xref: /openbmc/qemu/hw/ppc/spapr.c (revision ba6bb2ec)
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/qtest.h"
39 #include "sysemu/reset.h"
40 #include "sysemu/runstate.h"
41 #include "qemu/log.h"
42 #include "hw/fw-path-provider.h"
43 #include "elf.h"
44 #include "net/net.h"
45 #include "sysemu/device_tree.h"
46 #include "sysemu/cpus.h"
47 #include "sysemu/hw_accel.h"
48 #include "kvm_ppc.h"
49 #include "migration/misc.h"
50 #include "migration/qemu-file-types.h"
51 #include "migration/global_state.h"
52 #include "migration/register.h"
53 #include "migration/blocker.h"
54 #include "mmu-hash64.h"
55 #include "mmu-book3s-v3.h"
56 #include "cpu-models.h"
57 #include "hw/core/cpu.h"
58 
59 #include "hw/ppc/ppc.h"
60 #include "hw/loader.h"
61 
62 #include "hw/ppc/fdt.h"
63 #include "hw/ppc/spapr.h"
64 #include "hw/ppc/spapr_nested.h"
65 #include "hw/ppc/spapr_vio.h"
66 #include "hw/ppc/vof.h"
67 #include "hw/qdev-properties.h"
68 #include "hw/pci-host/spapr.h"
69 #include "hw/pci/msi.h"
70 
71 #include "hw/pci/pci.h"
72 #include "hw/scsi/scsi.h"
73 #include "hw/virtio/virtio-scsi.h"
74 #include "hw/virtio/vhost-scsi-common.h"
75 
76 #include "exec/ram_addr.h"
77 #include "hw/usb.h"
78 #include "qemu/config-file.h"
79 #include "qemu/error-report.h"
80 #include "trace.h"
81 #include "hw/nmi.h"
82 #include "hw/intc/intc.h"
83 
84 #include "hw/ppc/spapr_cpu_core.h"
85 #include "hw/mem/memory-device.h"
86 #include "hw/ppc/spapr_tpm_proxy.h"
87 #include "hw/ppc/spapr_nvdimm.h"
88 #include "hw/ppc/spapr_numa.h"
89 #include "hw/ppc/pef.h"
90 
91 #include "monitor/monitor.h"
92 
93 #include <libfdt.h>
94 
95 /* SLOF memory layout:
96  *
97  * SLOF raw image loaded at 0, copies its romfs right below the flat
98  * device-tree, then position SLOF itself 31M below that
99  *
100  * So we set FW_OVERHEAD to 40MB which should account for all of that
101  * and more
102  *
103  * We load our kernel at 4M, leaving space for SLOF initial image
104  */
105 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
106 #define FW_MAX_SIZE             0x400000
107 #define FW_FILE_NAME            "slof.bin"
108 #define FW_FILE_NAME_VOF        "vof.bin"
109 #define FW_OVERHEAD             0x2800000
110 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
111 
112 #define MIN_RMA_SLOF            (128 * MiB)
113 
114 #define PHANDLE_INTC            0x00001111
115 
116 /* These two functions implement the VCPU id numbering: one to compute them
117  * all and one to identify thread 0 of a VCORE. Any change to the first one
118  * is likely to have an impact on the second one, so let's keep them close.
119  */
120 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
121 {
122     MachineState *ms = MACHINE(spapr);
123     unsigned int smp_threads = ms->smp.threads;
124 
125     assert(spapr->vsmt);
126     return
127         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
128 }
129 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
130                                       PowerPCCPU *cpu)
131 {
132     assert(spapr->vsmt);
133     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
134 }
135 
136 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
137 {
138     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
139      * and newer QEMUs don't even have them. In both cases, we don't want
140      * to send anything on the wire.
141      */
142     return false;
143 }
144 
145 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
146     /*
147      * Hack ahead.  We can't have two devices with the same name and
148      * instance id.  So I rename this to pass make check.
149      * Real help from people who knows the hardware is needed.
150      */
151     .name = "icp/server",
152     .version_id = 1,
153     .minimum_version_id = 1,
154     .needed = pre_2_10_vmstate_dummy_icp_needed,
155     .fields = (const VMStateField[]) {
156         VMSTATE_UNUSED(4), /* uint32_t xirr */
157         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
158         VMSTATE_UNUSED(1), /* uint8_t mfrr */
159         VMSTATE_END_OF_LIST()
160     },
161 };
162 
163 /*
164  * See comment in hw/intc/xics.c:icp_realize()
165  *
166  * You have to remove vmstate_replace_hack_for_ppc() when you remove
167  * the machine types that need the following function.
168  */
169 static void pre_2_10_vmstate_register_dummy_icp(int i)
170 {
171     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
172                      (void *)(uintptr_t) i);
173 }
174 
175 /*
176  * See comment in hw/intc/xics.c:icp_realize()
177  *
178  * You have to remove vmstate_replace_hack_for_ppc() when you remove
179  * the machine types that need the following function.
180  */
181 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
182 {
183     /*
184      * This used to be:
185      *
186      *    vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
187      *                      (void *)(uintptr_t) i);
188      */
189 }
190 
191 int spapr_max_server_number(SpaprMachineState *spapr)
192 {
193     MachineState *ms = MACHINE(spapr);
194 
195     assert(spapr->vsmt);
196     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
197 }
198 
199 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
200                                   int smt_threads)
201 {
202     int i, ret = 0;
203     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
204     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
205     int index = spapr_get_vcpu_id(cpu);
206 
207     if (cpu->compat_pvr) {
208         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
209         if (ret < 0) {
210             return ret;
211         }
212     }
213 
214     /* Build interrupt servers and gservers properties */
215     for (i = 0; i < smt_threads; i++) {
216         servers_prop[i] = cpu_to_be32(index + i);
217         /* Hack, direct the group queues back to cpu 0 */
218         gservers_prop[i*2] = cpu_to_be32(index + i);
219         gservers_prop[i*2 + 1] = 0;
220     }
221     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
222                       servers_prop, sizeof(*servers_prop) * smt_threads);
223     if (ret < 0) {
224         return ret;
225     }
226     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
227                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
228 
229     return ret;
230 }
231 
232 static void spapr_dt_pa_features(SpaprMachineState *spapr,
233                                  PowerPCCPU *cpu,
234                                  void *fdt, int offset)
235 {
236     /*
237      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
238      * but not MTTCG, so disable it. To advertise it, a cap would have
239      * to be added, or support implemented for MTTCG.
240      *
241      * Copy/paste is not supported by TCG, so it is not advertised. KVM
242      * can execute them but it has no accelerator drivers which are usable,
243      * so there isn't much need for it anyway.
244      */
245 
246     /* These should be kept in sync with pnv */
247     uint8_t pa_features_206[] = { 6, 0,
248         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
249     uint8_t pa_features_207[] = { 24, 0,
250         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
251         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
252         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
254     uint8_t pa_features_300[] = { 66, 0,
255         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
256         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
257         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
258         /* 6: DS207 */
259         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
260         /* 16: Vector */
261         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
262         /* 18: Vec. Scalar, 20: Vec. XOR */
263         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
264         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
265         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
266         /* 32: LE atomic, 34: EBB + ext EBB */
267         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
268         /* 40: Radix MMU */
269         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
270         /* 42: PM, 44: PC RA, 46: SC vec'd */
271         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
272         /* 48: SIMD, 50: QP BFP, 52: String */
273         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
274         /* 54: DecFP, 56: DecI, 58: SHA */
275         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
276         /* 60: NM atomic, 62: RNG */
277         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
278     };
279     /* 3.1 removes SAO, HTM support */
280     uint8_t pa_features_31[] = { 74, 0,
281         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
282         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
283         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
284         /* 6: DS207 */
285         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
286         /* 16: Vector */
287         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
288         /* 18: Vec. Scalar, 20: Vec. XOR */
289         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
290         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
291         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
292         /* 32: LE atomic, 34: EBB + ext EBB */
293         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
294         /* 40: Radix MMU */
295         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
296         /* 42: PM, 44: PC RA, 46: SC vec'd */
297         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
298         /* 48: SIMD, 50: QP BFP, 52: String */
299         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
300         /* 54: DecFP, 56: DecI, 58: SHA */
301         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
302         /* 60: NM atomic, 62: RNG */
303         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
304         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
305         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
306         /* 72: [P]HASHST/[P]HASHCHK */
307         0x80, 0x00,                         /* 72 - 73 */
308     };
309     uint8_t *pa_features = NULL;
310     size_t pa_size;
311 
312     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
313         pa_features = pa_features_206;
314         pa_size = sizeof(pa_features_206);
315     }
316     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
317         pa_features = pa_features_207;
318         pa_size = sizeof(pa_features_207);
319     }
320     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
321         pa_features = pa_features_300;
322         pa_size = sizeof(pa_features_300);
323     }
324     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
325         pa_features = pa_features_31;
326         pa_size = sizeof(pa_features_31);
327     }
328     if (!pa_features) {
329         return;
330     }
331 
332     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
333         /*
334          * Note: we keep CI large pages off by default because a 64K capable
335          * guest provisioned with large pages might otherwise try to map a qemu
336          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
337          * even if that qemu runs on a 4k host.
338          * We dd this bit back here if we are confident this is not an issue
339          */
340         pa_features[3] |= 0x20;
341     }
342     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
343         pa_features[24] |= 0x80;    /* Transactional memory support */
344     }
345     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
346         /* Workaround for broken kernels that attempt (guest) radix
347          * mode when they can't handle it, if they see the radix bit set
348          * in pa-features. So hide it from them. */
349         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
350     }
351 
352     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
353 }
354 
355 static hwaddr spapr_node0_size(MachineState *machine)
356 {
357     if (machine->numa_state->num_nodes) {
358         int i;
359         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
360             if (machine->numa_state->nodes[i].node_mem) {
361                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
362                            machine->ram_size);
363             }
364         }
365     }
366     return machine->ram_size;
367 }
368 
369 static void add_str(GString *s, const gchar *s1)
370 {
371     g_string_append_len(s, s1, strlen(s1) + 1);
372 }
373 
374 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
375                                 hwaddr start, hwaddr size)
376 {
377     char mem_name[32];
378     uint64_t mem_reg_property[2];
379     int off;
380 
381     mem_reg_property[0] = cpu_to_be64(start);
382     mem_reg_property[1] = cpu_to_be64(size);
383 
384     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
385     off = fdt_add_subnode(fdt, 0, mem_name);
386     _FDT(off);
387     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
388     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
389                       sizeof(mem_reg_property))));
390     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
391     return off;
392 }
393 
394 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
395 {
396     MemoryDeviceInfoList *info;
397 
398     for (info = list; info; info = info->next) {
399         MemoryDeviceInfo *value = info->value;
400 
401         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
402             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
403 
404             if (addr >= pcdimm_info->addr &&
405                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
406                 return pcdimm_info->node;
407             }
408         }
409     }
410 
411     return -1;
412 }
413 
414 struct sPAPRDrconfCellV2 {
415      uint32_t seq_lmbs;
416      uint64_t base_addr;
417      uint32_t drc_index;
418      uint32_t aa_index;
419      uint32_t flags;
420 } QEMU_PACKED;
421 
422 typedef struct DrconfCellQueue {
423     struct sPAPRDrconfCellV2 cell;
424     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
425 } DrconfCellQueue;
426 
427 static DrconfCellQueue *
428 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
429                       uint32_t drc_index, uint32_t aa_index,
430                       uint32_t flags)
431 {
432     DrconfCellQueue *elem;
433 
434     elem = g_malloc0(sizeof(*elem));
435     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
436     elem->cell.base_addr = cpu_to_be64(base_addr);
437     elem->cell.drc_index = cpu_to_be32(drc_index);
438     elem->cell.aa_index = cpu_to_be32(aa_index);
439     elem->cell.flags = cpu_to_be32(flags);
440 
441     return elem;
442 }
443 
444 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
445                                       int offset, MemoryDeviceInfoList *dimms)
446 {
447     MachineState *machine = MACHINE(spapr);
448     uint8_t *int_buf, *cur_index;
449     int ret;
450     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
451     uint64_t addr, cur_addr, size;
452     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
453     uint64_t mem_end = machine->device_memory->base +
454                        memory_region_size(&machine->device_memory->mr);
455     uint32_t node, buf_len, nr_entries = 0;
456     SpaprDrc *drc;
457     DrconfCellQueue *elem, *next;
458     MemoryDeviceInfoList *info;
459     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
460         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
461 
462     /* Entry to cover RAM and the gap area */
463     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
464                                  SPAPR_LMB_FLAGS_RESERVED |
465                                  SPAPR_LMB_FLAGS_DRC_INVALID);
466     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
467     nr_entries++;
468 
469     cur_addr = machine->device_memory->base;
470     for (info = dimms; info; info = info->next) {
471         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
472 
473         addr = di->addr;
474         size = di->size;
475         node = di->node;
476 
477         /*
478          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
479          * area is marked hotpluggable in the next iteration for the bigger
480          * chunk including the NVDIMM occupied area.
481          */
482         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
483             continue;
484 
485         /* Entry for hot-pluggable area */
486         if (cur_addr < addr) {
487             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
488             g_assert(drc);
489             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
490                                          cur_addr, spapr_drc_index(drc), -1, 0);
491             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
492             nr_entries++;
493         }
494 
495         /* Entry for DIMM */
496         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
497         g_assert(drc);
498         elem = spapr_get_drconf_cell(size / lmb_size, addr,
499                                      spapr_drc_index(drc), node,
500                                      (SPAPR_LMB_FLAGS_ASSIGNED |
501                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
502         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
503         nr_entries++;
504         cur_addr = addr + size;
505     }
506 
507     /* Entry for remaining hotpluggable area */
508     if (cur_addr < mem_end) {
509         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
510         g_assert(drc);
511         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
512                                      cur_addr, spapr_drc_index(drc), -1, 0);
513         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
514         nr_entries++;
515     }
516 
517     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
518     int_buf = cur_index = g_malloc0(buf_len);
519     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
520     cur_index += sizeof(nr_entries);
521 
522     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
523         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
524         cur_index += sizeof(elem->cell);
525         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
526         g_free(elem);
527     }
528 
529     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
530     g_free(int_buf);
531     if (ret < 0) {
532         return -1;
533     }
534     return 0;
535 }
536 
537 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
538                                    int offset, MemoryDeviceInfoList *dimms)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int i, ret;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
544     uint32_t nr_lmbs = (machine->device_memory->base +
545                        memory_region_size(&machine->device_memory->mr)) /
546                        lmb_size;
547     uint32_t *int_buf, *cur_index, buf_len;
548 
549     /*
550      * Allocate enough buffer size to fit in ibm,dynamic-memory
551      */
552     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
553     cur_index = int_buf = g_malloc0(buf_len);
554     int_buf[0] = cpu_to_be32(nr_lmbs);
555     cur_index++;
556     for (i = 0; i < nr_lmbs; i++) {
557         uint64_t addr = i * lmb_size;
558         uint32_t *dynamic_memory = cur_index;
559 
560         if (i >= device_lmb_start) {
561             SpaprDrc *drc;
562 
563             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
564             g_assert(drc);
565 
566             dynamic_memory[0] = cpu_to_be32(addr >> 32);
567             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
568             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
569             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
570             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
571             if (memory_region_present(get_system_memory(), addr)) {
572                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
573             } else {
574                 dynamic_memory[5] = cpu_to_be32(0);
575             }
576         } else {
577             /*
578              * LMB information for RMA, boot time RAM and gap b/n RAM and
579              * device memory region -- all these are marked as reserved
580              * and as having no valid DRC.
581              */
582             dynamic_memory[0] = cpu_to_be32(addr >> 32);
583             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
584             dynamic_memory[2] = cpu_to_be32(0);
585             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
586             dynamic_memory[4] = cpu_to_be32(-1);
587             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
588                                             SPAPR_LMB_FLAGS_DRC_INVALID);
589         }
590 
591         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
592     }
593     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
594     g_free(int_buf);
595     if (ret < 0) {
596         return -1;
597     }
598     return 0;
599 }
600 
601 /*
602  * Adds ibm,dynamic-reconfiguration-memory node.
603  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
604  * of this device tree node.
605  */
606 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
607                                                    void *fdt)
608 {
609     MachineState *machine = MACHINE(spapr);
610     int ret, offset;
611     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
612     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
613                                 cpu_to_be32(lmb_size & 0xffffffff)};
614     MemoryDeviceInfoList *dimms = NULL;
615 
616     /* Don't create the node if there is no device memory. */
617     if (!machine->device_memory) {
618         return 0;
619     }
620 
621     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
622 
623     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
624                     sizeof(prop_lmb_size));
625     if (ret < 0) {
626         return ret;
627     }
628 
629     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
630     if (ret < 0) {
631         return ret;
632     }
633 
634     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
635     if (ret < 0) {
636         return ret;
637     }
638 
639     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
640     dimms = qmp_memory_device_list();
641     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
642         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
643     } else {
644         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
645     }
646     qapi_free_MemoryDeviceInfoList(dimms);
647 
648     if (ret < 0) {
649         return ret;
650     }
651 
652     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
653 
654     return ret;
655 }
656 
657 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
658 {
659     MachineState *machine = MACHINE(spapr);
660     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
661     hwaddr mem_start, node_size;
662     int i, nb_nodes = machine->numa_state->num_nodes;
663     NodeInfo *nodes = machine->numa_state->nodes;
664 
665     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
666         if (!nodes[i].node_mem) {
667             continue;
668         }
669         if (mem_start >= machine->ram_size) {
670             node_size = 0;
671         } else {
672             node_size = nodes[i].node_mem;
673             if (node_size > machine->ram_size - mem_start) {
674                 node_size = machine->ram_size - mem_start;
675             }
676         }
677         if (!mem_start) {
678             /* spapr_machine_init() checks for rma_size <= node0_size
679              * already */
680             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
681             mem_start += spapr->rma_size;
682             node_size -= spapr->rma_size;
683         }
684         for ( ; node_size; ) {
685             hwaddr sizetmp = pow2floor(node_size);
686 
687             /* mem_start != 0 here */
688             if (ctzl(mem_start) < ctzl(sizetmp)) {
689                 sizetmp = 1ULL << ctzl(mem_start);
690             }
691 
692             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
693             node_size -= sizetmp;
694             mem_start += sizetmp;
695         }
696     }
697 
698     /* Generate ibm,dynamic-reconfiguration-memory node if required */
699     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
700         int ret;
701 
702         g_assert(smc->dr_lmb_enabled);
703         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
704         if (ret) {
705             return ret;
706         }
707     }
708 
709     return 0;
710 }
711 
712 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
713                          SpaprMachineState *spapr)
714 {
715     MachineState *ms = MACHINE(spapr);
716     PowerPCCPU *cpu = POWERPC_CPU(cs);
717     CPUPPCState *env = &cpu->env;
718     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
719     int index = spapr_get_vcpu_id(cpu);
720     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
721                        0xffffffff, 0xffffffff};
722     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
723         : SPAPR_TIMEBASE_FREQ;
724     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
725     uint32_t page_sizes_prop[64];
726     size_t page_sizes_prop_size;
727     unsigned int smp_threads = ms->smp.threads;
728     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
729     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
730     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
731     SpaprDrc *drc;
732     int drc_index;
733     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
734     int i;
735 
736     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
737     if (drc) {
738         drc_index = spapr_drc_index(drc);
739         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
740     }
741 
742     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
743     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
744 
745     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
746     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
747                            env->dcache_line_size)));
748     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
749                            env->dcache_line_size)));
750     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
751                            env->icache_line_size)));
752     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
753                            env->icache_line_size)));
754 
755     if (pcc->l1_dcache_size) {
756         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
757                                pcc->l1_dcache_size)));
758     } else {
759         warn_report("Unknown L1 dcache size for cpu");
760     }
761     if (pcc->l1_icache_size) {
762         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
763                                pcc->l1_icache_size)));
764     } else {
765         warn_report("Unknown L1 icache size for cpu");
766     }
767 
768     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
769     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
770     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
771     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
772     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
773     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
774 
775     if (ppc_has_spr(cpu, SPR_PURR)) {
776         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
777     }
778     if (ppc_has_spr(cpu, SPR_PURR)) {
779         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
780     }
781 
782     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
783         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
784                           segs, sizeof(segs))));
785     }
786 
787     /* Advertise VSX (vector extensions) if available
788      *   1               == VMX / Altivec available
789      *   2               == VSX available
790      *
791      * Only CPUs for which we create core types in spapr_cpu_core.c
792      * are possible, and all of those have VMX */
793     if (env->insns_flags & PPC_ALTIVEC) {
794         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
795             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
796         } else {
797             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
798         }
799     }
800 
801     /* Advertise DFP (Decimal Floating Point) if available
802      *   0 / no property == no DFP
803      *   1               == DFP available */
804     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
805         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
806     }
807 
808     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
809                                                       sizeof(page_sizes_prop));
810     if (page_sizes_prop_size) {
811         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
812                           page_sizes_prop, page_sizes_prop_size)));
813     }
814 
815     spapr_dt_pa_features(spapr, cpu, fdt, offset);
816 
817     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
818                            cs->cpu_index / vcpus_per_socket)));
819 
820     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
821                       pft_size_prop, sizeof(pft_size_prop))));
822 
823     if (ms->numa_state->num_nodes > 1) {
824         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
825     }
826 
827     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
828 
829     if (pcc->radix_page_info) {
830         for (i = 0; i < pcc->radix_page_info->count; i++) {
831             radix_AP_encodings[i] =
832                 cpu_to_be32(pcc->radix_page_info->entries[i]);
833         }
834         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
835                           radix_AP_encodings,
836                           pcc->radix_page_info->count *
837                           sizeof(radix_AP_encodings[0]))));
838     }
839 
840     /*
841      * We set this property to let the guest know that it can use the large
842      * decrementer and its width in bits.
843      */
844     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
845         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
846                               pcc->lrg_decr_bits)));
847 }
848 
849 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
850                              int cpus_offset)
851 {
852     PowerPCCPU *cpu = POWERPC_CPU(cs);
853     int index = spapr_get_vcpu_id(cpu);
854     DeviceClass *dc = DEVICE_GET_CLASS(cs);
855     g_autofree char *nodename = NULL;
856     int offset;
857 
858     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
859         return;
860     }
861 
862     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
863     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
864     _FDT(offset);
865     spapr_dt_cpu(cs, fdt, offset, spapr);
866 }
867 
868 
869 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
870 {
871     CPUState **rev;
872     CPUState *cs;
873     int n_cpus;
874     int cpus_offset;
875     int i;
876 
877     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
878     _FDT(cpus_offset);
879     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
880     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
881 
882     /*
883      * We walk the CPUs in reverse order to ensure that CPU DT nodes
884      * created by fdt_add_subnode() end up in the right order in FDT
885      * for the guest kernel the enumerate the CPUs correctly.
886      *
887      * The CPU list cannot be traversed in reverse order, so we need
888      * to do extra work.
889      */
890     n_cpus = 0;
891     rev = NULL;
892     CPU_FOREACH(cs) {
893         rev = g_renew(CPUState *, rev, n_cpus + 1);
894         rev[n_cpus++] = cs;
895     }
896 
897     for (i = n_cpus - 1; i >= 0; i--) {
898         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
899     }
900 
901     g_free(rev);
902 }
903 
904 static int spapr_dt_rng(void *fdt)
905 {
906     int node;
907     int ret;
908 
909     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
910     if (node <= 0) {
911         return -1;
912     }
913     ret = fdt_setprop_string(fdt, node, "device_type",
914                              "ibm,platform-facilities");
915     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
916     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
917 
918     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
919     if (node <= 0) {
920         return -1;
921     }
922     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
923 
924     return ret ? -1 : 0;
925 }
926 
927 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
928 {
929     MachineState *ms = MACHINE(spapr);
930     int rtas;
931     GString *hypertas = g_string_sized_new(256);
932     GString *qemu_hypertas = g_string_sized_new(256);
933     uint32_t lrdr_capacity[] = {
934         0,
935         0,
936         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
937         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
938         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
939     };
940 
941     /* Do we have device memory? */
942     if (MACHINE(spapr)->device_memory) {
943         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
944             memory_region_size(&MACHINE(spapr)->device_memory->mr);
945 
946         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
947         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
948     }
949 
950     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
951 
952     /* hypertas */
953     add_str(hypertas, "hcall-pft");
954     add_str(hypertas, "hcall-term");
955     add_str(hypertas, "hcall-dabr");
956     add_str(hypertas, "hcall-interrupt");
957     add_str(hypertas, "hcall-tce");
958     add_str(hypertas, "hcall-vio");
959     add_str(hypertas, "hcall-splpar");
960     add_str(hypertas, "hcall-join");
961     add_str(hypertas, "hcall-bulk");
962     add_str(hypertas, "hcall-set-mode");
963     add_str(hypertas, "hcall-sprg0");
964     add_str(hypertas, "hcall-copy");
965     add_str(hypertas, "hcall-debug");
966     add_str(hypertas, "hcall-vphn");
967     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
968         add_str(hypertas, "hcall-rpt-invalidate");
969     }
970 
971     add_str(qemu_hypertas, "hcall-memop1");
972 
973     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
974         add_str(hypertas, "hcall-multi-tce");
975     }
976 
977     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
978         add_str(hypertas, "hcall-hpt-resize");
979     }
980 
981     add_str(hypertas, "hcall-watchdog");
982 
983     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
984                      hypertas->str, hypertas->len));
985     g_string_free(hypertas, TRUE);
986     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
987                      qemu_hypertas->str, qemu_hypertas->len));
988     g_string_free(qemu_hypertas, TRUE);
989 
990     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
991 
992     /*
993      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
994      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
995      *
996      * The system reset requirements are driven by existing Linux and PowerVM
997      * implementation which (contrary to PAPR) saves r3 in the error log
998      * structure like machine check, so Linux expects to find the saved r3
999      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
1000      * does not look at the error value).
1001      *
1002      * System reset interrupts are not subject to interlock like machine
1003      * check, so this memory area could be corrupted if the sreset is
1004      * interrupted by a machine check (or vice versa) if it was shared. To
1005      * prevent this, system reset uses per-CPU areas for the sreset save
1006      * area. A system reset that interrupts a system reset handler could
1007      * still overwrite this area, but Linux doesn't try to recover in that
1008      * case anyway.
1009      *
1010      * The extra 8 bytes is required because Linux's FWNMI error log check
1011      * is off-by-one.
1012      *
1013      * RTAS_MIN_SIZE is required for the RTAS blob itself.
1014      */
1015     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
1016                           RTAS_ERROR_LOG_MAX +
1017                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
1018                           sizeof(uint64_t)));
1019     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1020                           RTAS_ERROR_LOG_MAX));
1021     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1022                           RTAS_EVENT_SCAN_RATE));
1023 
1024     g_assert(msi_nonbroken);
1025     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1026 
1027     /*
1028      * According to PAPR, rtas ibm,os-term does not guarantee a return
1029      * back to the guest cpu.
1030      *
1031      * While an additional ibm,extended-os-term property indicates
1032      * that rtas call return will always occur. Set this property.
1033      */
1034     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1035 
1036     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1037                      lrdr_capacity, sizeof(lrdr_capacity)));
1038 
1039     spapr_dt_rtas_tokens(fdt, rtas);
1040 }
1041 
1042 /*
1043  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1044  * and the XIVE features that the guest may request and thus the valid
1045  * values for bytes 23..26 of option vector 5:
1046  */
1047 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1048                                           int chosen)
1049 {
1050     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1051 
1052     char val[2 * 4] = {
1053         23, 0x00, /* XICS / XIVE mode */
1054         24, 0x00, /* Hash/Radix, filled in below. */
1055         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1056         26, 0x40, /* Radix options: GTSE == yes. */
1057     };
1058 
1059     if (spapr->irq->xics && spapr->irq->xive) {
1060         val[1] = SPAPR_OV5_XIVE_BOTH;
1061     } else if (spapr->irq->xive) {
1062         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1063     } else {
1064         assert(spapr->irq->xics);
1065         val[1] = SPAPR_OV5_XIVE_LEGACY;
1066     }
1067 
1068     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1069                           first_ppc_cpu->compat_pvr)) {
1070         /*
1071          * If we're in a pre POWER9 compat mode then the guest should
1072          * do hash and use the legacy interrupt mode
1073          */
1074         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1075         val[3] = 0x00; /* Hash */
1076         spapr_check_mmu_mode(false);
1077     } else if (kvm_enabled()) {
1078         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1079             val[3] = 0x80; /* OV5_MMU_BOTH */
1080         } else if (kvmppc_has_cap_mmu_radix()) {
1081             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1082         } else {
1083             val[3] = 0x00; /* Hash */
1084         }
1085     } else {
1086         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1087         val[3] = 0xC0;
1088     }
1089     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1090                      val, sizeof(val)));
1091 }
1092 
1093 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1094 {
1095     MachineState *machine = MACHINE(spapr);
1096     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1097     int chosen;
1098 
1099     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1100 
1101     if (reset) {
1102         const char *boot_device = spapr->boot_device;
1103         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1104         size_t cb = 0;
1105         g_autofree char *bootlist = get_boot_devices_list(&cb);
1106 
1107         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1108             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1109                                     machine->kernel_cmdline));
1110         }
1111 
1112         if (spapr->initrd_size) {
1113             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1114                                   spapr->initrd_base));
1115             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1116                                   spapr->initrd_base + spapr->initrd_size));
1117         }
1118 
1119         if (spapr->kernel_size) {
1120             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1121                                   cpu_to_be64(spapr->kernel_size) };
1122 
1123             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1124                          &kprop, sizeof(kprop)));
1125             if (spapr->kernel_le) {
1126                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1127             }
1128         }
1129         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1130             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1131         }
1132         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1133         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1134         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1135 
1136         if (cb && bootlist) {
1137             int i;
1138 
1139             for (i = 0; i < cb; i++) {
1140                 if (bootlist[i] == '\n') {
1141                     bootlist[i] = ' ';
1142                 }
1143             }
1144             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1145         }
1146 
1147         if (boot_device && strlen(boot_device)) {
1148             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1149         }
1150 
1151         if (spapr->want_stdout_path && stdout_path) {
1152             /*
1153              * "linux,stdout-path" and "stdout" properties are
1154              * deprecated by linux kernel. New platforms should only
1155              * use the "stdout-path" property. Set the new property
1156              * and continue using older property to remain compatible
1157              * with the existing firmware.
1158              */
1159             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1160             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1161         }
1162 
1163         /*
1164          * We can deal with BAR reallocation just fine, advertise it
1165          * to the guest
1166          */
1167         if (smc->linux_pci_probe) {
1168             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1169         }
1170 
1171         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1172     }
1173 
1174     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1175 
1176     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1177 }
1178 
1179 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1180 {
1181     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1182      * KVM to work under pHyp with some guest co-operation */
1183     int hypervisor;
1184     uint8_t hypercall[16];
1185 
1186     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1187     /* indicate KVM hypercall interface */
1188     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1189     if (kvmppc_has_cap_fixup_hcalls()) {
1190         /*
1191          * Older KVM versions with older guest kernels were broken
1192          * with the magic page, don't allow the guest to map it.
1193          */
1194         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1195                                   sizeof(hypercall))) {
1196             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1197                              hypercall, sizeof(hypercall)));
1198         }
1199     }
1200 }
1201 
1202 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1203 {
1204     MachineState *machine = MACHINE(spapr);
1205     MachineClass *mc = MACHINE_GET_CLASS(machine);
1206     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1207     uint32_t root_drc_type_mask = 0;
1208     int ret;
1209     void *fdt;
1210     SpaprPhbState *phb;
1211     char *buf;
1212 
1213     fdt = g_malloc0(space);
1214     _FDT((fdt_create_empty_tree(fdt, space)));
1215 
1216     /* Root node */
1217     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1218     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1219     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1220 
1221     /* Guest UUID & Name*/
1222     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1223     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1224     if (qemu_uuid_set) {
1225         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1226     }
1227     g_free(buf);
1228 
1229     if (qemu_get_vm_name()) {
1230         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1231                                 qemu_get_vm_name()));
1232     }
1233 
1234     /* Host Model & Serial Number */
1235     if (spapr->host_model) {
1236         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1237     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1238         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1239         g_free(buf);
1240     }
1241 
1242     if (spapr->host_serial) {
1243         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1244     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1245         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1246         g_free(buf);
1247     }
1248 
1249     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1250     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1251 
1252     /* /interrupt controller */
1253     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1254 
1255     ret = spapr_dt_memory(spapr, fdt);
1256     if (ret < 0) {
1257         error_report("couldn't setup memory nodes in fdt");
1258         exit(1);
1259     }
1260 
1261     /* /vdevice */
1262     spapr_dt_vdevice(spapr->vio_bus, fdt);
1263 
1264     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1265         ret = spapr_dt_rng(fdt);
1266         if (ret < 0) {
1267             error_report("could not set up rng device in the fdt");
1268             exit(1);
1269         }
1270     }
1271 
1272     QLIST_FOREACH(phb, &spapr->phbs, list) {
1273         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1274         if (ret < 0) {
1275             error_report("couldn't setup PCI devices in fdt");
1276             exit(1);
1277         }
1278     }
1279 
1280     spapr_dt_cpus(fdt, spapr);
1281 
1282     /* ibm,drc-indexes and friends */
1283     if (smc->dr_lmb_enabled) {
1284         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1285     }
1286     if (smc->dr_phb_enabled) {
1287         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1288     }
1289     if (mc->nvdimm_supported) {
1290         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1291     }
1292     if (root_drc_type_mask) {
1293         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1294     }
1295 
1296     if (mc->has_hotpluggable_cpus) {
1297         int offset = fdt_path_offset(fdt, "/cpus");
1298         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1299         if (ret < 0) {
1300             error_report("Couldn't set up CPU DR device tree properties");
1301             exit(1);
1302         }
1303     }
1304 
1305     /* /event-sources */
1306     spapr_dt_events(spapr, fdt);
1307 
1308     /* /rtas */
1309     spapr_dt_rtas(spapr, fdt);
1310 
1311     /* /chosen */
1312     spapr_dt_chosen(spapr, fdt, reset);
1313 
1314     /* /hypervisor */
1315     if (kvm_enabled()) {
1316         spapr_dt_hypervisor(spapr, fdt);
1317     }
1318 
1319     /* Build memory reserve map */
1320     if (reset) {
1321         if (spapr->kernel_size) {
1322             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1323                                   spapr->kernel_size)));
1324         }
1325         if (spapr->initrd_size) {
1326             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1327                                   spapr->initrd_size)));
1328         }
1329     }
1330 
1331     /* NVDIMM devices */
1332     if (mc->nvdimm_supported) {
1333         spapr_dt_persistent_memory(spapr, fdt);
1334     }
1335 
1336     return fdt;
1337 }
1338 
1339 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1340 {
1341     SpaprMachineState *spapr = opaque;
1342 
1343     return (addr & 0x0fffffff) + spapr->kernel_addr;
1344 }
1345 
1346 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1347                                     PowerPCCPU *cpu)
1348 {
1349     CPUPPCState *env = &cpu->env;
1350 
1351     /* The TCG path should also be holding the BQL at this point */
1352     g_assert(bql_locked());
1353 
1354     g_assert(!vhyp_cpu_in_nested(cpu));
1355 
1356     if (FIELD_EX64(env->msr, MSR, PR)) {
1357         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1358         env->gpr[3] = H_PRIVILEGE;
1359     } else {
1360         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1361     }
1362 }
1363 
1364 struct LPCRSyncState {
1365     target_ulong value;
1366     target_ulong mask;
1367 };
1368 
1369 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1370 {
1371     struct LPCRSyncState *s = arg.host_ptr;
1372     PowerPCCPU *cpu = POWERPC_CPU(cs);
1373     CPUPPCState *env = &cpu->env;
1374     target_ulong lpcr;
1375 
1376     cpu_synchronize_state(cs);
1377     lpcr = env->spr[SPR_LPCR];
1378     lpcr &= ~s->mask;
1379     lpcr |= s->value;
1380     ppc_store_lpcr(cpu, lpcr);
1381 }
1382 
1383 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1384 {
1385     CPUState *cs;
1386     struct LPCRSyncState s = {
1387         .value = value,
1388         .mask = mask
1389     };
1390     CPU_FOREACH(cs) {
1391         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1392     }
1393 }
1394 
1395 /* May be used when the machine is not running */
1396 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1397 {
1398     CPUState *cs;
1399     CPU_FOREACH(cs) {
1400         PowerPCCPU *cpu = POWERPC_CPU(cs);
1401         CPUPPCState *env = &cpu->env;
1402         target_ulong lpcr;
1403 
1404         lpcr = env->spr[SPR_LPCR];
1405         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1406         ppc_store_lpcr(cpu, lpcr);
1407     }
1408 }
1409 
1410 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1411                            target_ulong lpid, ppc_v3_pate_t *entry)
1412 {
1413     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1414     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1415 
1416     if (!spapr_cpu->in_nested) {
1417         assert(lpid == 0);
1418 
1419         /* Copy PATE1:GR into PATE0:HR */
1420         entry->dw0 = spapr->patb_entry & PATE0_HR;
1421         entry->dw1 = spapr->patb_entry;
1422         return true;
1423     } else {
1424         if (spapr_nested_api(spapr) == NESTED_API_KVM_HV) {
1425             return spapr_get_pate_nested_hv(spapr, cpu, lpid, entry);
1426         } else if (spapr_nested_api(spapr) == NESTED_API_PAPR) {
1427             return spapr_get_pate_nested_papr(spapr, cpu, lpid, entry);
1428         } else {
1429             g_assert_not_reached();
1430         }
1431     }
1432 }
1433 
1434 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1435 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1436 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1437 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1438 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1439 
1440 /*
1441  * Get the fd to access the kernel htab, re-opening it if necessary
1442  */
1443 static int get_htab_fd(SpaprMachineState *spapr)
1444 {
1445     Error *local_err = NULL;
1446 
1447     if (spapr->htab_fd >= 0) {
1448         return spapr->htab_fd;
1449     }
1450 
1451     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1452     if (spapr->htab_fd < 0) {
1453         error_report_err(local_err);
1454     }
1455 
1456     return spapr->htab_fd;
1457 }
1458 
1459 void close_htab_fd(SpaprMachineState *spapr)
1460 {
1461     if (spapr->htab_fd >= 0) {
1462         close(spapr->htab_fd);
1463     }
1464     spapr->htab_fd = -1;
1465 }
1466 
1467 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1468 {
1469     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1470 
1471     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1472 }
1473 
1474 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1475 {
1476     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1477 
1478     assert(kvm_enabled());
1479 
1480     if (!spapr->htab) {
1481         return 0;
1482     }
1483 
1484     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1485 }
1486 
1487 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1488                                                 hwaddr ptex, int n)
1489 {
1490     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1491     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1492 
1493     if (!spapr->htab) {
1494         /*
1495          * HTAB is controlled by KVM. Fetch into temporary buffer
1496          */
1497         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1498         kvmppc_read_hptes(hptes, ptex, n);
1499         return hptes;
1500     }
1501 
1502     /*
1503      * HTAB is controlled by QEMU. Just point to the internally
1504      * accessible PTEG.
1505      */
1506     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1507 }
1508 
1509 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1510                               const ppc_hash_pte64_t *hptes,
1511                               hwaddr ptex, int n)
1512 {
1513     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1514 
1515     if (!spapr->htab) {
1516         g_free((void *)hptes);
1517     }
1518 
1519     /* Nothing to do for qemu managed HPT */
1520 }
1521 
1522 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1523                       uint64_t pte0, uint64_t pte1)
1524 {
1525     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1526     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1527 
1528     if (!spapr->htab) {
1529         kvmppc_write_hpte(ptex, pte0, pte1);
1530     } else {
1531         if (pte0 & HPTE64_V_VALID) {
1532             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1533             /*
1534              * When setting valid, we write PTE1 first. This ensures
1535              * proper synchronization with the reading code in
1536              * ppc_hash64_pteg_search()
1537              */
1538             smp_wmb();
1539             stq_p(spapr->htab + offset, pte0);
1540         } else {
1541             stq_p(spapr->htab + offset, pte0);
1542             /*
1543              * When clearing it we set PTE0 first. This ensures proper
1544              * synchronization with the reading code in
1545              * ppc_hash64_pteg_search()
1546              */
1547             smp_wmb();
1548             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1549         }
1550     }
1551 }
1552 
1553 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1554                              uint64_t pte1)
1555 {
1556     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1557     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1558 
1559     if (!spapr->htab) {
1560         /* There should always be a hash table when this is called */
1561         error_report("spapr_hpte_set_c called with no hash table !");
1562         return;
1563     }
1564 
1565     /* The HW performs a non-atomic byte update */
1566     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1567 }
1568 
1569 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1570                              uint64_t pte1)
1571 {
1572     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1573     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1574 
1575     if (!spapr->htab) {
1576         /* There should always be a hash table when this is called */
1577         error_report("spapr_hpte_set_r called with no hash table !");
1578         return;
1579     }
1580 
1581     /* The HW performs a non-atomic byte update */
1582     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1583 }
1584 
1585 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1586 {
1587     int shift;
1588 
1589     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1590      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1591      * that's much more than is needed for Linux guests */
1592     shift = ctz64(pow2ceil(ramsize)) - 7;
1593     shift = MAX(shift, 18); /* Minimum architected size */
1594     shift = MIN(shift, 46); /* Maximum architected size */
1595     return shift;
1596 }
1597 
1598 void spapr_free_hpt(SpaprMachineState *spapr)
1599 {
1600     qemu_vfree(spapr->htab);
1601     spapr->htab = NULL;
1602     spapr->htab_shift = 0;
1603     close_htab_fd(spapr);
1604 }
1605 
1606 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1607 {
1608     ERRP_GUARD();
1609     long rc;
1610 
1611     /* Clean up any HPT info from a previous boot */
1612     spapr_free_hpt(spapr);
1613 
1614     rc = kvmppc_reset_htab(shift);
1615 
1616     if (rc == -EOPNOTSUPP) {
1617         error_setg(errp, "HPT not supported in nested guests");
1618         return -EOPNOTSUPP;
1619     }
1620 
1621     if (rc < 0) {
1622         /* kernel-side HPT needed, but couldn't allocate one */
1623         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1624                          shift);
1625         error_append_hint(errp, "Try smaller maxmem?\n");
1626         return -errno;
1627     } else if (rc > 0) {
1628         /* kernel-side HPT allocated */
1629         if (rc != shift) {
1630             error_setg(errp,
1631                        "Requested order %d HPT, but kernel allocated order %ld",
1632                        shift, rc);
1633             error_append_hint(errp, "Try smaller maxmem?\n");
1634             return -ENOSPC;
1635         }
1636 
1637         spapr->htab_shift = shift;
1638         spapr->htab = NULL;
1639     } else {
1640         /* kernel-side HPT not needed, allocate in userspace instead */
1641         size_t size = 1ULL << shift;
1642         int i;
1643 
1644         spapr->htab = qemu_memalign(size, size);
1645         memset(spapr->htab, 0, size);
1646         spapr->htab_shift = shift;
1647 
1648         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1649             DIRTY_HPTE(HPTE(spapr->htab, i));
1650         }
1651     }
1652     /* We're setting up a hash table, so that means we're not radix */
1653     spapr->patb_entry = 0;
1654     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1655     return 0;
1656 }
1657 
1658 void spapr_setup_hpt(SpaprMachineState *spapr)
1659 {
1660     int hpt_shift;
1661 
1662     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1663         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1664     } else {
1665         uint64_t current_ram_size;
1666 
1667         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1668         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1669     }
1670     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1671 
1672     if (kvm_enabled()) {
1673         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1674 
1675         /* Check our RMA fits in the possible VRMA */
1676         if (vrma_limit < spapr->rma_size) {
1677             error_report("Unable to create %" HWADDR_PRIu
1678                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1679                          spapr->rma_size / MiB, vrma_limit / MiB);
1680             exit(EXIT_FAILURE);
1681         }
1682     }
1683 }
1684 
1685 void spapr_check_mmu_mode(bool guest_radix)
1686 {
1687     if (guest_radix) {
1688         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1689             error_report("Guest requested unavailable MMU mode (radix).");
1690             exit(EXIT_FAILURE);
1691         }
1692     } else {
1693         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1694             && !kvmppc_has_cap_mmu_hash_v3()) {
1695             error_report("Guest requested unavailable MMU mode (hash).");
1696             exit(EXIT_FAILURE);
1697         }
1698     }
1699 }
1700 
1701 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1702 {
1703     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1704     PowerPCCPU *first_ppc_cpu;
1705     hwaddr fdt_addr;
1706     void *fdt;
1707     int rc;
1708 
1709     if (reason != SHUTDOWN_CAUSE_SNAPSHOT_LOAD) {
1710         /*
1711          * Record-replay snapshot load must not consume random, this was
1712          * already replayed from initial machine reset.
1713          */
1714         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1715     }
1716 
1717     pef_kvm_reset(machine->cgs, &error_fatal);
1718     spapr_caps_apply(spapr);
1719     spapr_nested_reset(spapr);
1720 
1721     first_ppc_cpu = POWERPC_CPU(first_cpu);
1722     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1723         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1724                               spapr->max_compat_pvr)) {
1725         /*
1726          * If using KVM with radix mode available, VCPUs can be started
1727          * without a HPT because KVM will start them in radix mode.
1728          * Set the GR bit in PATE so that we know there is no HPT.
1729          */
1730         spapr->patb_entry = PATE1_GR;
1731         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1732     } else {
1733         spapr_setup_hpt(spapr);
1734     }
1735 
1736     qemu_devices_reset(reason);
1737 
1738     spapr_ovec_cleanup(spapr->ov5_cas);
1739     spapr->ov5_cas = spapr_ovec_new();
1740 
1741     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1742 
1743     /*
1744      * This is fixing some of the default configuration of the XIVE
1745      * devices. To be called after the reset of the machine devices.
1746      */
1747     spapr_irq_reset(spapr, &error_fatal);
1748 
1749     /*
1750      * There is no CAS under qtest. Simulate one to please the code that
1751      * depends on spapr->ov5_cas. This is especially needed to test device
1752      * unplug, so we do that before resetting the DRCs.
1753      */
1754     if (qtest_enabled()) {
1755         spapr_ovec_cleanup(spapr->ov5_cas);
1756         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1757     }
1758 
1759     spapr_nvdimm_finish_flushes();
1760 
1761     /* DRC reset may cause a device to be unplugged. This will cause troubles
1762      * if this device is used by another device (eg, a running vhost backend
1763      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1764      * situations, we reset DRCs after all devices have been reset.
1765      */
1766     spapr_drc_reset_all(spapr);
1767 
1768     spapr_clear_pending_events(spapr);
1769 
1770     /*
1771      * We place the device tree just below either the top of the RMA,
1772      * or just below 2GB, whichever is lower, so that it can be
1773      * processed with 32-bit real mode code if necessary
1774      */
1775     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1776 
1777     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1778     if (spapr->vof) {
1779         spapr_vof_reset(spapr, fdt, &error_fatal);
1780         /*
1781          * Do not pack the FDT as the client may change properties.
1782          * VOF client does not expect the FDT so we do not load it to the VM.
1783          */
1784     } else {
1785         rc = fdt_pack(fdt);
1786         /* Should only fail if we've built a corrupted tree */
1787         assert(rc == 0);
1788 
1789         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1790                                   0, fdt_addr, 0);
1791         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1792     }
1793     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1794 
1795     g_free(spapr->fdt_blob);
1796     spapr->fdt_size = fdt_totalsize(fdt);
1797     spapr->fdt_initial_size = spapr->fdt_size;
1798     spapr->fdt_blob = fdt;
1799 
1800     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1801     machine->fdt = fdt;
1802 
1803     /* Set up the entry state */
1804     first_ppc_cpu->env.gpr[5] = 0;
1805 
1806     spapr->fwnmi_system_reset_addr = -1;
1807     spapr->fwnmi_machine_check_addr = -1;
1808     spapr->fwnmi_machine_check_interlock = -1;
1809 
1810     /* Signal all vCPUs waiting on this condition */
1811     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1812 
1813     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1814 }
1815 
1816 static void spapr_create_nvram(SpaprMachineState *spapr)
1817 {
1818     DeviceState *dev = qdev_new("spapr-nvram");
1819     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1820 
1821     if (dinfo) {
1822         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1823                                 &error_fatal);
1824     }
1825 
1826     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1827 
1828     spapr->nvram = (struct SpaprNvram *)dev;
1829 }
1830 
1831 static void spapr_rtc_create(SpaprMachineState *spapr)
1832 {
1833     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1834                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1835                                        &error_fatal, NULL);
1836     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1837     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1838                               "date");
1839 }
1840 
1841 /* Returns whether we want to use VGA or not */
1842 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1843 {
1844     vga_interface_created = true;
1845     switch (vga_interface_type) {
1846     case VGA_NONE:
1847         return false;
1848     case VGA_DEVICE:
1849         return true;
1850     case VGA_STD:
1851     case VGA_VIRTIO:
1852     case VGA_CIRRUS:
1853         return pci_vga_init(pci_bus) != NULL;
1854     default:
1855         error_setg(errp,
1856                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1857         return false;
1858     }
1859 }
1860 
1861 static int spapr_pre_load(void *opaque)
1862 {
1863     int rc;
1864 
1865     rc = spapr_caps_pre_load(opaque);
1866     if (rc) {
1867         return rc;
1868     }
1869 
1870     return 0;
1871 }
1872 
1873 static int spapr_post_load(void *opaque, int version_id)
1874 {
1875     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1876     int err = 0;
1877 
1878     err = spapr_caps_post_migration(spapr);
1879     if (err) {
1880         return err;
1881     }
1882 
1883     /*
1884      * In earlier versions, there was no separate qdev for the PAPR
1885      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1886      * So when migrating from those versions, poke the incoming offset
1887      * value into the RTC device
1888      */
1889     if (version_id < 3) {
1890         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1891         if (err) {
1892             return err;
1893         }
1894     }
1895 
1896     if (kvm_enabled() && spapr->patb_entry) {
1897         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1898         bool radix = !!(spapr->patb_entry & PATE1_GR);
1899         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1900 
1901         /*
1902          * Update LPCR:HR and UPRT as they may not be set properly in
1903          * the stream
1904          */
1905         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1906                             LPCR_HR | LPCR_UPRT);
1907 
1908         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1909         if (err) {
1910             error_report("Process table config unsupported by the host");
1911             return -EINVAL;
1912         }
1913     }
1914 
1915     err = spapr_irq_post_load(spapr, version_id);
1916     if (err) {
1917         return err;
1918     }
1919 
1920     return err;
1921 }
1922 
1923 static int spapr_pre_save(void *opaque)
1924 {
1925     int rc;
1926 
1927     rc = spapr_caps_pre_save(opaque);
1928     if (rc) {
1929         return rc;
1930     }
1931 
1932     return 0;
1933 }
1934 
1935 static bool version_before_3(void *opaque, int version_id)
1936 {
1937     return version_id < 3;
1938 }
1939 
1940 static bool spapr_pending_events_needed(void *opaque)
1941 {
1942     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1943     return !QTAILQ_EMPTY(&spapr->pending_events);
1944 }
1945 
1946 static const VMStateDescription vmstate_spapr_event_entry = {
1947     .name = "spapr_event_log_entry",
1948     .version_id = 1,
1949     .minimum_version_id = 1,
1950     .fields = (const VMStateField[]) {
1951         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1952         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1953         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1954                                      NULL, extended_length),
1955         VMSTATE_END_OF_LIST()
1956     },
1957 };
1958 
1959 static const VMStateDescription vmstate_spapr_pending_events = {
1960     .name = "spapr_pending_events",
1961     .version_id = 1,
1962     .minimum_version_id = 1,
1963     .needed = spapr_pending_events_needed,
1964     .fields = (const VMStateField[]) {
1965         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1966                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1967         VMSTATE_END_OF_LIST()
1968     },
1969 };
1970 
1971 static bool spapr_ov5_cas_needed(void *opaque)
1972 {
1973     SpaprMachineState *spapr = opaque;
1974     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1975     bool cas_needed;
1976 
1977     /* Prior to the introduction of SpaprOptionVector, we had two option
1978      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1979      * Both of these options encode machine topology into the device-tree
1980      * in such a way that the now-booted OS should still be able to interact
1981      * appropriately with QEMU regardless of what options were actually
1982      * negotiatied on the source side.
1983      *
1984      * As such, we can avoid migrating the CAS-negotiated options if these
1985      * are the only options available on the current machine/platform.
1986      * Since these are the only options available for pseries-2.7 and
1987      * earlier, this allows us to maintain old->new/new->old migration
1988      * compatibility.
1989      *
1990      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1991      * via default pseries-2.8 machines and explicit command-line parameters.
1992      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1993      * of the actual CAS-negotiated values to continue working properly. For
1994      * example, availability of memory unplug depends on knowing whether
1995      * OV5_HP_EVT was negotiated via CAS.
1996      *
1997      * Thus, for any cases where the set of available CAS-negotiatable
1998      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1999      * include the CAS-negotiated options in the migration stream, unless
2000      * if they affect boot time behaviour only.
2001      */
2002     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2003     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2004     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2005 
2006     /* We need extra information if we have any bits outside the mask
2007      * defined above */
2008     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
2009 
2010     spapr_ovec_cleanup(ov5_mask);
2011 
2012     return cas_needed;
2013 }
2014 
2015 static const VMStateDescription vmstate_spapr_ov5_cas = {
2016     .name = "spapr_option_vector_ov5_cas",
2017     .version_id = 1,
2018     .minimum_version_id = 1,
2019     .needed = spapr_ov5_cas_needed,
2020     .fields = (const VMStateField[]) {
2021         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2022                                  vmstate_spapr_ovec, SpaprOptionVector),
2023         VMSTATE_END_OF_LIST()
2024     },
2025 };
2026 
2027 static bool spapr_patb_entry_needed(void *opaque)
2028 {
2029     SpaprMachineState *spapr = opaque;
2030 
2031     return !!spapr->patb_entry;
2032 }
2033 
2034 static const VMStateDescription vmstate_spapr_patb_entry = {
2035     .name = "spapr_patb_entry",
2036     .version_id = 1,
2037     .minimum_version_id = 1,
2038     .needed = spapr_patb_entry_needed,
2039     .fields = (const VMStateField[]) {
2040         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2041         VMSTATE_END_OF_LIST()
2042     },
2043 };
2044 
2045 static bool spapr_irq_map_needed(void *opaque)
2046 {
2047     SpaprMachineState *spapr = opaque;
2048 
2049     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2050 }
2051 
2052 static const VMStateDescription vmstate_spapr_irq_map = {
2053     .name = "spapr_irq_map",
2054     .version_id = 1,
2055     .minimum_version_id = 1,
2056     .needed = spapr_irq_map_needed,
2057     .fields = (const VMStateField[]) {
2058         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2059         VMSTATE_END_OF_LIST()
2060     },
2061 };
2062 
2063 static bool spapr_dtb_needed(void *opaque)
2064 {
2065     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2066 
2067     return smc->update_dt_enabled;
2068 }
2069 
2070 static int spapr_dtb_pre_load(void *opaque)
2071 {
2072     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2073 
2074     g_free(spapr->fdt_blob);
2075     spapr->fdt_blob = NULL;
2076     spapr->fdt_size = 0;
2077 
2078     return 0;
2079 }
2080 
2081 static const VMStateDescription vmstate_spapr_dtb = {
2082     .name = "spapr_dtb",
2083     .version_id = 1,
2084     .minimum_version_id = 1,
2085     .needed = spapr_dtb_needed,
2086     .pre_load = spapr_dtb_pre_load,
2087     .fields = (const VMStateField[]) {
2088         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2089         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2090         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2091                                      fdt_size),
2092         VMSTATE_END_OF_LIST()
2093     },
2094 };
2095 
2096 static bool spapr_fwnmi_needed(void *opaque)
2097 {
2098     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2099 
2100     return spapr->fwnmi_machine_check_addr != -1;
2101 }
2102 
2103 static int spapr_fwnmi_pre_save(void *opaque)
2104 {
2105     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2106 
2107     /*
2108      * Check if machine check handling is in progress and print a
2109      * warning message.
2110      */
2111     if (spapr->fwnmi_machine_check_interlock != -1) {
2112         warn_report("A machine check is being handled during migration. The"
2113                 "handler may run and log hardware error on the destination");
2114     }
2115 
2116     return 0;
2117 }
2118 
2119 static const VMStateDescription vmstate_spapr_fwnmi = {
2120     .name = "spapr_fwnmi",
2121     .version_id = 1,
2122     .minimum_version_id = 1,
2123     .needed = spapr_fwnmi_needed,
2124     .pre_save = spapr_fwnmi_pre_save,
2125     .fields = (const VMStateField[]) {
2126         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2127         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2128         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2129         VMSTATE_END_OF_LIST()
2130     },
2131 };
2132 
2133 static const VMStateDescription vmstate_spapr = {
2134     .name = "spapr",
2135     .version_id = 3,
2136     .minimum_version_id = 1,
2137     .pre_load = spapr_pre_load,
2138     .post_load = spapr_post_load,
2139     .pre_save = spapr_pre_save,
2140     .fields = (const VMStateField[]) {
2141         /* used to be @next_irq */
2142         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2143 
2144         /* RTC offset */
2145         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2146 
2147         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2148         VMSTATE_END_OF_LIST()
2149     },
2150     .subsections = (const VMStateDescription * const []) {
2151         &vmstate_spapr_ov5_cas,
2152         &vmstate_spapr_patb_entry,
2153         &vmstate_spapr_pending_events,
2154         &vmstate_spapr_cap_htm,
2155         &vmstate_spapr_cap_vsx,
2156         &vmstate_spapr_cap_dfp,
2157         &vmstate_spapr_cap_cfpc,
2158         &vmstate_spapr_cap_sbbc,
2159         &vmstate_spapr_cap_ibs,
2160         &vmstate_spapr_cap_hpt_maxpagesize,
2161         &vmstate_spapr_irq_map,
2162         &vmstate_spapr_cap_nested_kvm_hv,
2163         &vmstate_spapr_dtb,
2164         &vmstate_spapr_cap_large_decr,
2165         &vmstate_spapr_cap_ccf_assist,
2166         &vmstate_spapr_cap_fwnmi,
2167         &vmstate_spapr_fwnmi,
2168         &vmstate_spapr_cap_rpt_invalidate,
2169         &vmstate_spapr_cap_nested_papr,
2170         NULL
2171     }
2172 };
2173 
2174 static int htab_save_setup(QEMUFile *f, void *opaque)
2175 {
2176     SpaprMachineState *spapr = opaque;
2177 
2178     /* "Iteration" header */
2179     if (!spapr->htab_shift) {
2180         qemu_put_be32(f, -1);
2181     } else {
2182         qemu_put_be32(f, spapr->htab_shift);
2183     }
2184 
2185     if (spapr->htab) {
2186         spapr->htab_save_index = 0;
2187         spapr->htab_first_pass = true;
2188     } else {
2189         if (spapr->htab_shift) {
2190             assert(kvm_enabled());
2191         }
2192     }
2193 
2194 
2195     return 0;
2196 }
2197 
2198 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2199                             int chunkstart, int n_valid, int n_invalid)
2200 {
2201     qemu_put_be32(f, chunkstart);
2202     qemu_put_be16(f, n_valid);
2203     qemu_put_be16(f, n_invalid);
2204     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2205                     HASH_PTE_SIZE_64 * n_valid);
2206 }
2207 
2208 static void htab_save_end_marker(QEMUFile *f)
2209 {
2210     qemu_put_be32(f, 0);
2211     qemu_put_be16(f, 0);
2212     qemu_put_be16(f, 0);
2213 }
2214 
2215 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2216                                  int64_t max_ns)
2217 {
2218     bool has_timeout = max_ns != -1;
2219     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2220     int index = spapr->htab_save_index;
2221     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2222 
2223     assert(spapr->htab_first_pass);
2224 
2225     do {
2226         int chunkstart;
2227 
2228         /* Consume invalid HPTEs */
2229         while ((index < htabslots)
2230                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2231             CLEAN_HPTE(HPTE(spapr->htab, index));
2232             index++;
2233         }
2234 
2235         /* Consume valid HPTEs */
2236         chunkstart = index;
2237         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2238                && HPTE_VALID(HPTE(spapr->htab, index))) {
2239             CLEAN_HPTE(HPTE(spapr->htab, index));
2240             index++;
2241         }
2242 
2243         if (index > chunkstart) {
2244             int n_valid = index - chunkstart;
2245 
2246             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2247 
2248             if (has_timeout &&
2249                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2250                 break;
2251             }
2252         }
2253     } while ((index < htabslots) && !migration_rate_exceeded(f));
2254 
2255     if (index >= htabslots) {
2256         assert(index == htabslots);
2257         index = 0;
2258         spapr->htab_first_pass = false;
2259     }
2260     spapr->htab_save_index = index;
2261 }
2262 
2263 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2264                                 int64_t max_ns)
2265 {
2266     bool final = max_ns < 0;
2267     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2268     int examined = 0, sent = 0;
2269     int index = spapr->htab_save_index;
2270     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2271 
2272     assert(!spapr->htab_first_pass);
2273 
2274     do {
2275         int chunkstart, invalidstart;
2276 
2277         /* Consume non-dirty HPTEs */
2278         while ((index < htabslots)
2279                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2280             index++;
2281             examined++;
2282         }
2283 
2284         chunkstart = index;
2285         /* Consume valid dirty HPTEs */
2286         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2287                && HPTE_DIRTY(HPTE(spapr->htab, index))
2288                && HPTE_VALID(HPTE(spapr->htab, index))) {
2289             CLEAN_HPTE(HPTE(spapr->htab, index));
2290             index++;
2291             examined++;
2292         }
2293 
2294         invalidstart = index;
2295         /* Consume invalid dirty HPTEs */
2296         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2297                && HPTE_DIRTY(HPTE(spapr->htab, index))
2298                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2299             CLEAN_HPTE(HPTE(spapr->htab, index));
2300             index++;
2301             examined++;
2302         }
2303 
2304         if (index > chunkstart) {
2305             int n_valid = invalidstart - chunkstart;
2306             int n_invalid = index - invalidstart;
2307 
2308             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2309             sent += index - chunkstart;
2310 
2311             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2312                 break;
2313             }
2314         }
2315 
2316         if (examined >= htabslots) {
2317             break;
2318         }
2319 
2320         if (index >= htabslots) {
2321             assert(index == htabslots);
2322             index = 0;
2323         }
2324     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2325 
2326     if (index >= htabslots) {
2327         assert(index == htabslots);
2328         index = 0;
2329     }
2330 
2331     spapr->htab_save_index = index;
2332 
2333     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2334 }
2335 
2336 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2337 #define MAX_KVM_BUF_SIZE    2048
2338 
2339 static int htab_save_iterate(QEMUFile *f, void *opaque)
2340 {
2341     SpaprMachineState *spapr = opaque;
2342     int fd;
2343     int rc = 0;
2344 
2345     /* Iteration header */
2346     if (!spapr->htab_shift) {
2347         qemu_put_be32(f, -1);
2348         return 1;
2349     } else {
2350         qemu_put_be32(f, 0);
2351     }
2352 
2353     if (!spapr->htab) {
2354         assert(kvm_enabled());
2355 
2356         fd = get_htab_fd(spapr);
2357         if (fd < 0) {
2358             return fd;
2359         }
2360 
2361         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2362         if (rc < 0) {
2363             return rc;
2364         }
2365     } else  if (spapr->htab_first_pass) {
2366         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2367     } else {
2368         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2369     }
2370 
2371     htab_save_end_marker(f);
2372 
2373     return rc;
2374 }
2375 
2376 static int htab_save_complete(QEMUFile *f, void *opaque)
2377 {
2378     SpaprMachineState *spapr = opaque;
2379     int fd;
2380 
2381     /* Iteration header */
2382     if (!spapr->htab_shift) {
2383         qemu_put_be32(f, -1);
2384         return 0;
2385     } else {
2386         qemu_put_be32(f, 0);
2387     }
2388 
2389     if (!spapr->htab) {
2390         int rc;
2391 
2392         assert(kvm_enabled());
2393 
2394         fd = get_htab_fd(spapr);
2395         if (fd < 0) {
2396             return fd;
2397         }
2398 
2399         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2400         if (rc < 0) {
2401             return rc;
2402         }
2403     } else {
2404         if (spapr->htab_first_pass) {
2405             htab_save_first_pass(f, spapr, -1);
2406         }
2407         htab_save_later_pass(f, spapr, -1);
2408     }
2409 
2410     /* End marker */
2411     htab_save_end_marker(f);
2412 
2413     return 0;
2414 }
2415 
2416 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2417 {
2418     SpaprMachineState *spapr = opaque;
2419     uint32_t section_hdr;
2420     int fd = -1;
2421     Error *local_err = NULL;
2422 
2423     if (version_id < 1 || version_id > 1) {
2424         error_report("htab_load() bad version");
2425         return -EINVAL;
2426     }
2427 
2428     section_hdr = qemu_get_be32(f);
2429 
2430     if (section_hdr == -1) {
2431         spapr_free_hpt(spapr);
2432         return 0;
2433     }
2434 
2435     if (section_hdr) {
2436         int ret;
2437 
2438         /* First section gives the htab size */
2439         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2440         if (ret < 0) {
2441             error_report_err(local_err);
2442             return ret;
2443         }
2444         return 0;
2445     }
2446 
2447     if (!spapr->htab) {
2448         assert(kvm_enabled());
2449 
2450         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2451         if (fd < 0) {
2452             error_report_err(local_err);
2453             return fd;
2454         }
2455     }
2456 
2457     while (true) {
2458         uint32_t index;
2459         uint16_t n_valid, n_invalid;
2460 
2461         index = qemu_get_be32(f);
2462         n_valid = qemu_get_be16(f);
2463         n_invalid = qemu_get_be16(f);
2464 
2465         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2466             /* End of Stream */
2467             break;
2468         }
2469 
2470         if ((index + n_valid + n_invalid) >
2471             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2472             /* Bad index in stream */
2473             error_report(
2474                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2475                 index, n_valid, n_invalid, spapr->htab_shift);
2476             return -EINVAL;
2477         }
2478 
2479         if (spapr->htab) {
2480             if (n_valid) {
2481                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2482                                 HASH_PTE_SIZE_64 * n_valid);
2483             }
2484             if (n_invalid) {
2485                 memset(HPTE(spapr->htab, index + n_valid), 0,
2486                        HASH_PTE_SIZE_64 * n_invalid);
2487             }
2488         } else {
2489             int rc;
2490 
2491             assert(fd >= 0);
2492 
2493             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2494                                         &local_err);
2495             if (rc < 0) {
2496                 error_report_err(local_err);
2497                 return rc;
2498             }
2499         }
2500     }
2501 
2502     if (!spapr->htab) {
2503         assert(fd >= 0);
2504         close(fd);
2505     }
2506 
2507     return 0;
2508 }
2509 
2510 static void htab_save_cleanup(void *opaque)
2511 {
2512     SpaprMachineState *spapr = opaque;
2513 
2514     close_htab_fd(spapr);
2515 }
2516 
2517 static SaveVMHandlers savevm_htab_handlers = {
2518     .save_setup = htab_save_setup,
2519     .save_live_iterate = htab_save_iterate,
2520     .save_live_complete_precopy = htab_save_complete,
2521     .save_cleanup = htab_save_cleanup,
2522     .load_state = htab_load,
2523 };
2524 
2525 static void spapr_boot_set(void *opaque, const char *boot_device,
2526                            Error **errp)
2527 {
2528     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2529 
2530     g_free(spapr->boot_device);
2531     spapr->boot_device = g_strdup(boot_device);
2532 }
2533 
2534 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2535 {
2536     MachineState *machine = MACHINE(spapr);
2537     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2538     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2539     int i;
2540 
2541     g_assert(!nr_lmbs || machine->device_memory);
2542     for (i = 0; i < nr_lmbs; i++) {
2543         uint64_t addr;
2544 
2545         addr = i * lmb_size + machine->device_memory->base;
2546         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2547                                addr / lmb_size);
2548     }
2549 }
2550 
2551 /*
2552  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2553  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2554  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2555  */
2556 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2557 {
2558     int i;
2559 
2560     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2561         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2562                    " is not aligned to %" PRIu64 " MiB",
2563                    machine->ram_size,
2564                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2565         return;
2566     }
2567 
2568     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2569         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2570                    " is not aligned to %" PRIu64 " MiB",
2571                    machine->ram_size,
2572                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2573         return;
2574     }
2575 
2576     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2577         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2578             error_setg(errp,
2579                        "Node %d memory size 0x%" PRIx64
2580                        " is not aligned to %" PRIu64 " MiB",
2581                        i, machine->numa_state->nodes[i].node_mem,
2582                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2583             return;
2584         }
2585     }
2586 }
2587 
2588 /* find cpu slot in machine->possible_cpus by core_id */
2589 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2590 {
2591     int index = id / ms->smp.threads;
2592 
2593     if (index >= ms->possible_cpus->len) {
2594         return NULL;
2595     }
2596     if (idx) {
2597         *idx = index;
2598     }
2599     return &ms->possible_cpus->cpus[index];
2600 }
2601 
2602 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2603 {
2604     MachineState *ms = MACHINE(spapr);
2605     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2606     Error *local_err = NULL;
2607     bool vsmt_user = !!spapr->vsmt;
2608     int kvm_smt = kvmppc_smt_threads();
2609     int ret;
2610     unsigned int smp_threads = ms->smp.threads;
2611 
2612     if (tcg_enabled()) {
2613         if (smp_threads > 1 &&
2614             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2615                                    spapr->max_compat_pvr)) {
2616             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2617             return;
2618         }
2619 
2620         if (smp_threads > 8) {
2621             error_setg(errp, "TCG cannot support more than 8 threads/core "
2622                        "on a pseries machine");
2623             return;
2624         }
2625     }
2626     if (!is_power_of_2(smp_threads)) {
2627         error_setg(errp, "Cannot support %d threads/core on a pseries "
2628                    "machine because it must be a power of 2", smp_threads);
2629         return;
2630     }
2631 
2632     /* Determine the VSMT mode to use: */
2633     if (vsmt_user) {
2634         if (spapr->vsmt < smp_threads) {
2635             error_setg(errp, "Cannot support VSMT mode %d"
2636                        " because it must be >= threads/core (%d)",
2637                        spapr->vsmt, smp_threads);
2638             return;
2639         }
2640         /* In this case, spapr->vsmt has been set by the command line */
2641     } else if (!smc->smp_threads_vsmt) {
2642         /*
2643          * Default VSMT value is tricky, because we need it to be as
2644          * consistent as possible (for migration), but this requires
2645          * changing it for at least some existing cases.  We pick 8 as
2646          * the value that we'd get with KVM on POWER8, the
2647          * overwhelmingly common case in production systems.
2648          */
2649         spapr->vsmt = MAX(8, smp_threads);
2650     } else {
2651         spapr->vsmt = smp_threads;
2652     }
2653 
2654     /* KVM: If necessary, set the SMT mode: */
2655     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2656         ret = kvmppc_set_smt_threads(spapr->vsmt);
2657         if (ret) {
2658             /* Looks like KVM isn't able to change VSMT mode */
2659             error_setg(&local_err,
2660                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2661                        spapr->vsmt, ret);
2662             /* We can live with that if the default one is big enough
2663              * for the number of threads, and a submultiple of the one
2664              * we want.  In this case we'll waste some vcpu ids, but
2665              * behaviour will be correct */
2666             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2667                 warn_report_err(local_err);
2668             } else {
2669                 if (!vsmt_user) {
2670                     error_append_hint(&local_err,
2671                                       "On PPC, a VM with %d threads/core"
2672                                       " on a host with %d threads/core"
2673                                       " requires the use of VSMT mode %d.\n",
2674                                       smp_threads, kvm_smt, spapr->vsmt);
2675                 }
2676                 kvmppc_error_append_smt_possible_hint(&local_err);
2677                 error_propagate(errp, local_err);
2678             }
2679         }
2680     }
2681     /* else TCG: nothing to do currently */
2682 }
2683 
2684 static void spapr_init_cpus(SpaprMachineState *spapr)
2685 {
2686     MachineState *machine = MACHINE(spapr);
2687     MachineClass *mc = MACHINE_GET_CLASS(machine);
2688     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2689     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2690     const CPUArchIdList *possible_cpus;
2691     unsigned int smp_cpus = machine->smp.cpus;
2692     unsigned int smp_threads = machine->smp.threads;
2693     unsigned int max_cpus = machine->smp.max_cpus;
2694     int boot_cores_nr = smp_cpus / smp_threads;
2695     int i;
2696 
2697     possible_cpus = mc->possible_cpu_arch_ids(machine);
2698     if (mc->has_hotpluggable_cpus) {
2699         if (smp_cpus % smp_threads) {
2700             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2701                          smp_cpus, smp_threads);
2702             exit(1);
2703         }
2704         if (max_cpus % smp_threads) {
2705             error_report("max_cpus (%u) must be multiple of threads (%u)",
2706                          max_cpus, smp_threads);
2707             exit(1);
2708         }
2709     } else {
2710         if (max_cpus != smp_cpus) {
2711             error_report("This machine version does not support CPU hotplug");
2712             exit(1);
2713         }
2714         boot_cores_nr = possible_cpus->len;
2715     }
2716 
2717     if (smc->pre_2_10_has_unused_icps) {
2718         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2719             /* Dummy entries get deregistered when real ICPState objects
2720              * are registered during CPU core hotplug.
2721              */
2722             pre_2_10_vmstate_register_dummy_icp(i);
2723         }
2724     }
2725 
2726     for (i = 0; i < possible_cpus->len; i++) {
2727         int core_id = i * smp_threads;
2728 
2729         if (mc->has_hotpluggable_cpus) {
2730             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2731                                    spapr_vcpu_id(spapr, core_id));
2732         }
2733 
2734         if (i < boot_cores_nr) {
2735             Object *core  = object_new(type);
2736             int nr_threads = smp_threads;
2737 
2738             /* Handle the partially filled core for older machine types */
2739             if ((i + 1) * smp_threads >= smp_cpus) {
2740                 nr_threads = smp_cpus - i * smp_threads;
2741             }
2742 
2743             object_property_set_int(core, "nr-threads", nr_threads,
2744                                     &error_fatal);
2745             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2746                                     &error_fatal);
2747             qdev_realize(DEVICE(core), NULL, &error_fatal);
2748 
2749             object_unref(core);
2750         }
2751     }
2752 }
2753 
2754 static PCIHostState *spapr_create_default_phb(void)
2755 {
2756     DeviceState *dev;
2757 
2758     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2759     qdev_prop_set_uint32(dev, "index", 0);
2760     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2761 
2762     return PCI_HOST_BRIDGE(dev);
2763 }
2764 
2765 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2766 {
2767     MachineState *machine = MACHINE(spapr);
2768     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2769     hwaddr rma_size = machine->ram_size;
2770     hwaddr node0_size = spapr_node0_size(machine);
2771 
2772     /* RMA has to fit in the first NUMA node */
2773     rma_size = MIN(rma_size, node0_size);
2774 
2775     /*
2776      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2777      * never exceed that
2778      */
2779     rma_size = MIN(rma_size, 1 * TiB);
2780 
2781     /*
2782      * Clamp the RMA size based on machine type.  This is for
2783      * migration compatibility with older qemu versions, which limited
2784      * the RMA size for complicated and mostly bad reasons.
2785      */
2786     if (smc->rma_limit) {
2787         rma_size = MIN(rma_size, smc->rma_limit);
2788     }
2789 
2790     if (rma_size < MIN_RMA_SLOF) {
2791         error_setg(errp,
2792                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2793                    "ldMiB guest RMA (Real Mode Area memory)",
2794                    MIN_RMA_SLOF / MiB);
2795         return 0;
2796     }
2797 
2798     return rma_size;
2799 }
2800 
2801 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2802 {
2803     MachineState *machine = MACHINE(spapr);
2804     int i;
2805 
2806     for (i = 0; i < machine->ram_slots; i++) {
2807         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2808     }
2809 }
2810 
2811 /* pSeries LPAR / sPAPR hardware init */
2812 static void spapr_machine_init(MachineState *machine)
2813 {
2814     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2815     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2816     MachineClass *mc = MACHINE_GET_CLASS(machine);
2817     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2818     const char *bios_name = machine->firmware ?: bios_default;
2819     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2820     const char *kernel_filename = machine->kernel_filename;
2821     const char *initrd_filename = machine->initrd_filename;
2822     PCIHostState *phb;
2823     bool has_vga;
2824     int i;
2825     MemoryRegion *sysmem = get_system_memory();
2826     long load_limit, fw_size;
2827     Error *resize_hpt_err = NULL;
2828     NICInfo *nd;
2829 
2830     if (!filename) {
2831         error_report("Could not find LPAR firmware '%s'", bios_name);
2832         exit(1);
2833     }
2834     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2835     if (fw_size <= 0) {
2836         error_report("Could not load LPAR firmware '%s'", filename);
2837         exit(1);
2838     }
2839 
2840     /*
2841      * if Secure VM (PEF) support is configured, then initialize it
2842      */
2843     pef_kvm_init(machine->cgs, &error_fatal);
2844 
2845     msi_nonbroken = true;
2846 
2847     QLIST_INIT(&spapr->phbs);
2848     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2849 
2850     /* Determine capabilities to run with */
2851     spapr_caps_init(spapr);
2852 
2853     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2854     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2855         /*
2856          * If the user explicitly requested a mode we should either
2857          * supply it, or fail completely (which we do below).  But if
2858          * it's not set explicitly, we reset our mode to something
2859          * that works
2860          */
2861         if (resize_hpt_err) {
2862             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2863             error_free(resize_hpt_err);
2864             resize_hpt_err = NULL;
2865         } else {
2866             spapr->resize_hpt = smc->resize_hpt_default;
2867         }
2868     }
2869 
2870     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2871 
2872     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2873         /*
2874          * User requested HPT resize, but this host can't supply it.  Bail out
2875          */
2876         error_report_err(resize_hpt_err);
2877         exit(1);
2878     }
2879     error_free(resize_hpt_err);
2880 
2881     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2882 
2883     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2884     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2885 
2886     /*
2887      * VSMT must be set in order to be able to compute VCPU ids, ie to
2888      * call spapr_max_server_number() or spapr_vcpu_id().
2889      */
2890     spapr_set_vsmt_mode(spapr, &error_fatal);
2891 
2892     /* Set up Interrupt Controller before we create the VCPUs */
2893     spapr_irq_init(spapr, &error_fatal);
2894 
2895     /* Set up containers for ibm,client-architecture-support negotiated options
2896      */
2897     spapr->ov5 = spapr_ovec_new();
2898     spapr->ov5_cas = spapr_ovec_new();
2899 
2900     if (smc->dr_lmb_enabled) {
2901         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2902         spapr_validate_node_memory(machine, &error_fatal);
2903     }
2904 
2905     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2906 
2907     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2908     if (!smc->pre_6_2_numa_affinity) {
2909         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2910     }
2911 
2912     /* advertise support for dedicated HP event source to guests */
2913     if (spapr->use_hotplug_event_source) {
2914         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2915     }
2916 
2917     /* advertise support for HPT resizing */
2918     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2919         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2920     }
2921 
2922     /* advertise support for ibm,dyamic-memory-v2 */
2923     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2924 
2925     /* advertise XIVE on POWER9 machines */
2926     if (spapr->irq->xive) {
2927         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2928     }
2929 
2930     /* init CPUs */
2931     spapr_init_cpus(spapr);
2932 
2933     /* Init numa_assoc_array */
2934     spapr_numa_associativity_init(spapr, machine);
2935 
2936     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2937         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2938                               spapr->max_compat_pvr)) {
2939         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2940         /* KVM and TCG always allow GTSE with radix... */
2941         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2942     }
2943     /* ... but not with hash (currently). */
2944 
2945     if (kvm_enabled()) {
2946         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2947         kvmppc_enable_logical_ci_hcalls();
2948         kvmppc_enable_set_mode_hcall();
2949 
2950         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2951         kvmppc_enable_clear_ref_mod_hcalls();
2952 
2953         /* Enable H_PAGE_INIT */
2954         kvmppc_enable_h_page_init();
2955     }
2956 
2957     /* map RAM */
2958     memory_region_add_subregion(sysmem, 0, machine->ram);
2959 
2960     /* initialize hotplug memory address space */
2961     if (machine->ram_size < machine->maxram_size) {
2962         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2963         hwaddr device_mem_base;
2964 
2965         /*
2966          * Limit the number of hotpluggable memory slots to half the number
2967          * slots that KVM supports, leaving the other half for PCI and other
2968          * devices. However ensure that number of slots doesn't drop below 32.
2969          */
2970         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2971                            SPAPR_MAX_RAM_SLOTS;
2972 
2973         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2974             max_memslots = SPAPR_MAX_RAM_SLOTS;
2975         }
2976         if (machine->ram_slots > max_memslots) {
2977             error_report("Specified number of memory slots %"
2978                          PRIu64" exceeds max supported %d",
2979                          machine->ram_slots, max_memslots);
2980             exit(1);
2981         }
2982 
2983         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
2984         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
2985     }
2986 
2987     if (smc->dr_lmb_enabled) {
2988         spapr_create_lmb_dr_connectors(spapr);
2989     }
2990 
2991     if (mc->nvdimm_supported) {
2992         spapr_create_nvdimm_dr_connectors(spapr);
2993     }
2994 
2995     /* Set up RTAS event infrastructure */
2996     spapr_events_init(spapr);
2997 
2998     /* Set up the RTC RTAS interfaces */
2999     spapr_rtc_create(spapr);
3000 
3001     /* Set up VIO bus */
3002     spapr->vio_bus = spapr_vio_bus_init();
3003 
3004     for (i = 0; serial_hd(i); i++) {
3005         spapr_vty_create(spapr->vio_bus, serial_hd(i));
3006     }
3007 
3008     /* We always have at least the nvram device on VIO */
3009     spapr_create_nvram(spapr);
3010 
3011     /*
3012      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3013      * connectors (described in root DT node's "ibm,drc-types" property)
3014      * are pre-initialized here. additional child connectors (such as
3015      * connectors for a PHBs PCI slots) are added as needed during their
3016      * parent's realization.
3017      */
3018     if (smc->dr_phb_enabled) {
3019         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3020             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3021         }
3022     }
3023 
3024     /* Set up PCI */
3025     spapr_pci_rtas_init();
3026 
3027     phb = spapr_create_default_phb();
3028 
3029     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3030         spapr_vlan_create(spapr->vio_bus, nd);
3031     }
3032 
3033     pci_init_nic_devices(phb->bus, NULL);
3034 
3035     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3036         spapr_vscsi_create(spapr->vio_bus);
3037     }
3038 
3039     /* Graphics */
3040     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3041     if (has_vga) {
3042         spapr->want_stdout_path = !machine->enable_graphics;
3043         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3044     } else {
3045         spapr->want_stdout_path = true;
3046     }
3047 
3048     if (machine->usb) {
3049         if (smc->use_ohci_by_default) {
3050             pci_create_simple(phb->bus, -1, "pci-ohci");
3051         } else {
3052             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3053         }
3054 
3055         if (has_vga) {
3056             USBBus *usb_bus;
3057 
3058             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3059                                                               &error_abort));
3060             usb_create_simple(usb_bus, "usb-kbd");
3061             usb_create_simple(usb_bus, "usb-mouse");
3062         }
3063     }
3064 
3065     if (kernel_filename) {
3066         uint64_t loaded_addr = 0;
3067 
3068         spapr->kernel_size = load_elf(kernel_filename, NULL,
3069                                       translate_kernel_address, spapr,
3070                                       NULL, &loaded_addr, NULL, NULL, 1,
3071                                       PPC_ELF_MACHINE, 0, 0);
3072         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3073             spapr->kernel_size = load_elf(kernel_filename, NULL,
3074                                           translate_kernel_address, spapr,
3075                                           NULL, &loaded_addr, NULL, NULL, 0,
3076                                           PPC_ELF_MACHINE, 0, 0);
3077             spapr->kernel_le = spapr->kernel_size > 0;
3078         }
3079         if (spapr->kernel_size < 0) {
3080             error_report("error loading %s: %s", kernel_filename,
3081                          load_elf_strerror(spapr->kernel_size));
3082             exit(1);
3083         }
3084 
3085         if (spapr->kernel_addr != loaded_addr) {
3086             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3087                         " to 0x%"PRIx64,
3088                         spapr->kernel_addr, loaded_addr);
3089             spapr->kernel_addr = loaded_addr;
3090         }
3091 
3092         /* load initrd */
3093         if (initrd_filename) {
3094             /* Try to locate the initrd in the gap between the kernel
3095              * and the firmware. Add a bit of space just in case
3096              */
3097             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3098                                   + 0x1ffff) & ~0xffff;
3099             spapr->initrd_size = load_image_targphys(initrd_filename,
3100                                                      spapr->initrd_base,
3101                                                      load_limit
3102                                                      - spapr->initrd_base);
3103             if (spapr->initrd_size < 0) {
3104                 error_report("could not load initial ram disk '%s'",
3105                              initrd_filename);
3106                 exit(1);
3107             }
3108         }
3109     }
3110 
3111     /* FIXME: Should register things through the MachineState's qdev
3112      * interface, this is a legacy from the sPAPREnvironment structure
3113      * which predated MachineState but had a similar function */
3114     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3115     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3116                          &savevm_htab_handlers, spapr);
3117 
3118     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3119 
3120     qemu_register_boot_set(spapr_boot_set, spapr);
3121 
3122     /*
3123      * Nothing needs to be done to resume a suspended guest because
3124      * suspending does not change the machine state, so no need for
3125      * a ->wakeup method.
3126      */
3127     qemu_register_wakeup_support();
3128 
3129     if (kvm_enabled()) {
3130         /* to stop and start vmclock */
3131         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3132                                          &spapr->tb);
3133 
3134         kvmppc_spapr_enable_inkernel_multitce();
3135     }
3136 
3137     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3138     if (spapr->vof) {
3139         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3140         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3141     }
3142 
3143     spapr_watchdog_init(spapr);
3144 }
3145 
3146 #define DEFAULT_KVM_TYPE "auto"
3147 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3148 {
3149     /*
3150      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3151      * accommodate the 'HV' and 'PV' formats that exists in the
3152      * wild. The 'auto' mode is being introduced already as
3153      * lower-case, thus we don't need to bother checking for
3154      * "AUTO".
3155      */
3156     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3157         return 0;
3158     }
3159 
3160     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3161         return 1;
3162     }
3163 
3164     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3165         return 2;
3166     }
3167 
3168     error_report("Unknown kvm-type specified '%s'", vm_type);
3169     return -1;
3170 }
3171 
3172 /*
3173  * Implementation of an interface to adjust firmware path
3174  * for the bootindex property handling.
3175  */
3176 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3177                                    DeviceState *dev)
3178 {
3179 #define CAST(type, obj, name) \
3180     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3181     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3182     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3183     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3184     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3185 
3186     if (d && bus) {
3187         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3188         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3189         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3190 
3191         if (spapr) {
3192             /*
3193              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3194              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3195              * 0x8000 | (target << 8) | (bus << 5) | lun
3196              * (see the "Logical unit addressing format" table in SAM5)
3197              */
3198             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3199             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3200                                    (uint64_t)id << 48);
3201         } else if (virtio) {
3202             /*
3203              * We use SRP luns of the form 01000000 | (target << 8) | lun
3204              * in the top 32 bits of the 64-bit LUN
3205              * Note: the quote above is from SLOF and it is wrong,
3206              * the actual binding is:
3207              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3208              */
3209             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3210             if (d->lun >= 256) {
3211                 /* Use the LUN "flat space addressing method" */
3212                 id |= 0x4000;
3213             }
3214             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3215                                    (uint64_t)id << 32);
3216         } else if (usb) {
3217             /*
3218              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3219              * in the top 32 bits of the 64-bit LUN
3220              */
3221             unsigned usb_port = atoi(usb->port->path);
3222             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3223             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3224                                    (uint64_t)id << 32);
3225         }
3226     }
3227 
3228     /*
3229      * SLOF probes the USB devices, and if it recognizes that the device is a
3230      * storage device, it changes its name to "storage" instead of "usb-host",
3231      * and additionally adds a child node for the SCSI LUN, so the correct
3232      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3233      */
3234     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3235         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3236         if (usb_device_is_scsi_storage(usbdev)) {
3237             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3238         }
3239     }
3240 
3241     if (phb) {
3242         /* Replace "pci" with "pci@800000020000000" */
3243         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3244     }
3245 
3246     if (vsc) {
3247         /* Same logic as virtio above */
3248         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3249         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3250     }
3251 
3252     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3253         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3254         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3255         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3256     }
3257 
3258     if (pcidev) {
3259         return spapr_pci_fw_dev_name(pcidev);
3260     }
3261 
3262     return NULL;
3263 }
3264 
3265 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3266 {
3267     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3268 
3269     return g_strdup(spapr->kvm_type);
3270 }
3271 
3272 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3273 {
3274     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3275 
3276     g_free(spapr->kvm_type);
3277     spapr->kvm_type = g_strdup(value);
3278 }
3279 
3280 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     return spapr->use_hotplug_event_source;
3285 }
3286 
3287 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3288                                             Error **errp)
3289 {
3290     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3291 
3292     spapr->use_hotplug_event_source = value;
3293 }
3294 
3295 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3296 {
3297     return true;
3298 }
3299 
3300 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3301 {
3302     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3303 
3304     switch (spapr->resize_hpt) {
3305     case SPAPR_RESIZE_HPT_DEFAULT:
3306         return g_strdup("default");
3307     case SPAPR_RESIZE_HPT_DISABLED:
3308         return g_strdup("disabled");
3309     case SPAPR_RESIZE_HPT_ENABLED:
3310         return g_strdup("enabled");
3311     case SPAPR_RESIZE_HPT_REQUIRED:
3312         return g_strdup("required");
3313     }
3314     g_assert_not_reached();
3315 }
3316 
3317 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3318 {
3319     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3320 
3321     if (strcmp(value, "default") == 0) {
3322         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3323     } else if (strcmp(value, "disabled") == 0) {
3324         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3325     } else if (strcmp(value, "enabled") == 0) {
3326         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3327     } else if (strcmp(value, "required") == 0) {
3328         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3329     } else {
3330         error_setg(errp, "Bad value for \"resize-hpt\" property");
3331     }
3332 }
3333 
3334 static bool spapr_get_vof(Object *obj, Error **errp)
3335 {
3336     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3337 
3338     return spapr->vof != NULL;
3339 }
3340 
3341 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3342 {
3343     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3344 
3345     if (spapr->vof) {
3346         vof_cleanup(spapr->vof);
3347         g_free(spapr->vof);
3348         spapr->vof = NULL;
3349     }
3350     if (!value) {
3351         return;
3352     }
3353     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3354 }
3355 
3356 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3357 {
3358     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3359 
3360     if (spapr->irq == &spapr_irq_xics_legacy) {
3361         return g_strdup("legacy");
3362     } else if (spapr->irq == &spapr_irq_xics) {
3363         return g_strdup("xics");
3364     } else if (spapr->irq == &spapr_irq_xive) {
3365         return g_strdup("xive");
3366     } else if (spapr->irq == &spapr_irq_dual) {
3367         return g_strdup("dual");
3368     }
3369     g_assert_not_reached();
3370 }
3371 
3372 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3373 {
3374     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3375 
3376     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3377         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3378         return;
3379     }
3380 
3381     /* The legacy IRQ backend can not be set */
3382     if (strcmp(value, "xics") == 0) {
3383         spapr->irq = &spapr_irq_xics;
3384     } else if (strcmp(value, "xive") == 0) {
3385         spapr->irq = &spapr_irq_xive;
3386     } else if (strcmp(value, "dual") == 0) {
3387         spapr->irq = &spapr_irq_dual;
3388     } else {
3389         error_setg(errp, "Bad value for \"ic-mode\" property");
3390     }
3391 }
3392 
3393 static char *spapr_get_host_model(Object *obj, Error **errp)
3394 {
3395     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3396 
3397     return g_strdup(spapr->host_model);
3398 }
3399 
3400 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3401 {
3402     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3403 
3404     g_free(spapr->host_model);
3405     spapr->host_model = g_strdup(value);
3406 }
3407 
3408 static char *spapr_get_host_serial(Object *obj, Error **errp)
3409 {
3410     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3411 
3412     return g_strdup(spapr->host_serial);
3413 }
3414 
3415 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3416 {
3417     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3418 
3419     g_free(spapr->host_serial);
3420     spapr->host_serial = g_strdup(value);
3421 }
3422 
3423 static void spapr_instance_init(Object *obj)
3424 {
3425     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3426     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3427     MachineState *ms = MACHINE(spapr);
3428     MachineClass *mc = MACHINE_GET_CLASS(ms);
3429 
3430     /*
3431      * NVDIMM support went live in 5.1 without considering that, in
3432      * other archs, the user needs to enable NVDIMM support with the
3433      * 'nvdimm' machine option and the default behavior is NVDIMM
3434      * support disabled. It is too late to roll back to the standard
3435      * behavior without breaking 5.1 guests.
3436      */
3437     if (mc->nvdimm_supported) {
3438         ms->nvdimms_state->is_enabled = true;
3439     }
3440 
3441     spapr->htab_fd = -1;
3442     spapr->use_hotplug_event_source = true;
3443     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3444     object_property_add_str(obj, "kvm-type",
3445                             spapr_get_kvm_type, spapr_set_kvm_type);
3446     object_property_set_description(obj, "kvm-type",
3447                                     "Specifies the KVM virtualization mode (auto,"
3448                                     " hv, pr). Defaults to 'auto'. This mode will use"
3449                                     " any available KVM module loaded in the host,"
3450                                     " where kvm_hv takes precedence if both kvm_hv and"
3451                                     " kvm_pr are loaded.");
3452     object_property_add_bool(obj, "modern-hotplug-events",
3453                             spapr_get_modern_hotplug_events,
3454                             spapr_set_modern_hotplug_events);
3455     object_property_set_description(obj, "modern-hotplug-events",
3456                                     "Use dedicated hotplug event mechanism in"
3457                                     " place of standard EPOW events when possible"
3458                                     " (required for memory hot-unplug support)");
3459     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3460                             "Maximum permitted CPU compatibility mode");
3461 
3462     object_property_add_str(obj, "resize-hpt",
3463                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3464     object_property_set_description(obj, "resize-hpt",
3465                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3466     object_property_add_uint32_ptr(obj, "vsmt",
3467                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3468     object_property_set_description(obj, "vsmt",
3469                                     "Virtual SMT: KVM behaves as if this were"
3470                                     " the host's SMT mode");
3471 
3472     object_property_add_bool(obj, "vfio-no-msix-emulation",
3473                              spapr_get_msix_emulation, NULL);
3474 
3475     object_property_add_uint64_ptr(obj, "kernel-addr",
3476                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3477     object_property_set_description(obj, "kernel-addr",
3478                                     stringify(KERNEL_LOAD_ADDR)
3479                                     " for -kernel is the default");
3480     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3481 
3482     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3483     object_property_set_description(obj, "x-vof",
3484                                     "Enable Virtual Open Firmware (experimental)");
3485 
3486     /* The machine class defines the default interrupt controller mode */
3487     spapr->irq = smc->irq;
3488     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3489                             spapr_set_ic_mode);
3490     object_property_set_description(obj, "ic-mode",
3491                  "Specifies the interrupt controller mode (xics, xive, dual)");
3492 
3493     object_property_add_str(obj, "host-model",
3494         spapr_get_host_model, spapr_set_host_model);
3495     object_property_set_description(obj, "host-model",
3496         "Host model to advertise in guest device tree");
3497     object_property_add_str(obj, "host-serial",
3498         spapr_get_host_serial, spapr_set_host_serial);
3499     object_property_set_description(obj, "host-serial",
3500         "Host serial number to advertise in guest device tree");
3501 }
3502 
3503 static void spapr_machine_finalizefn(Object *obj)
3504 {
3505     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3506 
3507     g_free(spapr->kvm_type);
3508 }
3509 
3510 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3511 {
3512     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3513     CPUPPCState *env = cpu_env(cs);
3514 
3515     cpu_synchronize_state(cs);
3516     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3517     if (spapr->fwnmi_system_reset_addr != -1) {
3518         uint64_t rtas_addr, addr;
3519 
3520         /* get rtas addr from fdt */
3521         rtas_addr = spapr_get_rtas_addr();
3522         if (!rtas_addr) {
3523             qemu_system_guest_panicked(NULL);
3524             return;
3525         }
3526 
3527         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3528         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3529         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3530         env->gpr[3] = addr;
3531     }
3532     ppc_cpu_do_system_reset(cs);
3533     if (spapr->fwnmi_system_reset_addr != -1) {
3534         env->nip = spapr->fwnmi_system_reset_addr;
3535     }
3536 }
3537 
3538 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3539 {
3540     CPUState *cs;
3541 
3542     CPU_FOREACH(cs) {
3543         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3544     }
3545 }
3546 
3547 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3548                           void *fdt, int *fdt_start_offset, Error **errp)
3549 {
3550     uint64_t addr;
3551     uint32_t node;
3552 
3553     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3554     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3555                                     &error_abort);
3556     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3557                                              SPAPR_MEMORY_BLOCK_SIZE);
3558     return 0;
3559 }
3560 
3561 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3562                            bool dedicated_hp_event_source)
3563 {
3564     SpaprDrc *drc;
3565     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3566     int i;
3567     uint64_t addr = addr_start;
3568     bool hotplugged = spapr_drc_hotplugged(dev);
3569 
3570     for (i = 0; i < nr_lmbs; i++) {
3571         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3572                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3573         g_assert(drc);
3574 
3575         /*
3576          * memory_device_get_free_addr() provided a range of free addresses
3577          * that doesn't overlap with any existing mapping at pre-plug. The
3578          * corresponding LMB DRCs are thus assumed to be all attachable.
3579          */
3580         spapr_drc_attach(drc, dev);
3581         if (!hotplugged) {
3582             spapr_drc_reset(drc);
3583         }
3584         addr += SPAPR_MEMORY_BLOCK_SIZE;
3585     }
3586     /* send hotplug notification to the
3587      * guest only in case of hotplugged memory
3588      */
3589     if (hotplugged) {
3590         if (dedicated_hp_event_source) {
3591             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3592                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3593             g_assert(drc);
3594             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3595                                                    nr_lmbs,
3596                                                    spapr_drc_index(drc));
3597         } else {
3598             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3599                                            nr_lmbs);
3600         }
3601     }
3602 }
3603 
3604 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3605 {
3606     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3607     PCDIMMDevice *dimm = PC_DIMM(dev);
3608     uint64_t size, addr;
3609     int64_t slot;
3610     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3611 
3612     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3613 
3614     pc_dimm_plug(dimm, MACHINE(ms));
3615 
3616     if (!is_nvdimm) {
3617         addr = object_property_get_uint(OBJECT(dimm),
3618                                         PC_DIMM_ADDR_PROP, &error_abort);
3619         spapr_add_lmbs(dev, addr, size,
3620                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3621     } else {
3622         slot = object_property_get_int(OBJECT(dimm),
3623                                        PC_DIMM_SLOT_PROP, &error_abort);
3624         /* We should have valid slot number at this point */
3625         g_assert(slot >= 0);
3626         spapr_add_nvdimm(dev, slot);
3627     }
3628 }
3629 
3630 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3631                                   Error **errp)
3632 {
3633     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3634     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3635     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3636     PCDIMMDevice *dimm = PC_DIMM(dev);
3637     Error *local_err = NULL;
3638     uint64_t size;
3639     Object *memdev;
3640     hwaddr pagesize;
3641 
3642     if (!smc->dr_lmb_enabled) {
3643         error_setg(errp, "Memory hotplug not supported for this machine");
3644         return;
3645     }
3646 
3647     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3648     if (local_err) {
3649         error_propagate(errp, local_err);
3650         return;
3651     }
3652 
3653     if (is_nvdimm) {
3654         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3655             return;
3656         }
3657     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3658         error_setg(errp, "Hotplugged memory size must be a multiple of "
3659                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3660         return;
3661     }
3662 
3663     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3664                                       &error_abort);
3665     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3666     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3667         return;
3668     }
3669 
3670     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3671 }
3672 
3673 struct SpaprDimmState {
3674     PCDIMMDevice *dimm;
3675     uint32_t nr_lmbs;
3676     QTAILQ_ENTRY(SpaprDimmState) next;
3677 };
3678 
3679 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3680                                                        PCDIMMDevice *dimm)
3681 {
3682     SpaprDimmState *dimm_state = NULL;
3683 
3684     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3685         if (dimm_state->dimm == dimm) {
3686             break;
3687         }
3688     }
3689     return dimm_state;
3690 }
3691 
3692 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3693                                                       uint32_t nr_lmbs,
3694                                                       PCDIMMDevice *dimm)
3695 {
3696     SpaprDimmState *ds = NULL;
3697 
3698     /*
3699      * If this request is for a DIMM whose removal had failed earlier
3700      * (due to guest's refusal to remove the LMBs), we would have this
3701      * dimm already in the pending_dimm_unplugs list. In that
3702      * case don't add again.
3703      */
3704     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3705     if (!ds) {
3706         ds = g_new0(SpaprDimmState, 1);
3707         ds->nr_lmbs = nr_lmbs;
3708         ds->dimm = dimm;
3709         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3710     }
3711     return ds;
3712 }
3713 
3714 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3715                                               SpaprDimmState *dimm_state)
3716 {
3717     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3718     g_free(dimm_state);
3719 }
3720 
3721 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3722                                                         PCDIMMDevice *dimm)
3723 {
3724     SpaprDrc *drc;
3725     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3726                                                   &error_abort);
3727     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3728     uint32_t avail_lmbs = 0;
3729     uint64_t addr_start, addr;
3730     int i;
3731 
3732     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3733                                           &error_abort);
3734 
3735     addr = addr_start;
3736     for (i = 0; i < nr_lmbs; i++) {
3737         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3738                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3739         g_assert(drc);
3740         if (drc->dev) {
3741             avail_lmbs++;
3742         }
3743         addr += SPAPR_MEMORY_BLOCK_SIZE;
3744     }
3745 
3746     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3747 }
3748 
3749 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3750 {
3751     SpaprDimmState *ds;
3752     PCDIMMDevice *dimm;
3753     SpaprDrc *drc;
3754     uint32_t nr_lmbs;
3755     uint64_t size, addr_start, addr;
3756     g_autofree char *qapi_error = NULL;
3757     int i;
3758 
3759     if (!dev) {
3760         return;
3761     }
3762 
3763     dimm = PC_DIMM(dev);
3764     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3765 
3766     /*
3767      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3768      * unplug state, but one of its DRC is marked as unplug_requested.
3769      * This is bad and weird enough to g_assert() out.
3770      */
3771     g_assert(ds);
3772 
3773     spapr_pending_dimm_unplugs_remove(spapr, ds);
3774 
3775     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3776     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3777 
3778     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3779                                           &error_abort);
3780 
3781     addr = addr_start;
3782     for (i = 0; i < nr_lmbs; i++) {
3783         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3784                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3785         g_assert(drc);
3786 
3787         drc->unplug_requested = false;
3788         addr += SPAPR_MEMORY_BLOCK_SIZE;
3789     }
3790 
3791     /*
3792      * Tell QAPI that something happened and the memory
3793      * hotunplug wasn't successful. Keep sending
3794      * MEM_UNPLUG_ERROR even while sending
3795      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3796      * MEM_UNPLUG_ERROR is due.
3797      */
3798     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3799                                  "for device %s", dev->id);
3800 
3801     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3802 
3803     qapi_event_send_device_unplug_guest_error(dev->id,
3804                                               dev->canonical_path);
3805 }
3806 
3807 /* Callback to be called during DRC release. */
3808 void spapr_lmb_release(DeviceState *dev)
3809 {
3810     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3811     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3812     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3813 
3814     /* This information will get lost if a migration occurs
3815      * during the unplug process. In this case recover it. */
3816     if (ds == NULL) {
3817         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3818         g_assert(ds);
3819         /* The DRC being examined by the caller at least must be counted */
3820         g_assert(ds->nr_lmbs);
3821     }
3822 
3823     if (--ds->nr_lmbs) {
3824         return;
3825     }
3826 
3827     /*
3828      * Now that all the LMBs have been removed by the guest, call the
3829      * unplug handler chain. This can never fail.
3830      */
3831     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3832     object_unparent(OBJECT(dev));
3833 }
3834 
3835 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3836 {
3837     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3838     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3839 
3840     /* We really shouldn't get this far without anything to unplug */
3841     g_assert(ds);
3842 
3843     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3844     qdev_unrealize(dev);
3845     spapr_pending_dimm_unplugs_remove(spapr, ds);
3846 }
3847 
3848 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3849                                         DeviceState *dev, Error **errp)
3850 {
3851     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3852     PCDIMMDevice *dimm = PC_DIMM(dev);
3853     uint32_t nr_lmbs;
3854     uint64_t size, addr_start, addr;
3855     int i;
3856     SpaprDrc *drc;
3857 
3858     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3859         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3860         return;
3861     }
3862 
3863     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3864     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3865 
3866     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3867                                           &error_abort);
3868 
3869     /*
3870      * An existing pending dimm state for this DIMM means that there is an
3871      * unplug operation in progress, waiting for the spapr_lmb_release
3872      * callback to complete the job (BQL can't cover that far). In this case,
3873      * bail out to avoid detaching DRCs that were already released.
3874      */
3875     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3876         error_setg(errp, "Memory unplug already in progress for device %s",
3877                    dev->id);
3878         return;
3879     }
3880 
3881     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3882 
3883     addr = addr_start;
3884     for (i = 0; i < nr_lmbs; i++) {
3885         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3886                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3887         g_assert(drc);
3888 
3889         spapr_drc_unplug_request(drc);
3890         addr += SPAPR_MEMORY_BLOCK_SIZE;
3891     }
3892 
3893     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3894                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3895     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3896                                               nr_lmbs, spapr_drc_index(drc));
3897 }
3898 
3899 /* Callback to be called during DRC release. */
3900 void spapr_core_release(DeviceState *dev)
3901 {
3902     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3903 
3904     /* Call the unplug handler chain. This can never fail. */
3905     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3906     object_unparent(OBJECT(dev));
3907 }
3908 
3909 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3910 {
3911     MachineState *ms = MACHINE(hotplug_dev);
3912     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3913     CPUCore *cc = CPU_CORE(dev);
3914     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3915 
3916     if (smc->pre_2_10_has_unused_icps) {
3917         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3918         int i;
3919 
3920         for (i = 0; i < cc->nr_threads; i++) {
3921             CPUState *cs = CPU(sc->threads[i]);
3922 
3923             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3924         }
3925     }
3926 
3927     assert(core_slot);
3928     core_slot->cpu = NULL;
3929     qdev_unrealize(dev);
3930 }
3931 
3932 static
3933 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3934                                Error **errp)
3935 {
3936     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3937     int index;
3938     SpaprDrc *drc;
3939     CPUCore *cc = CPU_CORE(dev);
3940 
3941     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3942         error_setg(errp, "Unable to find CPU core with core-id: %d",
3943                    cc->core_id);
3944         return;
3945     }
3946     if (index == 0) {
3947         error_setg(errp, "Boot CPU core may not be unplugged");
3948         return;
3949     }
3950 
3951     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3952                           spapr_vcpu_id(spapr, cc->core_id));
3953     g_assert(drc);
3954 
3955     if (!spapr_drc_unplug_requested(drc)) {
3956         spapr_drc_unplug_request(drc);
3957     }
3958 
3959     /*
3960      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3961      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3962      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3963      * attempt (e.g. the kernel will refuse to remove the last online
3964      * CPU), we will never attempt it again because unplug_requested
3965      * will still be 'true' in that case.
3966      */
3967     spapr_hotplug_req_remove_by_index(drc);
3968 }
3969 
3970 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3971                            void *fdt, int *fdt_start_offset, Error **errp)
3972 {
3973     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3974     CPUState *cs = CPU(core->threads[0]);
3975     PowerPCCPU *cpu = POWERPC_CPU(cs);
3976     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3977     int id = spapr_get_vcpu_id(cpu);
3978     g_autofree char *nodename = NULL;
3979     int offset;
3980 
3981     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3982     offset = fdt_add_subnode(fdt, 0, nodename);
3983 
3984     spapr_dt_cpu(cs, fdt, offset, spapr);
3985 
3986     /*
3987      * spapr_dt_cpu() does not fill the 'name' property in the
3988      * CPU node. The function is called during boot process, before
3989      * and after CAS, and overwriting the 'name' property written
3990      * by SLOF is not allowed.
3991      *
3992      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3993      * CPUs more compatible with the coldplugged ones, which have
3994      * the 'name' property. Linux Kernel also relies on this
3995      * property to identify CPU nodes.
3996      */
3997     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3998 
3999     *fdt_start_offset = offset;
4000     return 0;
4001 }
4002 
4003 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4004 {
4005     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4006     MachineClass *mc = MACHINE_GET_CLASS(spapr);
4007     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4008     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
4009     CPUCore *cc = CPU_CORE(dev);
4010     SpaprDrc *drc;
4011     CPUArchId *core_slot;
4012     int index;
4013     bool hotplugged = spapr_drc_hotplugged(dev);
4014     int i;
4015 
4016     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4017     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4018 
4019     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4020                           spapr_vcpu_id(spapr, cc->core_id));
4021 
4022     g_assert(drc || !mc->has_hotpluggable_cpus);
4023 
4024     if (drc) {
4025         /*
4026          * spapr_core_pre_plug() already buys us this is a brand new
4027          * core being plugged into a free slot. Nothing should already
4028          * be attached to the corresponding DRC.
4029          */
4030         spapr_drc_attach(drc, dev);
4031 
4032         if (hotplugged) {
4033             /*
4034              * Send hotplug notification interrupt to the guest only
4035              * in case of hotplugged CPUs.
4036              */
4037             spapr_hotplug_req_add_by_index(drc);
4038         } else {
4039             spapr_drc_reset(drc);
4040         }
4041     }
4042 
4043     core_slot->cpu = CPU(dev);
4044 
4045     /*
4046      * Set compatibility mode to match the boot CPU, which was either set
4047      * by the machine reset code or by CAS. This really shouldn't fail at
4048      * this point.
4049      */
4050     if (hotplugged) {
4051         for (i = 0; i < cc->nr_threads; i++) {
4052             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4053                            &error_abort);
4054         }
4055     }
4056 
4057     if (smc->pre_2_10_has_unused_icps) {
4058         for (i = 0; i < cc->nr_threads; i++) {
4059             CPUState *cs = CPU(core->threads[i]);
4060             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4061         }
4062     }
4063 }
4064 
4065 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4066                                 Error **errp)
4067 {
4068     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4069     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4070     CPUCore *cc = CPU_CORE(dev);
4071     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4072     const char *type = object_get_typename(OBJECT(dev));
4073     CPUArchId *core_slot;
4074     int index;
4075     unsigned int smp_threads = machine->smp.threads;
4076 
4077     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4078         error_setg(errp, "CPU hotplug not supported for this machine");
4079         return;
4080     }
4081 
4082     if (strcmp(base_core_type, type)) {
4083         error_setg(errp, "CPU core type should be %s", base_core_type);
4084         return;
4085     }
4086 
4087     if (cc->core_id % smp_threads) {
4088         error_setg(errp, "invalid core id %d", cc->core_id);
4089         return;
4090     }
4091 
4092     /*
4093      * In general we should have homogeneous threads-per-core, but old
4094      * (pre hotplug support) machine types allow the last core to have
4095      * reduced threads as a compatibility hack for when we allowed
4096      * total vcpus not a multiple of threads-per-core.
4097      */
4098     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4099         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4100                    smp_threads);
4101         return;
4102     }
4103 
4104     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4105     if (!core_slot) {
4106         error_setg(errp, "core id %d out of range", cc->core_id);
4107         return;
4108     }
4109 
4110     if (core_slot->cpu) {
4111         error_setg(errp, "core %d already populated", cc->core_id);
4112         return;
4113     }
4114 
4115     numa_cpu_pre_plug(core_slot, dev, errp);
4116 }
4117 
4118 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4119                           void *fdt, int *fdt_start_offset, Error **errp)
4120 {
4121     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4122     int intc_phandle;
4123 
4124     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4125     if (intc_phandle <= 0) {
4126         return -1;
4127     }
4128 
4129     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4130         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4131         return -1;
4132     }
4133 
4134     /* generally SLOF creates these, for hotplug it's up to QEMU */
4135     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4136 
4137     return 0;
4138 }
4139 
4140 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4141                                Error **errp)
4142 {
4143     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4144     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4145     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4146     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4147     SpaprDrc *drc;
4148 
4149     if (dev->hotplugged && !smc->dr_phb_enabled) {
4150         error_setg(errp, "PHB hotplug not supported for this machine");
4151         return false;
4152     }
4153 
4154     if (sphb->index == (uint32_t)-1) {
4155         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4156         return false;
4157     }
4158 
4159     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4160     if (drc && drc->dev) {
4161         error_setg(errp, "PHB %d already attached", sphb->index);
4162         return false;
4163     }
4164 
4165     /*
4166      * This will check that sphb->index doesn't exceed the maximum number of
4167      * PHBs for the current machine type.
4168      */
4169     return
4170         smc->phb_placement(spapr, sphb->index,
4171                            &sphb->buid, &sphb->io_win_addr,
4172                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4173                            windows_supported, sphb->dma_liobn,
4174                            errp);
4175 }
4176 
4177 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4178 {
4179     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4180     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4181     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4182     SpaprDrc *drc;
4183     bool hotplugged = spapr_drc_hotplugged(dev);
4184 
4185     if (!smc->dr_phb_enabled) {
4186         return;
4187     }
4188 
4189     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4190     /* hotplug hooks should check it's enabled before getting this far */
4191     assert(drc);
4192 
4193     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4194     spapr_drc_attach(drc, dev);
4195 
4196     if (hotplugged) {
4197         spapr_hotplug_req_add_by_index(drc);
4198     } else {
4199         spapr_drc_reset(drc);
4200     }
4201 }
4202 
4203 void spapr_phb_release(DeviceState *dev)
4204 {
4205     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4206 
4207     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4208     object_unparent(OBJECT(dev));
4209 }
4210 
4211 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4212 {
4213     qdev_unrealize(dev);
4214 }
4215 
4216 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4217                                      DeviceState *dev, Error **errp)
4218 {
4219     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4220     SpaprDrc *drc;
4221 
4222     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4223     assert(drc);
4224 
4225     if (!spapr_drc_unplug_requested(drc)) {
4226         spapr_drc_unplug_request(drc);
4227         spapr_hotplug_req_remove_by_index(drc);
4228     } else {
4229         error_setg(errp,
4230                    "PCI Host Bridge unplug already in progress for device %s",
4231                    dev->id);
4232     }
4233 }
4234 
4235 static
4236 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4237                               Error **errp)
4238 {
4239     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4240 
4241     if (spapr->tpm_proxy != NULL) {
4242         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4243         return false;
4244     }
4245 
4246     return true;
4247 }
4248 
4249 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4250 {
4251     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4252     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4253 
4254     /* Already checked in spapr_tpm_proxy_pre_plug() */
4255     g_assert(spapr->tpm_proxy == NULL);
4256 
4257     spapr->tpm_proxy = tpm_proxy;
4258 }
4259 
4260 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4261 {
4262     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4263 
4264     qdev_unrealize(dev);
4265     object_unparent(OBJECT(dev));
4266     spapr->tpm_proxy = NULL;
4267 }
4268 
4269 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4270                                       DeviceState *dev, Error **errp)
4271 {
4272     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4273         spapr_memory_plug(hotplug_dev, dev);
4274     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4275         spapr_core_plug(hotplug_dev, dev);
4276     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4277         spapr_phb_plug(hotplug_dev, dev);
4278     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4279         spapr_tpm_proxy_plug(hotplug_dev, dev);
4280     }
4281 }
4282 
4283 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4284                                         DeviceState *dev, Error **errp)
4285 {
4286     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4287         spapr_memory_unplug(hotplug_dev, dev);
4288     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4289         spapr_core_unplug(hotplug_dev, dev);
4290     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4291         spapr_phb_unplug(hotplug_dev, dev);
4292     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4293         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4294     }
4295 }
4296 
4297 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4298 {
4299     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4300         /*
4301          * CAS will process all pending unplug requests.
4302          *
4303          * HACK: a guest could theoretically have cleared all bits in OV5,
4304          * but none of the guests we care for do.
4305          */
4306         spapr_ovec_empty(spapr->ov5_cas);
4307 }
4308 
4309 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4310                                                 DeviceState *dev, Error **errp)
4311 {
4312     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4313     MachineClass *mc = MACHINE_GET_CLASS(sms);
4314     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4315 
4316     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4317         if (spapr_memory_hot_unplug_supported(sms)) {
4318             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4319         } else {
4320             error_setg(errp, "Memory hot unplug not supported for this guest");
4321         }
4322     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4323         if (!mc->has_hotpluggable_cpus) {
4324             error_setg(errp, "CPU hot unplug not supported on this machine");
4325             return;
4326         }
4327         spapr_core_unplug_request(hotplug_dev, dev, errp);
4328     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4329         if (!smc->dr_phb_enabled) {
4330             error_setg(errp, "PHB hot unplug not supported on this machine");
4331             return;
4332         }
4333         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4334     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4335         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4336     }
4337 }
4338 
4339 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4340                                           DeviceState *dev, Error **errp)
4341 {
4342     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4343         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4344     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4345         spapr_core_pre_plug(hotplug_dev, dev, errp);
4346     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4347         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4348     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4349         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4350     }
4351 }
4352 
4353 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4354                                                  DeviceState *dev)
4355 {
4356     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4357         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4358         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4359         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4360         return HOTPLUG_HANDLER(machine);
4361     }
4362     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4363         PCIDevice *pcidev = PCI_DEVICE(dev);
4364         PCIBus *root = pci_device_root_bus(pcidev);
4365         SpaprPhbState *phb =
4366             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4367                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4368 
4369         if (phb) {
4370             return HOTPLUG_HANDLER(phb);
4371         }
4372     }
4373     return NULL;
4374 }
4375 
4376 static CpuInstanceProperties
4377 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4378 {
4379     CPUArchId *core_slot;
4380     MachineClass *mc = MACHINE_GET_CLASS(machine);
4381 
4382     /* make sure possible_cpu are initialized */
4383     mc->possible_cpu_arch_ids(machine);
4384     /* get CPU core slot containing thread that matches cpu_index */
4385     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4386     assert(core_slot);
4387     return core_slot->props;
4388 }
4389 
4390 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4391 {
4392     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4393 }
4394 
4395 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4396 {
4397     int i;
4398     unsigned int smp_threads = machine->smp.threads;
4399     unsigned int smp_cpus = machine->smp.cpus;
4400     const char *core_type;
4401     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4402     MachineClass *mc = MACHINE_GET_CLASS(machine);
4403 
4404     if (!mc->has_hotpluggable_cpus) {
4405         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4406     }
4407     if (machine->possible_cpus) {
4408         assert(machine->possible_cpus->len == spapr_max_cores);
4409         return machine->possible_cpus;
4410     }
4411 
4412     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4413     if (!core_type) {
4414         error_report("Unable to find sPAPR CPU Core definition");
4415         exit(1);
4416     }
4417 
4418     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4419                              sizeof(CPUArchId) * spapr_max_cores);
4420     machine->possible_cpus->len = spapr_max_cores;
4421     for (i = 0; i < machine->possible_cpus->len; i++) {
4422         int core_id = i * smp_threads;
4423 
4424         machine->possible_cpus->cpus[i].type = core_type;
4425         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4426         machine->possible_cpus->cpus[i].arch_id = core_id;
4427         machine->possible_cpus->cpus[i].props.has_core_id = true;
4428         machine->possible_cpus->cpus[i].props.core_id = core_id;
4429     }
4430     return machine->possible_cpus;
4431 }
4432 
4433 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4434                                 uint64_t *buid, hwaddr *pio,
4435                                 hwaddr *mmio32, hwaddr *mmio64,
4436                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4437 {
4438     /*
4439      * New-style PHB window placement.
4440      *
4441      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4442      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4443      * windows.
4444      *
4445      * Some guest kernels can't work with MMIO windows above 1<<46
4446      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4447      *
4448      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4449      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4450      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4451      * 1TiB 64-bit MMIO windows for each PHB.
4452      */
4453     const uint64_t base_buid = 0x800000020000000ULL;
4454     int i;
4455 
4456     /* Sanity check natural alignments */
4457     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4458     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4459     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4460     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4461     /* Sanity check bounds */
4462     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4463                       SPAPR_PCI_MEM32_WIN_SIZE);
4464     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4465                       SPAPR_PCI_MEM64_WIN_SIZE);
4466 
4467     if (index >= SPAPR_MAX_PHBS) {
4468         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4469                    SPAPR_MAX_PHBS - 1);
4470         return false;
4471     }
4472 
4473     *buid = base_buid + index;
4474     for (i = 0; i < n_dma; ++i) {
4475         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4476     }
4477 
4478     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4479     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4480     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4481     return true;
4482 }
4483 
4484 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4485 {
4486     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4487 
4488     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4489 }
4490 
4491 static void spapr_ics_resend(XICSFabric *dev)
4492 {
4493     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4494 
4495     ics_resend(spapr->ics);
4496 }
4497 
4498 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4499 {
4500     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4501 
4502     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4503 }
4504 
4505 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4506                                  Monitor *mon)
4507 {
4508     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4509 
4510     spapr_irq_print_info(spapr, mon);
4511     monitor_printf(mon, "irqchip: %s\n",
4512                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4513 }
4514 
4515 /*
4516  * This is a XIVE only operation
4517  */
4518 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4519                            uint8_t nvt_blk, uint32_t nvt_idx,
4520                            bool cam_ignore, uint8_t priority,
4521                            uint32_t logic_serv, XiveTCTXMatch *match)
4522 {
4523     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4524     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4525     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4526     int count;
4527 
4528     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4529                            priority, logic_serv, match);
4530     if (count < 0) {
4531         return count;
4532     }
4533 
4534     /*
4535      * When we implement the save and restore of the thread interrupt
4536      * contexts in the enter/exit CPU handlers of the machine and the
4537      * escalations in QEMU, we should be able to handle non dispatched
4538      * vCPUs.
4539      *
4540      * Until this is done, the sPAPR machine should find at least one
4541      * matching context always.
4542      */
4543     if (count == 0) {
4544         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4545                       nvt_blk, nvt_idx);
4546     }
4547 
4548     return count;
4549 }
4550 
4551 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4552 {
4553     return cpu->vcpu_id;
4554 }
4555 
4556 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4557 {
4558     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4559     MachineState *ms = MACHINE(spapr);
4560     int vcpu_id;
4561 
4562     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4563 
4564     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4565         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4566         error_append_hint(errp, "Adjust the number of cpus to %d "
4567                           "or try to raise the number of threads per core\n",
4568                           vcpu_id * ms->smp.threads / spapr->vsmt);
4569         return false;
4570     }
4571 
4572     cpu->vcpu_id = vcpu_id;
4573     return true;
4574 }
4575 
4576 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4577 {
4578     CPUState *cs;
4579 
4580     CPU_FOREACH(cs) {
4581         PowerPCCPU *cpu = POWERPC_CPU(cs);
4582 
4583         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4584             return cpu;
4585         }
4586     }
4587 
4588     return NULL;
4589 }
4590 
4591 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4592 {
4593     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4594 
4595     return spapr_cpu->in_nested;
4596 }
4597 
4598 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4599 {
4600     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4601 
4602     /* These are only called by TCG, KVM maintains dispatch state */
4603 
4604     spapr_cpu->prod = false;
4605     if (spapr_cpu->vpa_addr) {
4606         CPUState *cs = CPU(cpu);
4607         uint32_t dispatch;
4608 
4609         dispatch = ldl_be_phys(cs->as,
4610                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4611         dispatch++;
4612         if ((dispatch & 1) != 0) {
4613             qemu_log_mask(LOG_GUEST_ERROR,
4614                           "VPA: incorrect dispatch counter value for "
4615                           "dispatched partition %u, correcting.\n", dispatch);
4616             dispatch++;
4617         }
4618         stl_be_phys(cs->as,
4619                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4620     }
4621 }
4622 
4623 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4624 {
4625     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4626 
4627     if (spapr_cpu->vpa_addr) {
4628         CPUState *cs = CPU(cpu);
4629         uint32_t dispatch;
4630 
4631         dispatch = ldl_be_phys(cs->as,
4632                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4633         dispatch++;
4634         if ((dispatch & 1) != 1) {
4635             qemu_log_mask(LOG_GUEST_ERROR,
4636                           "VPA: incorrect dispatch counter value for "
4637                           "preempted partition %u, correcting.\n", dispatch);
4638             dispatch++;
4639         }
4640         stl_be_phys(cs->as,
4641                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4642     }
4643 }
4644 
4645 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4646 {
4647     MachineClass *mc = MACHINE_CLASS(oc);
4648     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4649     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4650     NMIClass *nc = NMI_CLASS(oc);
4651     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4652     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4653     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4654     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4655     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4656     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4657 
4658     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4659     mc->ignore_boot_device_suffixes = true;
4660 
4661     /*
4662      * We set up the default / latest behaviour here.  The class_init
4663      * functions for the specific versioned machine types can override
4664      * these details for backwards compatibility
4665      */
4666     mc->init = spapr_machine_init;
4667     mc->reset = spapr_machine_reset;
4668     mc->block_default_type = IF_SCSI;
4669 
4670     /*
4671      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4672      * In TCG the limit is restricted by the range of CPU IPIs available.
4673      */
4674     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4675 
4676     mc->no_parallel = 1;
4677     mc->default_boot_order = "";
4678     mc->default_ram_size = 512 * MiB;
4679     mc->default_ram_id = "ppc_spapr.ram";
4680     mc->default_display = "std";
4681     mc->kvm_type = spapr_kvm_type;
4682     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4683     mc->pci_allow_0_address = true;
4684     assert(!mc->get_hotplug_handler);
4685     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4686     hc->pre_plug = spapr_machine_device_pre_plug;
4687     hc->plug = spapr_machine_device_plug;
4688     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4689     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4690     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4691     hc->unplug_request = spapr_machine_device_unplug_request;
4692     hc->unplug = spapr_machine_device_unplug;
4693 
4694     smc->dr_lmb_enabled = true;
4695     smc->update_dt_enabled = true;
4696     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4697     mc->has_hotpluggable_cpus = true;
4698     mc->nvdimm_supported = true;
4699     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4700     fwc->get_dev_path = spapr_get_fw_dev_path;
4701     nc->nmi_monitor_handler = spapr_nmi;
4702     smc->phb_placement = spapr_phb_placement;
4703     vhc->cpu_in_nested = spapr_cpu_in_nested;
4704     vhc->deliver_hv_excp = spapr_exit_nested;
4705     vhc->hypercall = emulate_spapr_hypercall;
4706     vhc->hpt_mask = spapr_hpt_mask;
4707     vhc->map_hptes = spapr_map_hptes;
4708     vhc->unmap_hptes = spapr_unmap_hptes;
4709     vhc->hpte_set_c = spapr_hpte_set_c;
4710     vhc->hpte_set_r = spapr_hpte_set_r;
4711     vhc->get_pate = spapr_get_pate;
4712     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4713     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4714     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4715     xic->ics_get = spapr_ics_get;
4716     xic->ics_resend = spapr_ics_resend;
4717     xic->icp_get = spapr_icp_get;
4718     ispc->print_info = spapr_pic_print_info;
4719     /* Force NUMA node memory size to be a multiple of
4720      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4721      * in which LMBs are represented and hot-added
4722      */
4723     mc->numa_mem_align_shift = 28;
4724     mc->auto_enable_numa = true;
4725 
4726     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4727     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4728     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4729     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4730     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4731     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4732     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4733     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4734     smc->default_caps.caps[SPAPR_CAP_NESTED_PAPR] = SPAPR_CAP_OFF;
4735     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4736     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4737     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4738     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4739 
4740     /*
4741      * This cap specifies whether the AIL 3 mode for
4742      * H_SET_RESOURCE is supported. The default is modified
4743      * by default_caps_with_cpu().
4744      */
4745     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4746     spapr_caps_add_properties(smc);
4747     smc->irq = &spapr_irq_dual;
4748     smc->dr_phb_enabled = true;
4749     smc->linux_pci_probe = true;
4750     smc->smp_threads_vsmt = true;
4751     smc->nr_xirqs = SPAPR_NR_XIRQS;
4752     xfc->match_nvt = spapr_match_nvt;
4753     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4754     vmc->quiesce = spapr_vof_quiesce;
4755     vmc->setprop = spapr_vof_setprop;
4756 }
4757 
4758 static const TypeInfo spapr_machine_info = {
4759     .name          = TYPE_SPAPR_MACHINE,
4760     .parent        = TYPE_MACHINE,
4761     .abstract      = true,
4762     .instance_size = sizeof(SpaprMachineState),
4763     .instance_init = spapr_instance_init,
4764     .instance_finalize = spapr_machine_finalizefn,
4765     .class_size    = sizeof(SpaprMachineClass),
4766     .class_init    = spapr_machine_class_init,
4767     .interfaces = (InterfaceInfo[]) {
4768         { TYPE_FW_PATH_PROVIDER },
4769         { TYPE_NMI },
4770         { TYPE_HOTPLUG_HANDLER },
4771         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4772         { TYPE_XICS_FABRIC },
4773         { TYPE_INTERRUPT_STATS_PROVIDER },
4774         { TYPE_XIVE_FABRIC },
4775         { TYPE_VOF_MACHINE_IF },
4776         { }
4777     },
4778 };
4779 
4780 static void spapr_machine_latest_class_options(MachineClass *mc)
4781 {
4782     mc->alias = "pseries";
4783     mc->is_default = true;
4784 }
4785 
4786 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4787     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4788                                                     void *data)      \
4789     {                                                                \
4790         MachineClass *mc = MACHINE_CLASS(oc);                        \
4791         spapr_machine_##suffix##_class_options(mc);                  \
4792         if (latest) {                                                \
4793             spapr_machine_latest_class_options(mc);                  \
4794         }                                                            \
4795     }                                                                \
4796     static const TypeInfo spapr_machine_##suffix##_info = {          \
4797         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4798         .parent = TYPE_SPAPR_MACHINE,                                \
4799         .class_init = spapr_machine_##suffix##_class_init,           \
4800     };                                                               \
4801     static void spapr_machine_register_##suffix(void)                \
4802     {                                                                \
4803         type_register(&spapr_machine_##suffix##_info);               \
4804     }                                                                \
4805     type_init(spapr_machine_register_##suffix)
4806 
4807 /*
4808  * pseries-9.0
4809  */
4810 static void spapr_machine_9_0_class_options(MachineClass *mc)
4811 {
4812     /* Defaults for the latest behaviour inherited from the base class */
4813 }
4814 
4815 DEFINE_SPAPR_MACHINE(9_0, "9.0", true);
4816 
4817 /*
4818  * pseries-8.2
4819  */
4820 static void spapr_machine_8_2_class_options(MachineClass *mc)
4821 {
4822     spapr_machine_9_0_class_options(mc);
4823     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4824 }
4825 
4826 DEFINE_SPAPR_MACHINE(8_2, "8.2", false);
4827 
4828 /*
4829  * pseries-8.1
4830  */
4831 static void spapr_machine_8_1_class_options(MachineClass *mc)
4832 {
4833     spapr_machine_8_2_class_options(mc);
4834     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4835 }
4836 
4837 DEFINE_SPAPR_MACHINE(8_1, "8.1", false);
4838 
4839 /*
4840  * pseries-8.0
4841  */
4842 static void spapr_machine_8_0_class_options(MachineClass *mc)
4843 {
4844     spapr_machine_8_1_class_options(mc);
4845     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4846 }
4847 
4848 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4849 
4850 /*
4851  * pseries-7.2
4852  */
4853 static void spapr_machine_7_2_class_options(MachineClass *mc)
4854 {
4855     spapr_machine_8_0_class_options(mc);
4856     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4857 }
4858 
4859 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4860 
4861 /*
4862  * pseries-7.1
4863  */
4864 static void spapr_machine_7_1_class_options(MachineClass *mc)
4865 {
4866     spapr_machine_7_2_class_options(mc);
4867     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4868 }
4869 
4870 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4871 
4872 /*
4873  * pseries-7.0
4874  */
4875 static void spapr_machine_7_0_class_options(MachineClass *mc)
4876 {
4877     spapr_machine_7_1_class_options(mc);
4878     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4879 }
4880 
4881 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4882 
4883 /*
4884  * pseries-6.2
4885  */
4886 static void spapr_machine_6_2_class_options(MachineClass *mc)
4887 {
4888     spapr_machine_7_0_class_options(mc);
4889     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4890 }
4891 
4892 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4893 
4894 /*
4895  * pseries-6.1
4896  */
4897 static void spapr_machine_6_1_class_options(MachineClass *mc)
4898 {
4899     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4900 
4901     spapr_machine_6_2_class_options(mc);
4902     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4903     smc->pre_6_2_numa_affinity = true;
4904     mc->smp_props.prefer_sockets = true;
4905 }
4906 
4907 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4908 
4909 /*
4910  * pseries-6.0
4911  */
4912 static void spapr_machine_6_0_class_options(MachineClass *mc)
4913 {
4914     spapr_machine_6_1_class_options(mc);
4915     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4916 }
4917 
4918 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4919 
4920 /*
4921  * pseries-5.2
4922  */
4923 static void spapr_machine_5_2_class_options(MachineClass *mc)
4924 {
4925     spapr_machine_6_0_class_options(mc);
4926     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4927 }
4928 
4929 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4930 
4931 /*
4932  * pseries-5.1
4933  */
4934 static void spapr_machine_5_1_class_options(MachineClass *mc)
4935 {
4936     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4937 
4938     spapr_machine_5_2_class_options(mc);
4939     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4940     smc->pre_5_2_numa_associativity = true;
4941 }
4942 
4943 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4944 
4945 /*
4946  * pseries-5.0
4947  */
4948 static void spapr_machine_5_0_class_options(MachineClass *mc)
4949 {
4950     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4951     static GlobalProperty compat[] = {
4952         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4953     };
4954 
4955     spapr_machine_5_1_class_options(mc);
4956     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4957     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4958     mc->numa_mem_supported = true;
4959     smc->pre_5_1_assoc_refpoints = true;
4960 }
4961 
4962 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4963 
4964 /*
4965  * pseries-4.2
4966  */
4967 static void spapr_machine_4_2_class_options(MachineClass *mc)
4968 {
4969     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4970 
4971     spapr_machine_5_0_class_options(mc);
4972     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4973     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4974     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4975     smc->rma_limit = 16 * GiB;
4976     mc->nvdimm_supported = false;
4977 }
4978 
4979 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4980 
4981 /*
4982  * pseries-4.1
4983  */
4984 static void spapr_machine_4_1_class_options(MachineClass *mc)
4985 {
4986     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4987     static GlobalProperty compat[] = {
4988         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4989         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4990     };
4991 
4992     spapr_machine_4_2_class_options(mc);
4993     smc->linux_pci_probe = false;
4994     smc->smp_threads_vsmt = false;
4995     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4996     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4997 }
4998 
4999 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
5000 
5001 /*
5002  * pseries-4.0
5003  */
5004 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
5005                               uint64_t *buid, hwaddr *pio,
5006                               hwaddr *mmio32, hwaddr *mmio64,
5007                               unsigned n_dma, uint32_t *liobns, Error **errp)
5008 {
5009     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
5010                              liobns, errp)) {
5011         return false;
5012     }
5013     return true;
5014 }
5015 static void spapr_machine_4_0_class_options(MachineClass *mc)
5016 {
5017     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5018 
5019     spapr_machine_4_1_class_options(mc);
5020     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5021     smc->phb_placement = phb_placement_4_0;
5022     smc->irq = &spapr_irq_xics;
5023     smc->pre_4_1_migration = true;
5024 }
5025 
5026 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
5027 
5028 /*
5029  * pseries-3.1
5030  */
5031 static void spapr_machine_3_1_class_options(MachineClass *mc)
5032 {
5033     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5034 
5035     spapr_machine_4_0_class_options(mc);
5036     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5037 
5038     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5039     smc->update_dt_enabled = false;
5040     smc->dr_phb_enabled = false;
5041     smc->broken_host_serial_model = true;
5042     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5043     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5044     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5045     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5046 }
5047 
5048 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
5049 
5050 /*
5051  * pseries-3.0
5052  */
5053 
5054 static void spapr_machine_3_0_class_options(MachineClass *mc)
5055 {
5056     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5057 
5058     spapr_machine_3_1_class_options(mc);
5059     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5060 
5061     smc->legacy_irq_allocation = true;
5062     smc->nr_xirqs = 0x400;
5063     smc->irq = &spapr_irq_xics_legacy;
5064 }
5065 
5066 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
5067 
5068 /*
5069  * pseries-2.12
5070  */
5071 static void spapr_machine_2_12_class_options(MachineClass *mc)
5072 {
5073     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5074     static GlobalProperty compat[] = {
5075         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5076         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5077     };
5078 
5079     spapr_machine_3_0_class_options(mc);
5080     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5081     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5082 
5083     /* We depend on kvm_enabled() to choose a default value for the
5084      * hpt-max-page-size capability. Of course we can't do it here
5085      * because this is too early and the HW accelerator isn't initialized
5086      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5087      */
5088     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5089 }
5090 
5091 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5092 
5093 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5094 {
5095     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5096 
5097     spapr_machine_2_12_class_options(mc);
5098     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5099     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5100     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5101 }
5102 
5103 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5104 
5105 /*
5106  * pseries-2.11
5107  */
5108 
5109 static void spapr_machine_2_11_class_options(MachineClass *mc)
5110 {
5111     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5112 
5113     spapr_machine_2_12_class_options(mc);
5114     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5115     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5116     mc->deprecation_reason = "old and not maintained - use a 2.12+ version";
5117 }
5118 
5119 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5120 
5121 /*
5122  * pseries-2.10
5123  */
5124 
5125 static void spapr_machine_2_10_class_options(MachineClass *mc)
5126 {
5127     spapr_machine_2_11_class_options(mc);
5128     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5129 }
5130 
5131 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5132 
5133 /*
5134  * pseries-2.9
5135  */
5136 
5137 static void spapr_machine_2_9_class_options(MachineClass *mc)
5138 {
5139     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5140     static GlobalProperty compat[] = {
5141         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5142     };
5143 
5144     spapr_machine_2_10_class_options(mc);
5145     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5146     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5147     smc->pre_2_10_has_unused_icps = true;
5148     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5149 }
5150 
5151 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5152 
5153 /*
5154  * pseries-2.8
5155  */
5156 
5157 static void spapr_machine_2_8_class_options(MachineClass *mc)
5158 {
5159     static GlobalProperty compat[] = {
5160         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5161     };
5162 
5163     spapr_machine_2_9_class_options(mc);
5164     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5165     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5166     mc->numa_mem_align_shift = 23;
5167 }
5168 
5169 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5170 
5171 /*
5172  * pseries-2.7
5173  */
5174 
5175 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5176                               uint64_t *buid, hwaddr *pio,
5177                               hwaddr *mmio32, hwaddr *mmio64,
5178                               unsigned n_dma, uint32_t *liobns, Error **errp)
5179 {
5180     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5181     const uint64_t base_buid = 0x800000020000000ULL;
5182     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5183     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5184     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5185     const uint32_t max_index = 255;
5186     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5187 
5188     uint64_t ram_top = MACHINE(spapr)->ram_size;
5189     hwaddr phb0_base, phb_base;
5190     int i;
5191 
5192     /* Do we have device memory? */
5193     if (MACHINE(spapr)->device_memory) {
5194         /* Can't just use maxram_size, because there may be an
5195          * alignment gap between normal and device memory regions
5196          */
5197         ram_top = MACHINE(spapr)->device_memory->base +
5198             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5199     }
5200 
5201     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5202 
5203     if (index > max_index) {
5204         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5205                    max_index);
5206         return false;
5207     }
5208 
5209     *buid = base_buid + index;
5210     for (i = 0; i < n_dma; ++i) {
5211         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5212     }
5213 
5214     phb_base = phb0_base + index * phb_spacing;
5215     *pio = phb_base + pio_offset;
5216     *mmio32 = phb_base + mmio_offset;
5217     /*
5218      * We don't set the 64-bit MMIO window, relying on the PHB's
5219      * fallback behaviour of automatically splitting a large "32-bit"
5220      * window into contiguous 32-bit and 64-bit windows
5221      */
5222 
5223     return true;
5224 }
5225 
5226 static void spapr_machine_2_7_class_options(MachineClass *mc)
5227 {
5228     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5229     static GlobalProperty compat[] = {
5230         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5231         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5232         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5233         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5234     };
5235 
5236     spapr_machine_2_8_class_options(mc);
5237     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5238     mc->default_machine_opts = "modern-hotplug-events=off";
5239     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5240     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5241     smc->phb_placement = phb_placement_2_7;
5242 }
5243 
5244 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5245 
5246 /*
5247  * pseries-2.6
5248  */
5249 
5250 static void spapr_machine_2_6_class_options(MachineClass *mc)
5251 {
5252     static GlobalProperty compat[] = {
5253         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5254     };
5255 
5256     spapr_machine_2_7_class_options(mc);
5257     mc->has_hotpluggable_cpus = false;
5258     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5259     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5260 }
5261 
5262 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5263 
5264 /*
5265  * pseries-2.5
5266  */
5267 
5268 static void spapr_machine_2_5_class_options(MachineClass *mc)
5269 {
5270     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5271     static GlobalProperty compat[] = {
5272         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5273     };
5274 
5275     spapr_machine_2_6_class_options(mc);
5276     smc->use_ohci_by_default = true;
5277     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5278     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5279 }
5280 
5281 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5282 
5283 /*
5284  * pseries-2.4
5285  */
5286 
5287 static void spapr_machine_2_4_class_options(MachineClass *mc)
5288 {
5289     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5290 
5291     spapr_machine_2_5_class_options(mc);
5292     smc->dr_lmb_enabled = false;
5293     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5294 }
5295 
5296 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5297 
5298 /*
5299  * pseries-2.3
5300  */
5301 
5302 static void spapr_machine_2_3_class_options(MachineClass *mc)
5303 {
5304     static GlobalProperty compat[] = {
5305         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5306     };
5307     spapr_machine_2_4_class_options(mc);
5308     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5309     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5310 }
5311 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5312 
5313 /*
5314  * pseries-2.2
5315  */
5316 
5317 static void spapr_machine_2_2_class_options(MachineClass *mc)
5318 {
5319     static GlobalProperty compat[] = {
5320         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5321     };
5322 
5323     spapr_machine_2_3_class_options(mc);
5324     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5325     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5326     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5327 }
5328 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5329 
5330 /*
5331  * pseries-2.1
5332  */
5333 
5334 static void spapr_machine_2_1_class_options(MachineClass *mc)
5335 {
5336     spapr_machine_2_2_class_options(mc);
5337     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5338 }
5339 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5340 
5341 static void spapr_machine_register_types(void)
5342 {
5343     type_register_static(&spapr_machine_info);
5344 }
5345 
5346 type_init(spapr_machine_register_types)
5347