*
4536      * When we implement the save and restore of the thread interrupt
4537      * contexts in the enter/exit CPU handlers of the machine and the
4538      * escalations in QEMU, we should be able to handle non dispatched
4539      * vCPUs.
4540      *
4541      * Until this is done, the sPAPR machine should find at least one
4542      * matching context always.
4543      */
4544     if (count == 0) {
4545         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4546                       nvt_blk, nvt_idx);
4547     }
4548 
4549     return count;
4550 }
4551 
4552 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4553 {
4554     return cpu->vcpu_id;
4555 }
4556 
4557 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4558 {
4559     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4560     MachineState *ms = MACHINE(spapr);
4561     int vcpu_id;
4562 
4563     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4564 
4565     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4566         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4567         error_append_hint(errp, "Adjust the number of cpus to %d "
4568                           "or try to raise the number of threads per core\n",
4569                           vcpu_id * ms->smp.threads / spapr->vsmt);
4570         return false;
4571     }
4572 
4573     cpu->vcpu_id = vcpu_id;
4574     return true;
4575 }
4576 
4577 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4578 {
4579     CPUState *cs;
4580 
4581     CPU_FOREACH(cs) {
4582         PowerPCCPU *cpu = POWERPC_CPU(cs);
4583 
4584         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4585             return cpu;
4586         }
4587     }
4588 
4589     return NULL;
4590 }
4591 
4592 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4593 {
4594     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4595 
4596     return spapr_cpu->in_nested;
4597 }
4598 
4599 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4600 {
4601     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4602 
4603     /* These are only called by TCG, KVM maintains dispatch state */
4604 
4605     spapr_cpu->prod = false;
4606     if (spapr_cpu->vpa_addr) {
4607         CPUState *cs = CPU(cpu);
4608         uint32_t dispatch;
4609 
4610         dispatch = ldl_be_phys(cs->as,
4611                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4612         dispatch++;
4613         if ((dispatch & 1) != 0) {
4614             qemu_log_mask(LOG_GUEST_ERROR,
4615                           "VPA: incorrect dispatch counter value for "
4616                           "dispatched partition %u, correcting.\n", dispatch);
4617             dispatch++;
4618         }
4619         stl_be_phys(cs->as,
4620                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4621     }
4622 }
4623 
4624 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4625 {
4626     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4627 
4628     if (spapr_cpu->vpa_addr) {
4629         CPUState *cs = CPU(cpu);
4630         uint32_t dispatch;
4631 
4632         dispatch = ldl_be_phys(cs->as,
4633                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4634         dispatch++;
4635         if ((dispatch & 1) != 1) {
4636             qemu_log_mask(LOG_GUEST_ERROR,
4637                           "VPA: incorrect dispatch counter value for "
4638                           "preempted partition %u, correcting.\n", dispatch);
4639             dispatch++;
4640         }
4641         stl_be_phys(cs->as,
4642                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4643     }
4644 }
4645 
4646 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4647 {
4648     MachineClass *mc = MACHINE_CLASS(oc);
4649     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4650     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4651     NMIClass *nc = NMI_CLASS(oc);
4652     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4653     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4654     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4655     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4656     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4657     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4658 
4659     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4660     mc->ignore_boot_device_suffixes = true;
4661 
4662     /*
4663      * We set up the default / latest behaviour here.  The class_init
4664      * functions for the specific versioned machine types can override
4665      * these details for backwards compatibility
4666      */
4667     mc->init = spapr_machine_init;
4668     mc->reset = spapr_machine_reset;
4669     mc->block_default_type = IF_SCSI;
4670 
4671     /*
4672      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4673      * In TCG the limit is restricted by the range of CPU IPIs available.
4674      */
4675     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4676 
4677     mc->no_parallel = 1;
4678     mc->default_boot_order = "";
4679     mc->default_ram_size = 512 * MiB;
4680     mc->default_ram_id = "ppc_spapr.ram";
4681     mc->default_display = "std";
4682     mc->kvm_type = spapr_kvm_type;
4683     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4684     mc->pci_allow_0_address = true;
4685     assert(!mc->get_hotplug_handler);
4686     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4687     hc->pre_plug = spapr_machine_device_pre_plug;
4688     hc->plug = spapr_machine_device_plug;
4689     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4690     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4691     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4692     hc->unplug_request = spapr_machine_device_unplug_request;
4693     hc->unplug = spapr_machine_device_unplug;
4694 
4695     smc->dr_lmb_enabled = true;
4696     smc->update_dt_enabled = true;
4697     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4698     mc->has_hotpluggable_cpus = true;
4699     mc->nvdimm_supported = true;
4700     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4701     fwc->get_dev_path = spapr_get_fw_dev_path;
4702     nc->nmi_monitor_handler = spapr_nmi;
4703     smc->phb_placement = spapr_phb_placement;
4704     vhc->cpu_in_nested = spapr_cpu_in_nested;
4705     vhc->deliver_hv_excp = spapr_exit_nested;
4706     vhc->hypercall = emulate_spapr_hypercall;
4707     vhc->hpt_mask = spapr_hpt_mask;
4708     vhc->map_hptes = spapr_map_hptes;
4709     vhc->unmap_hptes = spapr_unmap_hptes;
4710     vhc->hpte_set_c = spapr_hpte_set_c;
4711     vhc->hpte_set_r = spapr_hpte_set_r;
4712     vhc->get_pate = spapr_get_pate;
4713     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4714     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4715     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4716     xic->ics_get = spapr_ics_get;
4717     xic->ics_resend = spapr_ics_resend;
4718     xic->icp_get = spapr_icp_get;
4719     ispc->print_info = spapr_pic_print_info;
4720     /* Force NUMA node memory size to be a multiple of
4721      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4722      * in which LMBs are represented and hot-added
4723      */
4724     mc->numa_mem_align_shift = 28;
4725     mc->auto_enable_numa = true;
4726 
4727     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4728     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4729     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4730     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4731     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4732     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4733     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4734     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4735     smc->default_caps.caps[SPAPR_CAP_NESTED_PAPR] = SPAPR_CAP_OFF;
4736     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4737     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4738     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4739     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4740 
4741     /*
4742      * This cap specifies whether the AIL 3 mode for
4743      * H_SET_RESOURCE is supported. The default is modified
4744      * by default_caps_with_cpu().
4745      */
4746     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4747     spapr_caps_add_properties(smc);
4748     smc->irq = &spapr_irq_dual;
4749     smc->dr_phb_enabled = true;
4750     smc->linux_pci_probe = true;
4751     smc->smp_threads_vsmt = true;
4752     smc->nr_xirqs = SPAPR_NR_XIRQS;
4753     xfc->match_nvt = spapr_match_nvt;
4754     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4755     vmc->quiesce = spapr_vof_quiesce;
4756     vmc->setprop = spapr_vof_setprop;
4757 }
4758 
4759 static const TypeInfo spapr_machine_info = {
4760     .name          = TYPE_SPAPR_MACHINE,
4761     .parent        = TYPE_MACHINE,
4762     .abstract      = true,
4763     .instance_size = sizeof(SpaprMachineState),
4764     .instance_init = spapr_instance_init,
4765     .instance_finalize = spapr_machine_finalizefn,
4766     .class_size    = sizeof(SpaprMachineClass),
4767     .class_init    = spapr_machine_class_init,
4768     .interfaces = (InterfaceInfo[]) {
4769         { TYPE_FW_PATH_PROVIDER },
4770         { TYPE_NMI },
4771         { TYPE_HOTPLUG_HANDLER },
4772         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4773         { TYPE_XICS_FABRIC },
4774         { TYPE_INTERRUPT_STATS_PROVIDER },
4775         { TYPE_XIVE_FABRIC },
4776         { TYPE_VOF_MACHINE_IF },
4777         { }
4778     },
4779 };
4780 
4781 static void spapr_machine_latest_class_options(MachineClass *mc)
4782 {
4783     mc->alias = "pseries";
4784     mc->is_default = true;
4785 }
4786 
4787 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4788     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4789                                                     void *data)      \
4790     {                                                                \
4791         MachineClass *mc = MACHINE_CLASS(oc);                        \
4792         spapr_machine_##suffix##_class_options(mc);                  \
4793         if (latest) {                                                \
4794             spapr_machine_latest_class_options(mc);                  \
4795         }                                                            \
4796     }                                                                \
4797     static const TypeInfo spapr_machine_##suffix##_info = {          \
4798         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4799         .parent = TYPE_SPAPR_MACHINE,                                \
4800         .class_init = spapr_machine_##suffix##_class_init,           \
4801     };                                                               \
4802     static void spapr_machine_register_##suffix(void)                \
4803     {                                                                \
4804         type_register(&spapr_machine_##suffix##_info);               \
4805     }                                                                \
4806     type_init(spapr_machine_register_##suffix)
4807 
4808 /*
4809  * pseries-9.0
4810  */
4811 static void spapr_machine_9_0_class_options(MachineClass *mc)
4812 {
4813     /* Defaults for the latest behaviour inherited from the base class */
4814 }
4815 
4816 DEFINE_SPAPR_MACHINE(9_0, "9.0", true);
4817 
4818 /*
4819  * pseries-8.2
4820  */
4821 static void spapr_machine_8_2_class_options(MachineClass *mc)
4822 {
4823     spapr_machine_9_0_class_options(mc);
4824     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4825 }
4826 
4827 DEFINE_SPAPR_MACHINE(8_2, "8.2", false);
4828 
4829 /*
4830  * pseries-8.1
4831  */
4832 static void spapr_machine_8_1_class_options(MachineClass *mc)
4833 {
4834     spapr_machine_8_2_class_options(mc);
4835     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4836 }
4837 
4838 DEFINE_SPAPR_MACHINE(8_1, "8.1", false);
4839 
4840 /*
4841  * pseries-8.0
4842  */
4843 static void spapr_machine_8_0_class_options(MachineClass *mc)
4844 {
4845     spapr_machine_8_1_class_options(mc);
4846     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4847 }
4848 
4849 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4850 
4851 /*
4852  * pseries-7.2
4853  */
4854 static void spapr_machine_7_2_class_options(MachineClass *mc)
4855 {
4856     spapr_machine_8_0_class_options(mc);
4857     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4858 }
4859 
4860 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4861 
4862 /*
4863  * pseries-7.1
4864  */
4865 static void spapr_machine_7_1_class_options(MachineClass *mc)
4866 {
4867     spapr_machine_7_2_class_options(mc);
4868     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4869 }
4870 
4871 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4872 
4873 /*
4874  * pseries-7.0
4875  */
4876 static void spapr_machine_7_0_class_options(MachineClass *mc)
4877 {
4878     spapr_machine_7_1_class_options(mc);
4879     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4880 }
4881 
4882 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4883 
4884 /*
4885  * pseries-6.2
4886  */
4887 static void spapr_machine_6_2_class_options(MachineClass *mc)
4888 {
4889     spapr_machine_7_0_class_options(mc);
4890     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4891 }
4892 
4893 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4894 
4895 /*
4896  * pseries-6.1
4897  */
4898 static void spapr_machine_6_1_class_options(MachineClass *mc)
4899 {
4900     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4901 
4902     spapr_machine_6_2_class_options(mc);
4903     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4904     smc->pre_6_2_numa_affinity = true;
4905     mc->smp_props.prefer_sockets = true;
4906 }
4907 
4908 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4909 
4910 /*
4911  * pseries-6.0
4912  */
4913 static void spapr_machine_6_0_class_options(MachineClass *mc)
4914 {
4915     spapr_machine_6_1_class_options(mc);
4916     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4917 }
4918 
4919 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4920 
4921 /*
4922  * pseries-5.2
4923  */
4924 static void spapr_machine_5_2_class_options(MachineClass *mc)
4925 {
4926     spapr_machine_6_0_class_options(mc);
4927     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4928 }
4929 
4930 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4931 
4932 /*
4933  * pseries-5.1
4934  */
4935 static void spapr_machine_5_1_class_options(MachineClass *mc)
4936 {
4937     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4938 
4939     spapr_machine_5_2_class_options(mc);
4940     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4941     smc->pre_5_2_numa_associativity = true;
4942 }
4943 
4944 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4945 
4946 /*
4947  * pseries-5.0
4948  */
4949 static void spapr_machine_5_0_class_options(MachineClass *mc)
4950 {
4951     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4952     static GlobalProperty compat[] = {
4953         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4954     };
4955 
4956     spapr_machine_5_1_class_options(mc);
4957     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4958     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4959     mc->numa_mem_supported = true;
4960     smc->pre_5_1_assoc_refpoints = true;
4961 }
4962 
4963 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4964 
4965 /*
4966  * pseries-4.2
4967  */
4968 static void spapr_machine_4_2_class_options(MachineClass *mc)
4969 {
4970     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4971 
4972     spapr_machine_5_0_class_options(mc);
4973     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4974     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4975     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4976     smc->rma_limit = 16 * GiB;
4977     mc->nvdimm_supported = false;
4978 }
4979 
4980 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4981 
4982 /*
4983  * pseries-4.1
4984  */
4985 static void spapr_machine_4_1_class_options(MachineClass *mc)
4986 {
4987     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4988     static GlobalProperty compat[] = {
4989         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4990         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4991     };
4992 
4993     spapr_machine_4_2_class_options(mc);
4994     smc->linux_pci_probe = false;
4995     smc->smp_threads_vsmt = false;
4996     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4997     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4998 }
4999 
5000 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
5001 
5002 /*
5003  * pseries-4.0
5004  */
5005 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
5006                               uint64_t *buid, hwaddr *pio,
5007                               hwaddr *mmio32, hwaddr *mmio64,
5008                               unsigned n_dma, uint32_t *liobns, Error **errp)
5009 {
5010     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
5011                              liobns, errp)) {
5012         return false;
5013     }
5014     return true;
5015 }
5016 static void spapr_machine_4_0_class_options(MachineClass *mc)
5017 {
5018     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5019 
5020     spapr_machine_4_1_class_options(mc);
5021     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5022     smc->phb_placement = phb_placement_4_0;
5023     smc->irq = &spapr_irq_xics;
5024     smc->pre_4_1_migration = true;
5025 }
5026 
5027 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
5028 
5029 /*
5030  * pseries-3.1
5031  */
5032 static void spapr_machine_3_1_class_options(MachineClass *mc)
5033 {
5034     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5035 
5036     spapr_machine_4_0_class_options(mc);
5037     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5038 
5039     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5040     smc->update_dt_enabled = false;
5041     smc->dr_phb_enabled = false;
5042     smc->broken_host_serial_model = true;
5043     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5044     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5045     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5046     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5047 }
5048 
5049 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
5050 
5051 /*
5052  * pseries-3.0
5053  */
5054 
5055 static void spapr_machine_3_0_class_options(MachineClass *mc)
5056 {
5057     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5058 
5059     spapr_machine_3_1_class_options(mc);
5060     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5061 
5062     smc->legacy_irq_allocation = true;
5063     smc->nr_xirqs = 0x400;
5064     smc->irq = &spapr_irq_xics_legacy;
5065 }
5066 
5067 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
5068 
5069 /*
5070  * pseries-2.12
5071  */
5072 static void spapr_machine_2_12_class_options(MachineClass *mc)
5073 {
5074     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5075     static GlobalProperty compat[] = {
5076         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5077         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5078     };
5079 
5080     spapr_machine_3_0_class_options(mc);
5081     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5082     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5083 
5084     /* We depend on kvm_enabled() to choose a default value for the
5085      * hpt-max-page-size capability. Of course we can't do it here
5086      * because this is too early and the HW accelerator isn't initialized
5087      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5088      */
5089     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5090 }
5091 
5092 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5093 
5094 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5095 {
5096     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5097 
5098     spapr_machine_2_12_class_options(mc);
5099     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5100     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5101     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5102 }
5103 
5104 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5105 
5106 /*
5107  * pseries-2.11
5108  */
5109 
5110 static void spapr_machine_2_11_class_options(MachineClass *mc)
5111 {
5112     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5113 
5114     spapr_machine_2_12_class_options(mc);
5115     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5116     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5117     mc->deprecation_reason = "old and not maintained - use a 2.12+ version";
5118 }
5119 
5120 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5121 
5122 /*
5123  * pseries-2.10
5124  */
5125 
5126 static void spapr_machine_2_10_class_options(MachineClass *mc)
5127 {
5128     spapr_machine_2_11_class_options(mc);
5129     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5130 }
5131 
5132 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5133 
5134 /*
5135  * pseries-2.9
5136  */
5137 
5138 static void spapr_machine_2_9_class_options(MachineClass *mc)
5139 {
5140     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5141     static GlobalProperty compat[] = {
5142         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5143     };
5144 
5145     spapr_machine_2_10_class_options(mc);
5146     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5147     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5148     smc->pre_2_10_has_unused_icps = true;
5149     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5150 }
5151 
5152 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5153 
5154 /*
5155  * pseries-2.8
5156  */
5157 
5158 static void spapr_machine_2_8_class_options(MachineClass *mc)
5159 {
5160     static GlobalProperty compat[] = {
5161         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5162     };
5163 
5164     spapr_machine_2_9_class_options(mc);
5165     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5166     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5167     mc->numa_mem_align_shift = 23;
5168 }
5169 
5170 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5171 
5172 /*
5173  * pseries-2.7
5174  */
5175 
5176 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5177                               uint64_t *buid, hwaddr *pio,
5178                               hwaddr *mmio32, hwaddr *mmio64,
5179                               unsigned n_dma, uint32_t *liobns, Error **errp)
5180 {
5181     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5182     const uint64_t base_buid = 0x800000020000000ULL;
5183     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5184     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5185     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5186     const uint32_t max_index = 255;
5187     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5188 
5189     uint64_t ram_top = MACHINE(spapr)->ram_size;
5190     hwaddr phb0_base, phb_base;
5191     int i;
5192 
5193     /* Do we have device memory? */
5194     if (MACHINE(spapr)->device_memory) {
5195         /* Can't just use maxram_size, because there may be an
5196          * alignment gap between normal and device memory regions
5197          */
5198         ram_top = MACHINE(spapr)->device_memory->base +
5199             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5200     }
5201 
5202     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5203 
5204     if (index > max_index) {
5205         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5206                    max_index);
5207         return false;
5208     }
5209 
5210     *buid = base_buid + index;
5211     for (i = 0; i < n_dma; ++i) {
5212         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5213     }
5214 
5215     phb_base = phb0_base + index * phb_spacing;
5216     *pio = phb_base + pio_offset;
5217     *mmio32 = phb_base + mmio_offset;
5218     /*
5219      * We don't set the 64-bit MMIO window, relying on the PHB's
5220      * fallback behaviour of automatically splitting a large "32-bit"
5221      * window into contiguous 32-bit and 64-bit windows
5222      */
5223 
5224     return true;
5225 }
5226 
5227 static void spapr_machine_2_7_class_options(MachineClass *mc)
5228 {
5229     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5230     static GlobalProperty compat[] = {
5231         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5232         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5233         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5234         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5235     };
5236 
5237     spapr_machine_2_8_class_options(mc);
5238     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5239     mc->default_machine_opts = "modern-hotplug-events=off";
5240     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5241     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5242     smc->phb_placement = phb_placement_2_7;
5243 }
5244 
5245 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5246 
5247 /*
5248  * pseries-2.6
5249  */
5250 
5251 static void spapr_machine_2_6_class_options(MachineClass *mc)
5252 {
5253     static GlobalProperty compat[] = {
5254         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5255     };
5256 
5257     spapr_machine_2_7_class_options(mc);
5258     mc->has_hotpluggable_cpus = false;
5259     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5260     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5261 }
5262 
5263 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5264 
5265 /*
5266  * pseries-2.5
5267  */
5268 
5269 static void spapr_machine_2_5_class_options(MachineClass *mc)
5270 {
5271     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5272     static GlobalProperty compat[] = {
5273         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5274     };
5275 
5276     spapr_machine_2_6_class_options(mc);
5277     smc->use_ohci_by_default = true;
5278     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5279     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5280 }
5281 
5282 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5283 
5284 /*
5285  * pseries-2.4
5286  */
5287 
5288 static void spapr_machine_2_4_class_options(MachineClass *mc)
5289 {
5290     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5291 
5292     spapr_machine_2_5_class_options(mc);
5293     smc->dr_lmb_enabled = false;
5294     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5295 }
5296 
5297 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5298 
5299 /*
5300  * pseries-2.3
5301  */
5302 
5303 static void spapr_machine_2_3_class_options(MachineClass *mc)
5304 {
5305     static GlobalProperty compat[] = {
5306         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5307     };
5308     spapr_machine_2_4_class_options(mc);
5309     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5310     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5311 }
5312 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5313 
5314 /*
5315  * pseries-2.2
5316  */
5317 
5318 static void spapr_machine_2_2_class_options(MachineClass *mc)
5319 {
5320     static GlobalProperty compat[] = {
5321         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5322     };
5323 
5324     spapr_machine_2_3_class_options(mc);
5325     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5326     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5327     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5328 }
5329 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5330 
5331 /*
5332  * pseries-2.1
5333  */
5334 
5335 static void spapr_machine_2_1_class_options(MachineClass *mc)
5336 {
5337     spapr_machine_2_2_class_options(mc);
5338     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5339 }
5340 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5341 
5342 static void spapr_machine_register_types(void)
5343 {
5344     type_register_static(&spapr_machine_info);
5345 }
5346 
5347 type_init(spapr_machine_register_types)
5348