xref: /openbmc/qemu/hw/ppc/spapr.c (revision 29320530)
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 "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/qapi-events-qdev.h"
33 #include "qapi/visitor.h"
34 #include "sysemu/sysemu.h"
35 #include "sysemu/hostmem.h"
36 #include "sysemu/numa.h"
37 #include "sysemu/qtest.h"
38 #include "sysemu/reset.h"
39 #include "sysemu/runstate.h"
40 #include "qemu/log.h"
41 #include "hw/fw-path-provider.h"
42 #include "elf.h"
43 #include "net/net.h"
44 #include "sysemu/device_tree.h"
45 #include "sysemu/cpus.h"
46 #include "sysemu/hw_accel.h"
47 #include "kvm_ppc.h"
48 #include "migration/misc.h"
49 #include "migration/qemu-file-types.h"
50 #include "migration/global_state.h"
51 #include "migration/register.h"
52 #include "migration/blocker.h"
53 #include "mmu-hash64.h"
54 #include "mmu-book3s-v3.h"
55 #include "cpu-models.h"
56 #include "hw/core/cpu.h"
57 
58 #include "hw/ppc/ppc.h"
59 #include "hw/loader.h"
60 
61 #include "hw/ppc/fdt.h"
62 #include "hw/ppc/spapr.h"
63 #include "hw/ppc/spapr_vio.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/pci-host/spapr.h"
66 #include "hw/pci/msi.h"
67 
68 #include "hw/pci/pci.h"
69 #include "hw/scsi/scsi.h"
70 #include "hw/virtio/virtio-scsi.h"
71 #include "hw/virtio/vhost-scsi-common.h"
72 
73 #include "exec/ram_addr.h"
74 #include "hw/usb.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
77 #include "trace.h"
78 #include "hw/nmi.h"
79 #include "hw/intc/intc.h"
80 
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
87 
88 #include "monitor/monitor.h"
89 
90 #include <libfdt.h>
91 
92 /* SLOF memory layout:
93  *
94  * SLOF raw image loaded at 0, copies its romfs right below the flat
95  * device-tree, then position SLOF itself 31M below that
96  *
97  * So we set FW_OVERHEAD to 40MB which should account for all of that
98  * and more
99  *
100  * We load our kernel at 4M, leaving space for SLOF initial image
101  */
102 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
103 #define FW_MAX_SIZE             0x400000
104 #define FW_FILE_NAME            "slof.bin"
105 #define FW_FILE_NAME_VOF        "vof.bin"
106 #define FW_OVERHEAD             0x2800000
107 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
108 
109 #define MIN_RMA_SLOF            (128 * MiB)
110 
111 #define PHANDLE_INTC            0x00001111
112 
113 /* These two functions implement the VCPU id numbering: one to compute them
114  * all and one to identify thread 0 of a VCORE. Any change to the first one
115  * is likely to have an impact on the second one, so let's keep them close.
116  */
117 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
118 {
119     MachineState *ms = MACHINE(spapr);
120     unsigned int smp_threads = ms->smp.threads;
121 
122     assert(spapr->vsmt);
123     return
124         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
125 }
126 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
127                                       PowerPCCPU *cpu)
128 {
129     assert(spapr->vsmt);
130     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
131 }
132 
133 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
134 {
135     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
136      * and newer QEMUs don't even have them. In both cases, we don't want
137      * to send anything on the wire.
138      */
139     return false;
140 }
141 
142 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
143     .name = "icp/server",
144     .version_id = 1,
145     .minimum_version_id = 1,
146     .needed = pre_2_10_vmstate_dummy_icp_needed,
147     .fields = (VMStateField[]) {
148         VMSTATE_UNUSED(4), /* uint32_t xirr */
149         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
150         VMSTATE_UNUSED(1), /* uint8_t mfrr */
151         VMSTATE_END_OF_LIST()
152     },
153 };
154 
155 static void pre_2_10_vmstate_register_dummy_icp(int i)
156 {
157     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
158                      (void *)(uintptr_t) i);
159 }
160 
161 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
162 {
163     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
164                        (void *)(uintptr_t) i);
165 }
166 
167 int spapr_max_server_number(SpaprMachineState *spapr)
168 {
169     MachineState *ms = MACHINE(spapr);
170 
171     assert(spapr->vsmt);
172     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
173 }
174 
175 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
176                                   int smt_threads)
177 {
178     int i, ret = 0;
179     uint32_t servers_prop[smt_threads];
180     uint32_t gservers_prop[smt_threads * 2];
181     int index = spapr_get_vcpu_id(cpu);
182 
183     if (cpu->compat_pvr) {
184         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
185         if (ret < 0) {
186             return ret;
187         }
188     }
189 
190     /* Build interrupt servers and gservers properties */
191     for (i = 0; i < smt_threads; i++) {
192         servers_prop[i] = cpu_to_be32(index + i);
193         /* Hack, direct the group queues back to cpu 0 */
194         gservers_prop[i*2] = cpu_to_be32(index + i);
195         gservers_prop[i*2 + 1] = 0;
196     }
197     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
198                       servers_prop, sizeof(servers_prop));
199     if (ret < 0) {
200         return ret;
201     }
202     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
203                       gservers_prop, sizeof(gservers_prop));
204 
205     return ret;
206 }
207 
208 static void spapr_dt_pa_features(SpaprMachineState *spapr,
209                                  PowerPCCPU *cpu,
210                                  void *fdt, int offset)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static hwaddr spapr_node0_size(MachineState *machine)
287 {
288     if (machine->numa_state->num_nodes) {
289         int i;
290         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
291             if (machine->numa_state->nodes[i].node_mem) {
292                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
293                            machine->ram_size);
294             }
295         }
296     }
297     return machine->ram_size;
298 }
299 
300 static void add_str(GString *s, const gchar *s1)
301 {
302     g_string_append_len(s, s1, strlen(s1) + 1);
303 }
304 
305 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
306                                 hwaddr start, hwaddr size)
307 {
308     char mem_name[32];
309     uint64_t mem_reg_property[2];
310     int off;
311 
312     mem_reg_property[0] = cpu_to_be64(start);
313     mem_reg_property[1] = cpu_to_be64(size);
314 
315     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
316     off = fdt_add_subnode(fdt, 0, mem_name);
317     _FDT(off);
318     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
319     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
320                       sizeof(mem_reg_property))));
321     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
322     return off;
323 }
324 
325 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
326 {
327     MemoryDeviceInfoList *info;
328 
329     for (info = list; info; info = info->next) {
330         MemoryDeviceInfo *value = info->value;
331 
332         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
333             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
334 
335             if (addr >= pcdimm_info->addr &&
336                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
337                 return pcdimm_info->node;
338             }
339         }
340     }
341 
342     return -1;
343 }
344 
345 struct sPAPRDrconfCellV2 {
346      uint32_t seq_lmbs;
347      uint64_t base_addr;
348      uint32_t drc_index;
349      uint32_t aa_index;
350      uint32_t flags;
351 } QEMU_PACKED;
352 
353 typedef struct DrconfCellQueue {
354     struct sPAPRDrconfCellV2 cell;
355     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
356 } DrconfCellQueue;
357 
358 static DrconfCellQueue *
359 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
360                       uint32_t drc_index, uint32_t aa_index,
361                       uint32_t flags)
362 {
363     DrconfCellQueue *elem;
364 
365     elem = g_malloc0(sizeof(*elem));
366     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
367     elem->cell.base_addr = cpu_to_be64(base_addr);
368     elem->cell.drc_index = cpu_to_be32(drc_index);
369     elem->cell.aa_index = cpu_to_be32(aa_index);
370     elem->cell.flags = cpu_to_be32(flags);
371 
372     return elem;
373 }
374 
375 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
376                                       int offset, MemoryDeviceInfoList *dimms)
377 {
378     MachineState *machine = MACHINE(spapr);
379     uint8_t *int_buf, *cur_index;
380     int ret;
381     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
382     uint64_t addr, cur_addr, size;
383     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
384     uint64_t mem_end = machine->device_memory->base +
385                        memory_region_size(&machine->device_memory->mr);
386     uint32_t node, buf_len, nr_entries = 0;
387     SpaprDrc *drc;
388     DrconfCellQueue *elem, *next;
389     MemoryDeviceInfoList *info;
390     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
391         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
392 
393     /* Entry to cover RAM and the gap area */
394     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
395                                  SPAPR_LMB_FLAGS_RESERVED |
396                                  SPAPR_LMB_FLAGS_DRC_INVALID);
397     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
398     nr_entries++;
399 
400     cur_addr = machine->device_memory->base;
401     for (info = dimms; info; info = info->next) {
402         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
403 
404         addr = di->addr;
405         size = di->size;
406         node = di->node;
407 
408         /*
409          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
410          * area is marked hotpluggable in the next iteration for the bigger
411          * chunk including the NVDIMM occupied area.
412          */
413         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
414             continue;
415 
416         /* Entry for hot-pluggable area */
417         if (cur_addr < addr) {
418             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
419             g_assert(drc);
420             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
421                                          cur_addr, spapr_drc_index(drc), -1, 0);
422             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
423             nr_entries++;
424         }
425 
426         /* Entry for DIMM */
427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
428         g_assert(drc);
429         elem = spapr_get_drconf_cell(size / lmb_size, addr,
430                                      spapr_drc_index(drc), node,
431                                      (SPAPR_LMB_FLAGS_ASSIGNED |
432                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
433         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
434         nr_entries++;
435         cur_addr = addr + size;
436     }
437 
438     /* Entry for remaining hotpluggable area */
439     if (cur_addr < mem_end) {
440         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
441         g_assert(drc);
442         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
443                                      cur_addr, spapr_drc_index(drc), -1, 0);
444         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
445         nr_entries++;
446     }
447 
448     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
449     int_buf = cur_index = g_malloc0(buf_len);
450     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
451     cur_index += sizeof(nr_entries);
452 
453     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
454         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
455         cur_index += sizeof(elem->cell);
456         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
457         g_free(elem);
458     }
459 
460     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
461     g_free(int_buf);
462     if (ret < 0) {
463         return -1;
464     }
465     return 0;
466 }
467 
468 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
469                                    int offset, MemoryDeviceInfoList *dimms)
470 {
471     MachineState *machine = MACHINE(spapr);
472     int i, ret;
473     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
474     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
475     uint32_t nr_lmbs = (machine->device_memory->base +
476                        memory_region_size(&machine->device_memory->mr)) /
477                        lmb_size;
478     uint32_t *int_buf, *cur_index, buf_len;
479 
480     /*
481      * Allocate enough buffer size to fit in ibm,dynamic-memory
482      */
483     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
484     cur_index = int_buf = g_malloc0(buf_len);
485     int_buf[0] = cpu_to_be32(nr_lmbs);
486     cur_index++;
487     for (i = 0; i < nr_lmbs; i++) {
488         uint64_t addr = i * lmb_size;
489         uint32_t *dynamic_memory = cur_index;
490 
491         if (i >= device_lmb_start) {
492             SpaprDrc *drc;
493 
494             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
495             g_assert(drc);
496 
497             dynamic_memory[0] = cpu_to_be32(addr >> 32);
498             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
499             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
500             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
501             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
502             if (memory_region_present(get_system_memory(), addr)) {
503                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
504             } else {
505                 dynamic_memory[5] = cpu_to_be32(0);
506             }
507         } else {
508             /*
509              * LMB information for RMA, boot time RAM and gap b/n RAM and
510              * device memory region -- all these are marked as reserved
511              * and as having no valid DRC.
512              */
513             dynamic_memory[0] = cpu_to_be32(addr >> 32);
514             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
515             dynamic_memory[2] = cpu_to_be32(0);
516             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
517             dynamic_memory[4] = cpu_to_be32(-1);
518             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
519                                             SPAPR_LMB_FLAGS_DRC_INVALID);
520         }
521 
522         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
523     }
524     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
525     g_free(int_buf);
526     if (ret < 0) {
527         return -1;
528     }
529     return 0;
530 }
531 
532 /*
533  * Adds ibm,dynamic-reconfiguration-memory node.
534  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
535  * of this device tree node.
536  */
537 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
538                                                    void *fdt)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int ret, offset;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
544                                 cpu_to_be32(lmb_size & 0xffffffff)};
545     MemoryDeviceInfoList *dimms = NULL;
546 
547     /*
548      * Don't create the node if there is no device memory
549      */
550     if (machine->ram_size == machine->maxram_size) {
551         return 0;
552     }
553 
554     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
555 
556     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557                     sizeof(prop_lmb_size));
558     if (ret < 0) {
559         return ret;
560     }
561 
562     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
563     if (ret < 0) {
564         return ret;
565     }
566 
567     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
568     if (ret < 0) {
569         return ret;
570     }
571 
572     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573     dimms = qmp_memory_device_list();
574     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
576     } else {
577         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
578     }
579     qapi_free_MemoryDeviceInfoList(dimms);
580 
581     if (ret < 0) {
582         return ret;
583     }
584 
585     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
586 
587     return ret;
588 }
589 
590 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
591 {
592     MachineState *machine = MACHINE(spapr);
593     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
594     hwaddr mem_start, node_size;
595     int i, nb_nodes = machine->numa_state->num_nodes;
596     NodeInfo *nodes = machine->numa_state->nodes;
597 
598     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
599         if (!nodes[i].node_mem) {
600             continue;
601         }
602         if (mem_start >= machine->ram_size) {
603             node_size = 0;
604         } else {
605             node_size = nodes[i].node_mem;
606             if (node_size > machine->ram_size - mem_start) {
607                 node_size = machine->ram_size - mem_start;
608             }
609         }
610         if (!mem_start) {
611             /* spapr_machine_init() checks for rma_size <= node0_size
612              * already */
613             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
614             mem_start += spapr->rma_size;
615             node_size -= spapr->rma_size;
616         }
617         for ( ; node_size; ) {
618             hwaddr sizetmp = pow2floor(node_size);
619 
620             /* mem_start != 0 here */
621             if (ctzl(mem_start) < ctzl(sizetmp)) {
622                 sizetmp = 1ULL << ctzl(mem_start);
623             }
624 
625             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
626             node_size -= sizetmp;
627             mem_start += sizetmp;
628         }
629     }
630 
631     /* Generate ibm,dynamic-reconfiguration-memory node if required */
632     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
633         int ret;
634 
635         g_assert(smc->dr_lmb_enabled);
636         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
637         if (ret) {
638             return ret;
639         }
640     }
641 
642     return 0;
643 }
644 
645 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
646                          SpaprMachineState *spapr)
647 {
648     MachineState *ms = MACHINE(spapr);
649     PowerPCCPU *cpu = POWERPC_CPU(cs);
650     CPUPPCState *env = &cpu->env;
651     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
652     int index = spapr_get_vcpu_id(cpu);
653     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
654                        0xffffffff, 0xffffffff};
655     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
656         : SPAPR_TIMEBASE_FREQ;
657     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
658     uint32_t page_sizes_prop[64];
659     size_t page_sizes_prop_size;
660     unsigned int smp_threads = ms->smp.threads;
661     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
662     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
663     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
664     SpaprDrc *drc;
665     int drc_index;
666     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
667     int i;
668 
669     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
670     if (drc) {
671         drc_index = spapr_drc_index(drc);
672         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
673     }
674 
675     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
676     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
677 
678     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
682                            env->dcache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
684                            env->icache_line_size)));
685     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
686                            env->icache_line_size)));
687 
688     if (pcc->l1_dcache_size) {
689         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
690                                pcc->l1_dcache_size)));
691     } else {
692         warn_report("Unknown L1 dcache size for cpu");
693     }
694     if (pcc->l1_icache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
696                                pcc->l1_icache_size)));
697     } else {
698         warn_report("Unknown L1 icache size for cpu");
699     }
700 
701     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
702     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
703     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
704     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
705     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
706     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
707 
708     if (ppc_has_spr(cpu, SPR_PURR)) {
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
710     }
711     if (ppc_has_spr(cpu, SPR_PURR)) {
712         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
713     }
714 
715     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
716         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
717                           segs, sizeof(segs))));
718     }
719 
720     /* Advertise VSX (vector extensions) if available
721      *   1               == VMX / Altivec available
722      *   2               == VSX available
723      *
724      * Only CPUs for which we create core types in spapr_cpu_core.c
725      * are possible, and all of those have VMX */
726     if (env->insns_flags & PPC_ALTIVEC) {
727         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
728             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
729         } else {
730             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
731         }
732     }
733 
734     /* Advertise DFP (Decimal Floating Point) if available
735      *   0 / no property == no DFP
736      *   1               == DFP available */
737     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
738         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
739     }
740 
741     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
742                                                       sizeof(page_sizes_prop));
743     if (page_sizes_prop_size) {
744         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
745                           page_sizes_prop, page_sizes_prop_size)));
746     }
747 
748     spapr_dt_pa_features(spapr, cpu, fdt, offset);
749 
750     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
751                            cs->cpu_index / vcpus_per_socket)));
752 
753     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
754                       pft_size_prop, sizeof(pft_size_prop))));
755 
756     if (ms->numa_state->num_nodes > 1) {
757         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
758     }
759 
760     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
761 
762     if (pcc->radix_page_info) {
763         for (i = 0; i < pcc->radix_page_info->count; i++) {
764             radix_AP_encodings[i] =
765                 cpu_to_be32(pcc->radix_page_info->entries[i]);
766         }
767         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
768                           radix_AP_encodings,
769                           pcc->radix_page_info->count *
770                           sizeof(radix_AP_encodings[0]))));
771     }
772 
773     /*
774      * We set this property to let the guest know that it can use the large
775      * decrementer and its width in bits.
776      */
777     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
778         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
779                               pcc->lrg_decr_bits)));
780 }
781 
782 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
783 {
784     CPUState **rev;
785     CPUState *cs;
786     int n_cpus;
787     int cpus_offset;
788     int i;
789 
790     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
791     _FDT(cpus_offset);
792     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
793     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
794 
795     /*
796      * We walk the CPUs in reverse order to ensure that CPU DT nodes
797      * created by fdt_add_subnode() end up in the right order in FDT
798      * for the guest kernel the enumerate the CPUs correctly.
799      *
800      * The CPU list cannot be traversed in reverse order, so we need
801      * to do extra work.
802      */
803     n_cpus = 0;
804     rev = NULL;
805     CPU_FOREACH(cs) {
806         rev = g_renew(CPUState *, rev, n_cpus + 1);
807         rev[n_cpus++] = cs;
808     }
809 
810     for (i = n_cpus - 1; i >= 0; i--) {
811         CPUState *cs = rev[i];
812         PowerPCCPU *cpu = POWERPC_CPU(cs);
813         int index = spapr_get_vcpu_id(cpu);
814         DeviceClass *dc = DEVICE_GET_CLASS(cs);
815         g_autofree char *nodename = NULL;
816         int offset;
817 
818         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
819             continue;
820         }
821 
822         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
823         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
824         _FDT(offset);
825         spapr_dt_cpu(cs, fdt, offset, spapr);
826     }
827 
828     g_free(rev);
829 }
830 
831 static int spapr_dt_rng(void *fdt)
832 {
833     int node;
834     int ret;
835 
836     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
837     if (node <= 0) {
838         return -1;
839     }
840     ret = fdt_setprop_string(fdt, node, "device_type",
841                              "ibm,platform-facilities");
842     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
843     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
844 
845     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
846     if (node <= 0) {
847         return -1;
848     }
849     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
850 
851     return ret ? -1 : 0;
852 }
853 
854 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
855 {
856     MachineState *ms = MACHINE(spapr);
857     int rtas;
858     GString *hypertas = g_string_sized_new(256);
859     GString *qemu_hypertas = g_string_sized_new(256);
860     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
861         memory_region_size(&MACHINE(spapr)->device_memory->mr);
862     uint32_t lrdr_capacity[] = {
863         cpu_to_be32(max_device_addr >> 32),
864         cpu_to_be32(max_device_addr & 0xffffffff),
865         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
866         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
867         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
868     };
869 
870     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
871 
872     /* hypertas */
873     add_str(hypertas, "hcall-pft");
874     add_str(hypertas, "hcall-term");
875     add_str(hypertas, "hcall-dabr");
876     add_str(hypertas, "hcall-interrupt");
877     add_str(hypertas, "hcall-tce");
878     add_str(hypertas, "hcall-vio");
879     add_str(hypertas, "hcall-splpar");
880     add_str(hypertas, "hcall-join");
881     add_str(hypertas, "hcall-bulk");
882     add_str(hypertas, "hcall-set-mode");
883     add_str(hypertas, "hcall-sprg0");
884     add_str(hypertas, "hcall-copy");
885     add_str(hypertas, "hcall-debug");
886     add_str(hypertas, "hcall-vphn");
887     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
888         add_str(hypertas, "hcall-rpt-invalidate");
889     }
890 
891     add_str(qemu_hypertas, "hcall-memop1");
892 
893     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
894         add_str(hypertas, "hcall-multi-tce");
895     }
896 
897     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
898         add_str(hypertas, "hcall-hpt-resize");
899     }
900 
901     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
902                      hypertas->str, hypertas->len));
903     g_string_free(hypertas, TRUE);
904     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
905                      qemu_hypertas->str, qemu_hypertas->len));
906     g_string_free(qemu_hypertas, TRUE);
907 
908     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
909 
910     /*
911      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
912      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
913      *
914      * The system reset requirements are driven by existing Linux and PowerVM
915      * implementation which (contrary to PAPR) saves r3 in the error log
916      * structure like machine check, so Linux expects to find the saved r3
917      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
918      * does not look at the error value).
919      *
920      * System reset interrupts are not subject to interlock like machine
921      * check, so this memory area could be corrupted if the sreset is
922      * interrupted by a machine check (or vice versa) if it was shared. To
923      * prevent this, system reset uses per-CPU areas for the sreset save
924      * area. A system reset that interrupts a system reset handler could
925      * still overwrite this area, but Linux doesn't try to recover in that
926      * case anyway.
927      *
928      * The extra 8 bytes is required because Linux's FWNMI error log check
929      * is off-by-one.
930      *
931      * RTAS_MIN_SIZE is required for the RTAS blob itself.
932      */
933     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
934                           RTAS_ERROR_LOG_MAX +
935                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
936                           sizeof(uint64_t)));
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
938                           RTAS_ERROR_LOG_MAX));
939     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
940                           RTAS_EVENT_SCAN_RATE));
941 
942     g_assert(msi_nonbroken);
943     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
944 
945     /*
946      * According to PAPR, rtas ibm,os-term does not guarantee a return
947      * back to the guest cpu.
948      *
949      * While an additional ibm,extended-os-term property indicates
950      * that rtas call return will always occur. Set this property.
951      */
952     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
953 
954     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
955                      lrdr_capacity, sizeof(lrdr_capacity)));
956 
957     spapr_dt_rtas_tokens(fdt, rtas);
958 }
959 
960 /*
961  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
962  * and the XIVE features that the guest may request and thus the valid
963  * values for bytes 23..26 of option vector 5:
964  */
965 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
966                                           int chosen)
967 {
968     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
969 
970     char val[2 * 4] = {
971         23, 0x00, /* XICS / XIVE mode */
972         24, 0x00, /* Hash/Radix, filled in below. */
973         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
974         26, 0x40, /* Radix options: GTSE == yes. */
975     };
976 
977     if (spapr->irq->xics && spapr->irq->xive) {
978         val[1] = SPAPR_OV5_XIVE_BOTH;
979     } else if (spapr->irq->xive) {
980         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
981     } else {
982         assert(spapr->irq->xics);
983         val[1] = SPAPR_OV5_XIVE_LEGACY;
984     }
985 
986     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
987                           first_ppc_cpu->compat_pvr)) {
988         /*
989          * If we're in a pre POWER9 compat mode then the guest should
990          * do hash and use the legacy interrupt mode
991          */
992         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
993         val[3] = 0x00; /* Hash */
994         spapr_check_mmu_mode(false);
995     } else if (kvm_enabled()) {
996         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
997             val[3] = 0x80; /* OV5_MMU_BOTH */
998         } else if (kvmppc_has_cap_mmu_radix()) {
999             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1000         } else {
1001             val[3] = 0x00; /* Hash */
1002         }
1003     } else {
1004         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1005         val[3] = 0xC0;
1006     }
1007     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1008                      val, sizeof(val)));
1009 }
1010 
1011 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1012 {
1013     MachineState *machine = MACHINE(spapr);
1014     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1015     int chosen;
1016 
1017     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1018 
1019     if (reset) {
1020         const char *boot_device = spapr->boot_device;
1021         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1022         size_t cb = 0;
1023         g_autofree char *bootlist = get_boot_devices_list(&cb);
1024 
1025         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1026             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1027                                     machine->kernel_cmdline));
1028         }
1029 
1030         if (spapr->initrd_size) {
1031             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1032                                   spapr->initrd_base));
1033             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1034                                   spapr->initrd_base + spapr->initrd_size));
1035         }
1036 
1037         if (spapr->kernel_size) {
1038             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1039                                   cpu_to_be64(spapr->kernel_size) };
1040 
1041             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1042                          &kprop, sizeof(kprop)));
1043             if (spapr->kernel_le) {
1044                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1045             }
1046         }
1047         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1048             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1049         }
1050         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1051         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1052         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1053 
1054         if (cb && bootlist) {
1055             int i;
1056 
1057             for (i = 0; i < cb; i++) {
1058                 if (bootlist[i] == '\n') {
1059                     bootlist[i] = ' ';
1060                 }
1061             }
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1063         }
1064 
1065         if (boot_device && strlen(boot_device)) {
1066             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1067         }
1068 
1069         if (!spapr->has_graphics && stdout_path) {
1070             /*
1071              * "linux,stdout-path" and "stdout" properties are
1072              * deprecated by linux kernel. New platforms should only
1073              * use the "stdout-path" property. Set the new property
1074              * and continue using older property to remain compatible
1075              * with the existing firmware.
1076              */
1077             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1078             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1079         }
1080 
1081         /*
1082          * We can deal with BAR reallocation just fine, advertise it
1083          * to the guest
1084          */
1085         if (smc->linux_pci_probe) {
1086             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1087         }
1088 
1089         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1090     }
1091 
1092     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1093 }
1094 
1095 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1096 {
1097     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1098      * KVM to work under pHyp with some guest co-operation */
1099     int hypervisor;
1100     uint8_t hypercall[16];
1101 
1102     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1103     /* indicate KVM hypercall interface */
1104     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1105     if (kvmppc_has_cap_fixup_hcalls()) {
1106         /*
1107          * Older KVM versions with older guest kernels were broken
1108          * with the magic page, don't allow the guest to map it.
1109          */
1110         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1111                                   sizeof(hypercall))) {
1112             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1113                              hypercall, sizeof(hypercall)));
1114         }
1115     }
1116 }
1117 
1118 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1119 {
1120     MachineState *machine = MACHINE(spapr);
1121     MachineClass *mc = MACHINE_GET_CLASS(machine);
1122     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1123     uint32_t root_drc_type_mask = 0;
1124     int ret;
1125     void *fdt;
1126     SpaprPhbState *phb;
1127     char *buf;
1128 
1129     fdt = g_malloc0(space);
1130     _FDT((fdt_create_empty_tree(fdt, space)));
1131 
1132     /* Root node */
1133     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1134     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1135     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1136 
1137     /* Guest UUID & Name*/
1138     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1139     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1140     if (qemu_uuid_set) {
1141         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1142     }
1143     g_free(buf);
1144 
1145     if (qemu_get_vm_name()) {
1146         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1147                                 qemu_get_vm_name()));
1148     }
1149 
1150     /* Host Model & Serial Number */
1151     if (spapr->host_model) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1153     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1154         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1155         g_free(buf);
1156     }
1157 
1158     if (spapr->host_serial) {
1159         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1160     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1161         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1162         g_free(buf);
1163     }
1164 
1165     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1166     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1167 
1168     /* /interrupt controller */
1169     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1170 
1171     ret = spapr_dt_memory(spapr, fdt);
1172     if (ret < 0) {
1173         error_report("couldn't setup memory nodes in fdt");
1174         exit(1);
1175     }
1176 
1177     /* /vdevice */
1178     spapr_dt_vdevice(spapr->vio_bus, fdt);
1179 
1180     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1181         ret = spapr_dt_rng(fdt);
1182         if (ret < 0) {
1183             error_report("could not set up rng device in the fdt");
1184             exit(1);
1185         }
1186     }
1187 
1188     QLIST_FOREACH(phb, &spapr->phbs, list) {
1189         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1190         if (ret < 0) {
1191             error_report("couldn't setup PCI devices in fdt");
1192             exit(1);
1193         }
1194     }
1195 
1196     spapr_dt_cpus(fdt, spapr);
1197 
1198     /* ibm,drc-indexes and friends */
1199     if (smc->dr_lmb_enabled) {
1200         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1201     }
1202     if (smc->dr_phb_enabled) {
1203         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1204     }
1205     if (mc->nvdimm_supported) {
1206         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1207     }
1208     if (root_drc_type_mask) {
1209         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1210     }
1211 
1212     if (mc->has_hotpluggable_cpus) {
1213         int offset = fdt_path_offset(fdt, "/cpus");
1214         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1215         if (ret < 0) {
1216             error_report("Couldn't set up CPU DR device tree properties");
1217             exit(1);
1218         }
1219     }
1220 
1221     /* /event-sources */
1222     spapr_dt_events(spapr, fdt);
1223 
1224     /* /rtas */
1225     spapr_dt_rtas(spapr, fdt);
1226 
1227     /* /chosen */
1228     spapr_dt_chosen(spapr, fdt, reset);
1229 
1230     /* /hypervisor */
1231     if (kvm_enabled()) {
1232         spapr_dt_hypervisor(spapr, fdt);
1233     }
1234 
1235     /* Build memory reserve map */
1236     if (reset) {
1237         if (spapr->kernel_size) {
1238             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1239                                   spapr->kernel_size)));
1240         }
1241         if (spapr->initrd_size) {
1242             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1243                                   spapr->initrd_size)));
1244         }
1245     }
1246 
1247     /* NVDIMM devices */
1248     if (mc->nvdimm_supported) {
1249         spapr_dt_persistent_memory(spapr, fdt);
1250     }
1251 
1252     return fdt;
1253 }
1254 
1255 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1256 {
1257     SpaprMachineState *spapr = opaque;
1258 
1259     return (addr & 0x0fffffff) + spapr->kernel_addr;
1260 }
1261 
1262 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1263                                     PowerPCCPU *cpu)
1264 {
1265     CPUPPCState *env = &cpu->env;
1266 
1267     /* The TCG path should also be holding the BQL at this point */
1268     g_assert(qemu_mutex_iothread_locked());
1269 
1270     g_assert(!vhyp_cpu_in_nested(cpu));
1271 
1272     if (FIELD_EX64(env->msr, MSR, PR)) {
1273         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1274         env->gpr[3] = H_PRIVILEGE;
1275     } else {
1276         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1277     }
1278 }
1279 
1280 struct LPCRSyncState {
1281     target_ulong value;
1282     target_ulong mask;
1283 };
1284 
1285 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1286 {
1287     struct LPCRSyncState *s = arg.host_ptr;
1288     PowerPCCPU *cpu = POWERPC_CPU(cs);
1289     CPUPPCState *env = &cpu->env;
1290     target_ulong lpcr;
1291 
1292     cpu_synchronize_state(cs);
1293     lpcr = env->spr[SPR_LPCR];
1294     lpcr &= ~s->mask;
1295     lpcr |= s->value;
1296     ppc_store_lpcr(cpu, lpcr);
1297 }
1298 
1299 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1300 {
1301     CPUState *cs;
1302     struct LPCRSyncState s = {
1303         .value = value,
1304         .mask = mask
1305     };
1306     CPU_FOREACH(cs) {
1307         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1308     }
1309 }
1310 
1311 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1312                            target_ulong lpid, ppc_v3_pate_t *entry)
1313 {
1314     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1315     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1316 
1317     if (!spapr_cpu->in_nested) {
1318         assert(lpid == 0);
1319 
1320         /* Copy PATE1:GR into PATE0:HR */
1321         entry->dw0 = spapr->patb_entry & PATE0_HR;
1322         entry->dw1 = spapr->patb_entry;
1323 
1324     } else {
1325         uint64_t patb, pats;
1326 
1327         assert(lpid != 0);
1328 
1329         patb = spapr->nested_ptcr & PTCR_PATB;
1330         pats = spapr->nested_ptcr & PTCR_PATS;
1331 
1332         /* Calculate number of entries */
1333         pats = 1ull << (pats + 12 - 4);
1334         if (pats <= lpid) {
1335             return false;
1336         }
1337 
1338         /* Grab entry */
1339         patb += 16 * lpid;
1340         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1341         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1342     }
1343 
1344     return true;
1345 }
1346 
1347 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1348 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1349 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1350 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1351 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1352 
1353 /*
1354  * Get the fd to access the kernel htab, re-opening it if necessary
1355  */
1356 static int get_htab_fd(SpaprMachineState *spapr)
1357 {
1358     Error *local_err = NULL;
1359 
1360     if (spapr->htab_fd >= 0) {
1361         return spapr->htab_fd;
1362     }
1363 
1364     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1365     if (spapr->htab_fd < 0) {
1366         error_report_err(local_err);
1367     }
1368 
1369     return spapr->htab_fd;
1370 }
1371 
1372 void close_htab_fd(SpaprMachineState *spapr)
1373 {
1374     if (spapr->htab_fd >= 0) {
1375         close(spapr->htab_fd);
1376     }
1377     spapr->htab_fd = -1;
1378 }
1379 
1380 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1381 {
1382     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1383 
1384     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1385 }
1386 
1387 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390 
1391     assert(kvm_enabled());
1392 
1393     if (!spapr->htab) {
1394         return 0;
1395     }
1396 
1397     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1398 }
1399 
1400 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1401                                                 hwaddr ptex, int n)
1402 {
1403     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1404     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1405 
1406     if (!spapr->htab) {
1407         /*
1408          * HTAB is controlled by KVM. Fetch into temporary buffer
1409          */
1410         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1411         kvmppc_read_hptes(hptes, ptex, n);
1412         return hptes;
1413     }
1414 
1415     /*
1416      * HTAB is controlled by QEMU. Just point to the internally
1417      * accessible PTEG.
1418      */
1419     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1420 }
1421 
1422 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1423                               const ppc_hash_pte64_t *hptes,
1424                               hwaddr ptex, int n)
1425 {
1426     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1427 
1428     if (!spapr->htab) {
1429         g_free((void *)hptes);
1430     }
1431 
1432     /* Nothing to do for qemu managed HPT */
1433 }
1434 
1435 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1436                       uint64_t pte0, uint64_t pte1)
1437 {
1438     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1439     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1440 
1441     if (!spapr->htab) {
1442         kvmppc_write_hpte(ptex, pte0, pte1);
1443     } else {
1444         if (pte0 & HPTE64_V_VALID) {
1445             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1446             /*
1447              * When setting valid, we write PTE1 first. This ensures
1448              * proper synchronization with the reading code in
1449              * ppc_hash64_pteg_search()
1450              */
1451             smp_wmb();
1452             stq_p(spapr->htab + offset, pte0);
1453         } else {
1454             stq_p(spapr->htab + offset, pte0);
1455             /*
1456              * When clearing it we set PTE0 first. This ensures proper
1457              * synchronization with the reading code in
1458              * ppc_hash64_pteg_search()
1459              */
1460             smp_wmb();
1461             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1462         }
1463     }
1464 }
1465 
1466 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1467                              uint64_t pte1)
1468 {
1469     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1470     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1471 
1472     if (!spapr->htab) {
1473         /* There should always be a hash table when this is called */
1474         error_report("spapr_hpte_set_c called with no hash table !");
1475         return;
1476     }
1477 
1478     /* The HW performs a non-atomic byte update */
1479     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1480 }
1481 
1482 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1483                              uint64_t pte1)
1484 {
1485     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1486     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1487 
1488     if (!spapr->htab) {
1489         /* There should always be a hash table when this is called */
1490         error_report("spapr_hpte_set_r called with no hash table !");
1491         return;
1492     }
1493 
1494     /* The HW performs a non-atomic byte update */
1495     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1496 }
1497 
1498 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1499 {
1500     int shift;
1501 
1502     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1503      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1504      * that's much more than is needed for Linux guests */
1505     shift = ctz64(pow2ceil(ramsize)) - 7;
1506     shift = MAX(shift, 18); /* Minimum architected size */
1507     shift = MIN(shift, 46); /* Maximum architected size */
1508     return shift;
1509 }
1510 
1511 void spapr_free_hpt(SpaprMachineState *spapr)
1512 {
1513     g_free(spapr->htab);
1514     spapr->htab = NULL;
1515     spapr->htab_shift = 0;
1516     close_htab_fd(spapr);
1517 }
1518 
1519 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1520 {
1521     ERRP_GUARD();
1522     long rc;
1523 
1524     /* Clean up any HPT info from a previous boot */
1525     spapr_free_hpt(spapr);
1526 
1527     rc = kvmppc_reset_htab(shift);
1528 
1529     if (rc == -EOPNOTSUPP) {
1530         error_setg(errp, "HPT not supported in nested guests");
1531         return -EOPNOTSUPP;
1532     }
1533 
1534     if (rc < 0) {
1535         /* kernel-side HPT needed, but couldn't allocate one */
1536         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1537                          shift);
1538         error_append_hint(errp, "Try smaller maxmem?\n");
1539         return -errno;
1540     } else if (rc > 0) {
1541         /* kernel-side HPT allocated */
1542         if (rc != shift) {
1543             error_setg(errp,
1544                        "Requested order %d HPT, but kernel allocated order %ld",
1545                        shift, rc);
1546             error_append_hint(errp, "Try smaller maxmem?\n");
1547             return -ENOSPC;
1548         }
1549 
1550         spapr->htab_shift = shift;
1551         spapr->htab = NULL;
1552     } else {
1553         /* kernel-side HPT not needed, allocate in userspace instead */
1554         size_t size = 1ULL << shift;
1555         int i;
1556 
1557         spapr->htab = qemu_memalign(size, size);
1558         memset(spapr->htab, 0, size);
1559         spapr->htab_shift = shift;
1560 
1561         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1562             DIRTY_HPTE(HPTE(spapr->htab, i));
1563         }
1564     }
1565     /* We're setting up a hash table, so that means we're not radix */
1566     spapr->patb_entry = 0;
1567     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1568     return 0;
1569 }
1570 
1571 void spapr_setup_hpt(SpaprMachineState *spapr)
1572 {
1573     int hpt_shift;
1574 
1575     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1576         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1577     } else {
1578         uint64_t current_ram_size;
1579 
1580         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1581         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1582     }
1583     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1584 
1585     if (kvm_enabled()) {
1586         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1587 
1588         /* Check our RMA fits in the possible VRMA */
1589         if (vrma_limit < spapr->rma_size) {
1590             error_report("Unable to create %" HWADDR_PRIu
1591                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1592                          spapr->rma_size / MiB, vrma_limit / MiB);
1593             exit(EXIT_FAILURE);
1594         }
1595     }
1596 }
1597 
1598 void spapr_check_mmu_mode(bool guest_radix)
1599 {
1600     if (guest_radix) {
1601         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1602             error_report("Guest requested unavailable MMU mode (radix).");
1603             exit(EXIT_FAILURE);
1604         }
1605     } else {
1606         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1607             && !kvmppc_has_cap_mmu_hash_v3()) {
1608             error_report("Guest requested unavailable MMU mode (hash).");
1609             exit(EXIT_FAILURE);
1610         }
1611     }
1612 }
1613 
1614 static void spapr_machine_reset(MachineState *machine)
1615 {
1616     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1617     PowerPCCPU *first_ppc_cpu;
1618     hwaddr fdt_addr;
1619     void *fdt;
1620     int rc;
1621 
1622     pef_kvm_reset(machine->cgs, &error_fatal);
1623     spapr_caps_apply(spapr);
1624 
1625     first_ppc_cpu = POWERPC_CPU(first_cpu);
1626     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1627         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1628                               spapr->max_compat_pvr)) {
1629         /*
1630          * If using KVM with radix mode available, VCPUs can be started
1631          * without a HPT because KVM will start them in radix mode.
1632          * Set the GR bit in PATE so that we know there is no HPT.
1633          */
1634         spapr->patb_entry = PATE1_GR;
1635         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1636     } else {
1637         spapr_setup_hpt(spapr);
1638     }
1639 
1640     qemu_devices_reset();
1641 
1642     spapr_ovec_cleanup(spapr->ov5_cas);
1643     spapr->ov5_cas = spapr_ovec_new();
1644 
1645     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1646 
1647     /*
1648      * This is fixing some of the default configuration of the XIVE
1649      * devices. To be called after the reset of the machine devices.
1650      */
1651     spapr_irq_reset(spapr, &error_fatal);
1652 
1653     /*
1654      * There is no CAS under qtest. Simulate one to please the code that
1655      * depends on spapr->ov5_cas. This is especially needed to test device
1656      * unplug, so we do that before resetting the DRCs.
1657      */
1658     if (qtest_enabled()) {
1659         spapr_ovec_cleanup(spapr->ov5_cas);
1660         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1661     }
1662 
1663     spapr_nvdimm_finish_flushes();
1664 
1665     /* DRC reset may cause a device to be unplugged. This will cause troubles
1666      * if this device is used by another device (eg, a running vhost backend
1667      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1668      * situations, we reset DRCs after all devices have been reset.
1669      */
1670     spapr_drc_reset_all(spapr);
1671 
1672     spapr_clear_pending_events(spapr);
1673 
1674     /*
1675      * We place the device tree just below either the top of the RMA,
1676      * or just below 2GB, whichever is lower, so that it can be
1677      * processed with 32-bit real mode code if necessary
1678      */
1679     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1680 
1681     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1682     if (spapr->vof) {
1683         spapr_vof_reset(spapr, fdt, &error_fatal);
1684         /*
1685          * Do not pack the FDT as the client may change properties.
1686          * VOF client does not expect the FDT so we do not load it to the VM.
1687          */
1688     } else {
1689         rc = fdt_pack(fdt);
1690         /* Should only fail if we've built a corrupted tree */
1691         assert(rc == 0);
1692 
1693         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1694                                   0, fdt_addr, 0);
1695         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1696     }
1697     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1698 
1699     g_free(spapr->fdt_blob);
1700     spapr->fdt_size = fdt_totalsize(fdt);
1701     spapr->fdt_initial_size = spapr->fdt_size;
1702     spapr->fdt_blob = fdt;
1703 
1704     /* Set up the entry state */
1705     first_ppc_cpu->env.gpr[5] = 0;
1706 
1707     spapr->fwnmi_system_reset_addr = -1;
1708     spapr->fwnmi_machine_check_addr = -1;
1709     spapr->fwnmi_machine_check_interlock = -1;
1710 
1711     /* Signal all vCPUs waiting on this condition */
1712     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1713 
1714     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1715 }
1716 
1717 static void spapr_create_nvram(SpaprMachineState *spapr)
1718 {
1719     DeviceState *dev = qdev_new("spapr-nvram");
1720     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1721 
1722     if (dinfo) {
1723         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1724                                 &error_fatal);
1725     }
1726 
1727     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1728 
1729     spapr->nvram = (struct SpaprNvram *)dev;
1730 }
1731 
1732 static void spapr_rtc_create(SpaprMachineState *spapr)
1733 {
1734     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1735                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1736                                        &error_fatal, NULL);
1737     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1738     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1739                               "date");
1740 }
1741 
1742 /* Returns whether we want to use VGA or not */
1743 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1744 {
1745     vga_interface_created = true;
1746     switch (vga_interface_type) {
1747     case VGA_NONE:
1748         return false;
1749     case VGA_DEVICE:
1750         return true;
1751     case VGA_STD:
1752     case VGA_VIRTIO:
1753     case VGA_CIRRUS:
1754         return pci_vga_init(pci_bus) != NULL;
1755     default:
1756         error_setg(errp,
1757                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1758         return false;
1759     }
1760 }
1761 
1762 static int spapr_pre_load(void *opaque)
1763 {
1764     int rc;
1765 
1766     rc = spapr_caps_pre_load(opaque);
1767     if (rc) {
1768         return rc;
1769     }
1770 
1771     return 0;
1772 }
1773 
1774 static int spapr_post_load(void *opaque, int version_id)
1775 {
1776     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1777     int err = 0;
1778 
1779     err = spapr_caps_post_migration(spapr);
1780     if (err) {
1781         return err;
1782     }
1783 
1784     /*
1785      * In earlier versions, there was no separate qdev for the PAPR
1786      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1787      * So when migrating from those versions, poke the incoming offset
1788      * value into the RTC device
1789      */
1790     if (version_id < 3) {
1791         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1792         if (err) {
1793             return err;
1794         }
1795     }
1796 
1797     if (kvm_enabled() && spapr->patb_entry) {
1798         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1799         bool radix = !!(spapr->patb_entry & PATE1_GR);
1800         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1801 
1802         /*
1803          * Update LPCR:HR and UPRT as they may not be set properly in
1804          * the stream
1805          */
1806         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1807                             LPCR_HR | LPCR_UPRT);
1808 
1809         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1810         if (err) {
1811             error_report("Process table config unsupported by the host");
1812             return -EINVAL;
1813         }
1814     }
1815 
1816     err = spapr_irq_post_load(spapr, version_id);
1817     if (err) {
1818         return err;
1819     }
1820 
1821     return err;
1822 }
1823 
1824 static int spapr_pre_save(void *opaque)
1825 {
1826     int rc;
1827 
1828     rc = spapr_caps_pre_save(opaque);
1829     if (rc) {
1830         return rc;
1831     }
1832 
1833     return 0;
1834 }
1835 
1836 static bool version_before_3(void *opaque, int version_id)
1837 {
1838     return version_id < 3;
1839 }
1840 
1841 static bool spapr_pending_events_needed(void *opaque)
1842 {
1843     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1844     return !QTAILQ_EMPTY(&spapr->pending_events);
1845 }
1846 
1847 static const VMStateDescription vmstate_spapr_event_entry = {
1848     .name = "spapr_event_log_entry",
1849     .version_id = 1,
1850     .minimum_version_id = 1,
1851     .fields = (VMStateField[]) {
1852         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1853         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1854         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1855                                      NULL, extended_length),
1856         VMSTATE_END_OF_LIST()
1857     },
1858 };
1859 
1860 static const VMStateDescription vmstate_spapr_pending_events = {
1861     .name = "spapr_pending_events",
1862     .version_id = 1,
1863     .minimum_version_id = 1,
1864     .needed = spapr_pending_events_needed,
1865     .fields = (VMStateField[]) {
1866         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1867                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1868         VMSTATE_END_OF_LIST()
1869     },
1870 };
1871 
1872 static bool spapr_ov5_cas_needed(void *opaque)
1873 {
1874     SpaprMachineState *spapr = opaque;
1875     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1876     bool cas_needed;
1877 
1878     /* Prior to the introduction of SpaprOptionVector, we had two option
1879      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1880      * Both of these options encode machine topology into the device-tree
1881      * in such a way that the now-booted OS should still be able to interact
1882      * appropriately with QEMU regardless of what options were actually
1883      * negotiatied on the source side.
1884      *
1885      * As such, we can avoid migrating the CAS-negotiated options if these
1886      * are the only options available on the current machine/platform.
1887      * Since these are the only options available for pseries-2.7 and
1888      * earlier, this allows us to maintain old->new/new->old migration
1889      * compatibility.
1890      *
1891      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1892      * via default pseries-2.8 machines and explicit command-line parameters.
1893      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1894      * of the actual CAS-negotiated values to continue working properly. For
1895      * example, availability of memory unplug depends on knowing whether
1896      * OV5_HP_EVT was negotiated via CAS.
1897      *
1898      * Thus, for any cases where the set of available CAS-negotiatable
1899      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1900      * include the CAS-negotiated options in the migration stream, unless
1901      * if they affect boot time behaviour only.
1902      */
1903     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1904     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1905     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1906 
1907     /* We need extra information if we have any bits outside the mask
1908      * defined above */
1909     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1910 
1911     spapr_ovec_cleanup(ov5_mask);
1912 
1913     return cas_needed;
1914 }
1915 
1916 static const VMStateDescription vmstate_spapr_ov5_cas = {
1917     .name = "spapr_option_vector_ov5_cas",
1918     .version_id = 1,
1919     .minimum_version_id = 1,
1920     .needed = spapr_ov5_cas_needed,
1921     .fields = (VMStateField[]) {
1922         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1923                                  vmstate_spapr_ovec, SpaprOptionVector),
1924         VMSTATE_END_OF_LIST()
1925     },
1926 };
1927 
1928 static bool spapr_patb_entry_needed(void *opaque)
1929 {
1930     SpaprMachineState *spapr = opaque;
1931 
1932     return !!spapr->patb_entry;
1933 }
1934 
1935 static const VMStateDescription vmstate_spapr_patb_entry = {
1936     .name = "spapr_patb_entry",
1937     .version_id = 1,
1938     .minimum_version_id = 1,
1939     .needed = spapr_patb_entry_needed,
1940     .fields = (VMStateField[]) {
1941         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1942         VMSTATE_END_OF_LIST()
1943     },
1944 };
1945 
1946 static bool spapr_irq_map_needed(void *opaque)
1947 {
1948     SpaprMachineState *spapr = opaque;
1949 
1950     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1951 }
1952 
1953 static const VMStateDescription vmstate_spapr_irq_map = {
1954     .name = "spapr_irq_map",
1955     .version_id = 1,
1956     .minimum_version_id = 1,
1957     .needed = spapr_irq_map_needed,
1958     .fields = (VMStateField[]) {
1959         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1960         VMSTATE_END_OF_LIST()
1961     },
1962 };
1963 
1964 static bool spapr_dtb_needed(void *opaque)
1965 {
1966     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1967 
1968     return smc->update_dt_enabled;
1969 }
1970 
1971 static int spapr_dtb_pre_load(void *opaque)
1972 {
1973     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1974 
1975     g_free(spapr->fdt_blob);
1976     spapr->fdt_blob = NULL;
1977     spapr->fdt_size = 0;
1978 
1979     return 0;
1980 }
1981 
1982 static const VMStateDescription vmstate_spapr_dtb = {
1983     .name = "spapr_dtb",
1984     .version_id = 1,
1985     .minimum_version_id = 1,
1986     .needed = spapr_dtb_needed,
1987     .pre_load = spapr_dtb_pre_load,
1988     .fields = (VMStateField[]) {
1989         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1990         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1991         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1992                                      fdt_size),
1993         VMSTATE_END_OF_LIST()
1994     },
1995 };
1996 
1997 static bool spapr_fwnmi_needed(void *opaque)
1998 {
1999     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2000 
2001     return spapr->fwnmi_machine_check_addr != -1;
2002 }
2003 
2004 static int spapr_fwnmi_pre_save(void *opaque)
2005 {
2006     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2007 
2008     /*
2009      * Check if machine check handling is in progress and print a
2010      * warning message.
2011      */
2012     if (spapr->fwnmi_machine_check_interlock != -1) {
2013         warn_report("A machine check is being handled during migration. The"
2014                 "handler may run and log hardware error on the destination");
2015     }
2016 
2017     return 0;
2018 }
2019 
2020 static const VMStateDescription vmstate_spapr_fwnmi = {
2021     .name = "spapr_fwnmi",
2022     .version_id = 1,
2023     .minimum_version_id = 1,
2024     .needed = spapr_fwnmi_needed,
2025     .pre_save = spapr_fwnmi_pre_save,
2026     .fields = (VMStateField[]) {
2027         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2028         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2029         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2030         VMSTATE_END_OF_LIST()
2031     },
2032 };
2033 
2034 static const VMStateDescription vmstate_spapr = {
2035     .name = "spapr",
2036     .version_id = 3,
2037     .minimum_version_id = 1,
2038     .pre_load = spapr_pre_load,
2039     .post_load = spapr_post_load,
2040     .pre_save = spapr_pre_save,
2041     .fields = (VMStateField[]) {
2042         /* used to be @next_irq */
2043         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2044 
2045         /* RTC offset */
2046         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2047 
2048         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2049         VMSTATE_END_OF_LIST()
2050     },
2051     .subsections = (const VMStateDescription*[]) {
2052         &vmstate_spapr_ov5_cas,
2053         &vmstate_spapr_patb_entry,
2054         &vmstate_spapr_pending_events,
2055         &vmstate_spapr_cap_htm,
2056         &vmstate_spapr_cap_vsx,
2057         &vmstate_spapr_cap_dfp,
2058         &vmstate_spapr_cap_cfpc,
2059         &vmstate_spapr_cap_sbbc,
2060         &vmstate_spapr_cap_ibs,
2061         &vmstate_spapr_cap_hpt_maxpagesize,
2062         &vmstate_spapr_irq_map,
2063         &vmstate_spapr_cap_nested_kvm_hv,
2064         &vmstate_spapr_dtb,
2065         &vmstate_spapr_cap_large_decr,
2066         &vmstate_spapr_cap_ccf_assist,
2067         &vmstate_spapr_cap_fwnmi,
2068         &vmstate_spapr_fwnmi,
2069         &vmstate_spapr_cap_rpt_invalidate,
2070         NULL
2071     }
2072 };
2073 
2074 static int htab_save_setup(QEMUFile *f, void *opaque)
2075 {
2076     SpaprMachineState *spapr = opaque;
2077 
2078     /* "Iteration" header */
2079     if (!spapr->htab_shift) {
2080         qemu_put_be32(f, -1);
2081     } else {
2082         qemu_put_be32(f, spapr->htab_shift);
2083     }
2084 
2085     if (spapr->htab) {
2086         spapr->htab_save_index = 0;
2087         spapr->htab_first_pass = true;
2088     } else {
2089         if (spapr->htab_shift) {
2090             assert(kvm_enabled());
2091         }
2092     }
2093 
2094 
2095     return 0;
2096 }
2097 
2098 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2099                             int chunkstart, int n_valid, int n_invalid)
2100 {
2101     qemu_put_be32(f, chunkstart);
2102     qemu_put_be16(f, n_valid);
2103     qemu_put_be16(f, n_invalid);
2104     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2105                     HASH_PTE_SIZE_64 * n_valid);
2106 }
2107 
2108 static void htab_save_end_marker(QEMUFile *f)
2109 {
2110     qemu_put_be32(f, 0);
2111     qemu_put_be16(f, 0);
2112     qemu_put_be16(f, 0);
2113 }
2114 
2115 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2116                                  int64_t max_ns)
2117 {
2118     bool has_timeout = max_ns != -1;
2119     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2120     int index = spapr->htab_save_index;
2121     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2122 
2123     assert(spapr->htab_first_pass);
2124 
2125     do {
2126         int chunkstart;
2127 
2128         /* Consume invalid HPTEs */
2129         while ((index < htabslots)
2130                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2131             CLEAN_HPTE(HPTE(spapr->htab, index));
2132             index++;
2133         }
2134 
2135         /* Consume valid HPTEs */
2136         chunkstart = index;
2137         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2138                && HPTE_VALID(HPTE(spapr->htab, index))) {
2139             CLEAN_HPTE(HPTE(spapr->htab, index));
2140             index++;
2141         }
2142 
2143         if (index > chunkstart) {
2144             int n_valid = index - chunkstart;
2145 
2146             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2147 
2148             if (has_timeout &&
2149                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2150                 break;
2151             }
2152         }
2153     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2154 
2155     if (index >= htabslots) {
2156         assert(index == htabslots);
2157         index = 0;
2158         spapr->htab_first_pass = false;
2159     }
2160     spapr->htab_save_index = index;
2161 }
2162 
2163 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2164                                 int64_t max_ns)
2165 {
2166     bool final = max_ns < 0;
2167     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2168     int examined = 0, sent = 0;
2169     int index = spapr->htab_save_index;
2170     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2171 
2172     assert(!spapr->htab_first_pass);
2173 
2174     do {
2175         int chunkstart, invalidstart;
2176 
2177         /* Consume non-dirty HPTEs */
2178         while ((index < htabslots)
2179                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2180             index++;
2181             examined++;
2182         }
2183 
2184         chunkstart = index;
2185         /* Consume valid dirty HPTEs */
2186         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2187                && HPTE_DIRTY(HPTE(spapr->htab, index))
2188                && HPTE_VALID(HPTE(spapr->htab, index))) {
2189             CLEAN_HPTE(HPTE(spapr->htab, index));
2190             index++;
2191             examined++;
2192         }
2193 
2194         invalidstart = index;
2195         /* Consume invalid dirty HPTEs */
2196         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2197                && HPTE_DIRTY(HPTE(spapr->htab, index))
2198                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2199             CLEAN_HPTE(HPTE(spapr->htab, index));
2200             index++;
2201             examined++;
2202         }
2203 
2204         if (index > chunkstart) {
2205             int n_valid = invalidstart - chunkstart;
2206             int n_invalid = index - invalidstart;
2207 
2208             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2209             sent += index - chunkstart;
2210 
2211             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2212                 break;
2213             }
2214         }
2215 
2216         if (examined >= htabslots) {
2217             break;
2218         }
2219 
2220         if (index >= htabslots) {
2221             assert(index == htabslots);
2222             index = 0;
2223         }
2224     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2225 
2226     if (index >= htabslots) {
2227         assert(index == htabslots);
2228         index = 0;
2229     }
2230 
2231     spapr->htab_save_index = index;
2232 
2233     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2234 }
2235 
2236 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2237 #define MAX_KVM_BUF_SIZE    2048
2238 
2239 static int htab_save_iterate(QEMUFile *f, void *opaque)
2240 {
2241     SpaprMachineState *spapr = opaque;
2242     int fd;
2243     int rc = 0;
2244 
2245     /* Iteration header */
2246     if (!spapr->htab_shift) {
2247         qemu_put_be32(f, -1);
2248         return 1;
2249     } else {
2250         qemu_put_be32(f, 0);
2251     }
2252 
2253     if (!spapr->htab) {
2254         assert(kvm_enabled());
2255 
2256         fd = get_htab_fd(spapr);
2257         if (fd < 0) {
2258             return fd;
2259         }
2260 
2261         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2262         if (rc < 0) {
2263             return rc;
2264         }
2265     } else  if (spapr->htab_first_pass) {
2266         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2267     } else {
2268         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2269     }
2270 
2271     htab_save_end_marker(f);
2272 
2273     return rc;
2274 }
2275 
2276 static int htab_save_complete(QEMUFile *f, void *opaque)
2277 {
2278     SpaprMachineState *spapr = opaque;
2279     int fd;
2280 
2281     /* Iteration header */
2282     if (!spapr->htab_shift) {
2283         qemu_put_be32(f, -1);
2284         return 0;
2285     } else {
2286         qemu_put_be32(f, 0);
2287     }
2288 
2289     if (!spapr->htab) {
2290         int rc;
2291 
2292         assert(kvm_enabled());
2293 
2294         fd = get_htab_fd(spapr);
2295         if (fd < 0) {
2296             return fd;
2297         }
2298 
2299         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2300         if (rc < 0) {
2301             return rc;
2302         }
2303     } else {
2304         if (spapr->htab_first_pass) {
2305             htab_save_first_pass(f, spapr, -1);
2306         }
2307         htab_save_later_pass(f, spapr, -1);
2308     }
2309 
2310     /* End marker */
2311     htab_save_end_marker(f);
2312 
2313     return 0;
2314 }
2315 
2316 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2317 {
2318     SpaprMachineState *spapr = opaque;
2319     uint32_t section_hdr;
2320     int fd = -1;
2321     Error *local_err = NULL;
2322 
2323     if (version_id < 1 || version_id > 1) {
2324         error_report("htab_load() bad version");
2325         return -EINVAL;
2326     }
2327 
2328     section_hdr = qemu_get_be32(f);
2329 
2330     if (section_hdr == -1) {
2331         spapr_free_hpt(spapr);
2332         return 0;
2333     }
2334 
2335     if (section_hdr) {
2336         int ret;
2337 
2338         /* First section gives the htab size */
2339         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2340         if (ret < 0) {
2341             error_report_err(local_err);
2342             return ret;
2343         }
2344         return 0;
2345     }
2346 
2347     if (!spapr->htab) {
2348         assert(kvm_enabled());
2349 
2350         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2351         if (fd < 0) {
2352             error_report_err(local_err);
2353             return fd;
2354         }
2355     }
2356 
2357     while (true) {
2358         uint32_t index;
2359         uint16_t n_valid, n_invalid;
2360 
2361         index = qemu_get_be32(f);
2362         n_valid = qemu_get_be16(f);
2363         n_invalid = qemu_get_be16(f);
2364 
2365         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2366             /* End of Stream */
2367             break;
2368         }
2369 
2370         if ((index + n_valid + n_invalid) >
2371             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2372             /* Bad index in stream */
2373             error_report(
2374                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2375                 index, n_valid, n_invalid, spapr->htab_shift);
2376             return -EINVAL;
2377         }
2378 
2379         if (spapr->htab) {
2380             if (n_valid) {
2381                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2382                                 HASH_PTE_SIZE_64 * n_valid);
2383             }
2384             if (n_invalid) {
2385                 memset(HPTE(spapr->htab, index + n_valid), 0,
2386                        HASH_PTE_SIZE_64 * n_invalid);
2387             }
2388         } else {
2389             int rc;
2390 
2391             assert(fd >= 0);
2392 
2393             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2394                                         &local_err);
2395             if (rc < 0) {
2396                 error_report_err(local_err);
2397                 return rc;
2398             }
2399         }
2400     }
2401 
2402     if (!spapr->htab) {
2403         assert(fd >= 0);
2404         close(fd);
2405     }
2406 
2407     return 0;
2408 }
2409 
2410 static void htab_save_cleanup(void *opaque)
2411 {
2412     SpaprMachineState *spapr = opaque;
2413 
2414     close_htab_fd(spapr);
2415 }
2416 
2417 static SaveVMHandlers savevm_htab_handlers = {
2418     .save_setup = htab_save_setup,
2419     .save_live_iterate = htab_save_iterate,
2420     .save_live_complete_precopy = htab_save_complete,
2421     .save_cleanup = htab_save_cleanup,
2422     .load_state = htab_load,
2423 };
2424 
2425 static void spapr_boot_set(void *opaque, const char *boot_device,
2426                            Error **errp)
2427 {
2428     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2429 
2430     g_free(spapr->boot_device);
2431     spapr->boot_device = g_strdup(boot_device);
2432 }
2433 
2434 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2435 {
2436     MachineState *machine = MACHINE(spapr);
2437     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2438     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2439     int i;
2440 
2441     for (i = 0; i < nr_lmbs; i++) {
2442         uint64_t addr;
2443 
2444         addr = i * lmb_size + machine->device_memory->base;
2445         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2446                                addr / lmb_size);
2447     }
2448 }
2449 
2450 /*
2451  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2452  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2453  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2454  */
2455 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2456 {
2457     int i;
2458 
2459     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2460         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2461                    " is not aligned to %" PRIu64 " MiB",
2462                    machine->ram_size,
2463                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2464         return;
2465     }
2466 
2467     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2468         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2469                    " is not aligned to %" PRIu64 " MiB",
2470                    machine->ram_size,
2471                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2472         return;
2473     }
2474 
2475     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2476         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2477             error_setg(errp,
2478                        "Node %d memory size 0x%" PRIx64
2479                        " is not aligned to %" PRIu64 " MiB",
2480                        i, machine->numa_state->nodes[i].node_mem,
2481                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2482             return;
2483         }
2484     }
2485 }
2486 
2487 /* find cpu slot in machine->possible_cpus by core_id */
2488 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2489 {
2490     int index = id / ms->smp.threads;
2491 
2492     if (index >= ms->possible_cpus->len) {
2493         return NULL;
2494     }
2495     if (idx) {
2496         *idx = index;
2497     }
2498     return &ms->possible_cpus->cpus[index];
2499 }
2500 
2501 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2502 {
2503     MachineState *ms = MACHINE(spapr);
2504     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2505     Error *local_err = NULL;
2506     bool vsmt_user = !!spapr->vsmt;
2507     int kvm_smt = kvmppc_smt_threads();
2508     int ret;
2509     unsigned int smp_threads = ms->smp.threads;
2510 
2511     if (!kvm_enabled() && (smp_threads > 1)) {
2512         error_setg(errp, "TCG cannot support more than 1 thread/core "
2513                    "on a pseries machine");
2514         return;
2515     }
2516     if (!is_power_of_2(smp_threads)) {
2517         error_setg(errp, "Cannot support %d threads/core on a pseries "
2518                    "machine because it must be a power of 2", smp_threads);
2519         return;
2520     }
2521 
2522     /* Detemine the VSMT mode to use: */
2523     if (vsmt_user) {
2524         if (spapr->vsmt < smp_threads) {
2525             error_setg(errp, "Cannot support VSMT mode %d"
2526                        " because it must be >= threads/core (%d)",
2527                        spapr->vsmt, smp_threads);
2528             return;
2529         }
2530         /* In this case, spapr->vsmt has been set by the command line */
2531     } else if (!smc->smp_threads_vsmt) {
2532         /*
2533          * Default VSMT value is tricky, because we need it to be as
2534          * consistent as possible (for migration), but this requires
2535          * changing it for at least some existing cases.  We pick 8 as
2536          * the value that we'd get with KVM on POWER8, the
2537          * overwhelmingly common case in production systems.
2538          */
2539         spapr->vsmt = MAX(8, smp_threads);
2540     } else {
2541         spapr->vsmt = smp_threads;
2542     }
2543 
2544     /* KVM: If necessary, set the SMT mode: */
2545     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2546         ret = kvmppc_set_smt_threads(spapr->vsmt);
2547         if (ret) {
2548             /* Looks like KVM isn't able to change VSMT mode */
2549             error_setg(&local_err,
2550                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2551                        spapr->vsmt, ret);
2552             /* We can live with that if the default one is big enough
2553              * for the number of threads, and a submultiple of the one
2554              * we want.  In this case we'll waste some vcpu ids, but
2555              * behaviour will be correct */
2556             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2557                 warn_report_err(local_err);
2558             } else {
2559                 if (!vsmt_user) {
2560                     error_append_hint(&local_err,
2561                                       "On PPC, a VM with %d threads/core"
2562                                       " on a host with %d threads/core"
2563                                       " requires the use of VSMT mode %d.\n",
2564                                       smp_threads, kvm_smt, spapr->vsmt);
2565                 }
2566                 kvmppc_error_append_smt_possible_hint(&local_err);
2567                 error_propagate(errp, local_err);
2568             }
2569         }
2570     }
2571     /* else TCG: nothing to do currently */
2572 }
2573 
2574 static void spapr_init_cpus(SpaprMachineState *spapr)
2575 {
2576     MachineState *machine = MACHINE(spapr);
2577     MachineClass *mc = MACHINE_GET_CLASS(machine);
2578     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2579     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2580     const CPUArchIdList *possible_cpus;
2581     unsigned int smp_cpus = machine->smp.cpus;
2582     unsigned int smp_threads = machine->smp.threads;
2583     unsigned int max_cpus = machine->smp.max_cpus;
2584     int boot_cores_nr = smp_cpus / smp_threads;
2585     int i;
2586 
2587     possible_cpus = mc->possible_cpu_arch_ids(machine);
2588     if (mc->has_hotpluggable_cpus) {
2589         if (smp_cpus % smp_threads) {
2590             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2591                          smp_cpus, smp_threads);
2592             exit(1);
2593         }
2594         if (max_cpus % smp_threads) {
2595             error_report("max_cpus (%u) must be multiple of threads (%u)",
2596                          max_cpus, smp_threads);
2597             exit(1);
2598         }
2599     } else {
2600         if (max_cpus != smp_cpus) {
2601             error_report("This machine version does not support CPU hotplug");
2602             exit(1);
2603         }
2604         boot_cores_nr = possible_cpus->len;
2605     }
2606 
2607     if (smc->pre_2_10_has_unused_icps) {
2608         int i;
2609 
2610         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2611             /* Dummy entries get deregistered when real ICPState objects
2612              * are registered during CPU core hotplug.
2613              */
2614             pre_2_10_vmstate_register_dummy_icp(i);
2615         }
2616     }
2617 
2618     for (i = 0; i < possible_cpus->len; i++) {
2619         int core_id = i * smp_threads;
2620 
2621         if (mc->has_hotpluggable_cpus) {
2622             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2623                                    spapr_vcpu_id(spapr, core_id));
2624         }
2625 
2626         if (i < boot_cores_nr) {
2627             Object *core  = object_new(type);
2628             int nr_threads = smp_threads;
2629 
2630             /* Handle the partially filled core for older machine types */
2631             if ((i + 1) * smp_threads >= smp_cpus) {
2632                 nr_threads = smp_cpus - i * smp_threads;
2633             }
2634 
2635             object_property_set_int(core, "nr-threads", nr_threads,
2636                                     &error_fatal);
2637             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2638                                     &error_fatal);
2639             qdev_realize(DEVICE(core), NULL, &error_fatal);
2640 
2641             object_unref(core);
2642         }
2643     }
2644 }
2645 
2646 static PCIHostState *spapr_create_default_phb(void)
2647 {
2648     DeviceState *dev;
2649 
2650     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2651     qdev_prop_set_uint32(dev, "index", 0);
2652     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2653 
2654     return PCI_HOST_BRIDGE(dev);
2655 }
2656 
2657 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2658 {
2659     MachineState *machine = MACHINE(spapr);
2660     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2661     hwaddr rma_size = machine->ram_size;
2662     hwaddr node0_size = spapr_node0_size(machine);
2663 
2664     /* RMA has to fit in the first NUMA node */
2665     rma_size = MIN(rma_size, node0_size);
2666 
2667     /*
2668      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2669      * never exceed that
2670      */
2671     rma_size = MIN(rma_size, 1 * TiB);
2672 
2673     /*
2674      * Clamp the RMA size based on machine type.  This is for
2675      * migration compatibility with older qemu versions, which limited
2676      * the RMA size for complicated and mostly bad reasons.
2677      */
2678     if (smc->rma_limit) {
2679         rma_size = MIN(rma_size, smc->rma_limit);
2680     }
2681 
2682     if (rma_size < MIN_RMA_SLOF) {
2683         error_setg(errp,
2684                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2685                    "ldMiB guest RMA (Real Mode Area memory)",
2686                    MIN_RMA_SLOF / MiB);
2687         return 0;
2688     }
2689 
2690     return rma_size;
2691 }
2692 
2693 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2694 {
2695     MachineState *machine = MACHINE(spapr);
2696     int i;
2697 
2698     for (i = 0; i < machine->ram_slots; i++) {
2699         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2700     }
2701 }
2702 
2703 /* pSeries LPAR / sPAPR hardware init */
2704 static void spapr_machine_init(MachineState *machine)
2705 {
2706     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2707     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2708     MachineClass *mc = MACHINE_GET_CLASS(machine);
2709     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2710     const char *bios_name = machine->firmware ?: bios_default;
2711     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2712     const char *kernel_filename = machine->kernel_filename;
2713     const char *initrd_filename = machine->initrd_filename;
2714     PCIHostState *phb;
2715     int i;
2716     MemoryRegion *sysmem = get_system_memory();
2717     long load_limit, fw_size;
2718     Error *resize_hpt_err = NULL;
2719 
2720     if (!filename) {
2721         error_report("Could not find LPAR firmware '%s'", bios_name);
2722         exit(1);
2723     }
2724     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2725     if (fw_size <= 0) {
2726         error_report("Could not load LPAR firmware '%s'", filename);
2727         exit(1);
2728     }
2729 
2730     /*
2731      * if Secure VM (PEF) support is configured, then initialize it
2732      */
2733     pef_kvm_init(machine->cgs, &error_fatal);
2734 
2735     msi_nonbroken = true;
2736 
2737     QLIST_INIT(&spapr->phbs);
2738     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2739 
2740     /* Determine capabilities to run with */
2741     spapr_caps_init(spapr);
2742 
2743     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2744     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2745         /*
2746          * If the user explicitly requested a mode we should either
2747          * supply it, or fail completely (which we do below).  But if
2748          * it's not set explicitly, we reset our mode to something
2749          * that works
2750          */
2751         if (resize_hpt_err) {
2752             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2753             error_free(resize_hpt_err);
2754             resize_hpt_err = NULL;
2755         } else {
2756             spapr->resize_hpt = smc->resize_hpt_default;
2757         }
2758     }
2759 
2760     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2761 
2762     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2763         /*
2764          * User requested HPT resize, but this host can't supply it.  Bail out
2765          */
2766         error_report_err(resize_hpt_err);
2767         exit(1);
2768     }
2769     error_free(resize_hpt_err);
2770 
2771     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2772 
2773     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2774     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2775 
2776     /*
2777      * VSMT must be set in order to be able to compute VCPU ids, ie to
2778      * call spapr_max_server_number() or spapr_vcpu_id().
2779      */
2780     spapr_set_vsmt_mode(spapr, &error_fatal);
2781 
2782     /* Set up Interrupt Controller before we create the VCPUs */
2783     spapr_irq_init(spapr, &error_fatal);
2784 
2785     /* Set up containers for ibm,client-architecture-support negotiated options
2786      */
2787     spapr->ov5 = spapr_ovec_new();
2788     spapr->ov5_cas = spapr_ovec_new();
2789 
2790     if (smc->dr_lmb_enabled) {
2791         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2792         spapr_validate_node_memory(machine, &error_fatal);
2793     }
2794 
2795     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2796 
2797     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2798     if (!smc->pre_6_2_numa_affinity) {
2799         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2800     }
2801 
2802     /* advertise support for dedicated HP event source to guests */
2803     if (spapr->use_hotplug_event_source) {
2804         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2805     }
2806 
2807     /* advertise support for HPT resizing */
2808     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2809         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2810     }
2811 
2812     /* advertise support for ibm,dyamic-memory-v2 */
2813     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2814 
2815     /* advertise XIVE on POWER9 machines */
2816     if (spapr->irq->xive) {
2817         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2818     }
2819 
2820     /* init CPUs */
2821     spapr_init_cpus(spapr);
2822 
2823     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2824 
2825     /* Init numa_assoc_array */
2826     spapr_numa_associativity_init(spapr, machine);
2827 
2828     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2829         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2830                               spapr->max_compat_pvr)) {
2831         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2832         /* KVM and TCG always allow GTSE with radix... */
2833         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2834     }
2835     /* ... but not with hash (currently). */
2836 
2837     if (kvm_enabled()) {
2838         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2839         kvmppc_enable_logical_ci_hcalls();
2840         kvmppc_enable_set_mode_hcall();
2841 
2842         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2843         kvmppc_enable_clear_ref_mod_hcalls();
2844 
2845         /* Enable H_PAGE_INIT */
2846         kvmppc_enable_h_page_init();
2847     }
2848 
2849     /* map RAM */
2850     memory_region_add_subregion(sysmem, 0, machine->ram);
2851 
2852     /* always allocate the device memory information */
2853     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2854 
2855     /* initialize hotplug memory address space */
2856     if (machine->ram_size < machine->maxram_size) {
2857         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2858         /*
2859          * Limit the number of hotpluggable memory slots to half the number
2860          * slots that KVM supports, leaving the other half for PCI and other
2861          * devices. However ensure that number of slots doesn't drop below 32.
2862          */
2863         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2864                            SPAPR_MAX_RAM_SLOTS;
2865 
2866         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2867             max_memslots = SPAPR_MAX_RAM_SLOTS;
2868         }
2869         if (machine->ram_slots > max_memslots) {
2870             error_report("Specified number of memory slots %"
2871                          PRIu64" exceeds max supported %d",
2872                          machine->ram_slots, max_memslots);
2873             exit(1);
2874         }
2875 
2876         machine->device_memory->base = ROUND_UP(machine->ram_size,
2877                                                 SPAPR_DEVICE_MEM_ALIGN);
2878         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2879                            "device-memory", device_mem_size);
2880         memory_region_add_subregion(sysmem, machine->device_memory->base,
2881                                     &machine->device_memory->mr);
2882     }
2883 
2884     if (smc->dr_lmb_enabled) {
2885         spapr_create_lmb_dr_connectors(spapr);
2886     }
2887 
2888     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2889         /* Create the error string for live migration blocker */
2890         error_setg(&spapr->fwnmi_migration_blocker,
2891             "A machine check is being handled during migration. The handler"
2892             "may run and log hardware error on the destination");
2893     }
2894 
2895     if (mc->nvdimm_supported) {
2896         spapr_create_nvdimm_dr_connectors(spapr);
2897     }
2898 
2899     /* Set up RTAS event infrastructure */
2900     spapr_events_init(spapr);
2901 
2902     /* Set up the RTC RTAS interfaces */
2903     spapr_rtc_create(spapr);
2904 
2905     /* Set up VIO bus */
2906     spapr->vio_bus = spapr_vio_bus_init();
2907 
2908     for (i = 0; serial_hd(i); i++) {
2909         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2910     }
2911 
2912     /* We always have at least the nvram device on VIO */
2913     spapr_create_nvram(spapr);
2914 
2915     /*
2916      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2917      * connectors (described in root DT node's "ibm,drc-types" property)
2918      * are pre-initialized here. additional child connectors (such as
2919      * connectors for a PHBs PCI slots) are added as needed during their
2920      * parent's realization.
2921      */
2922     if (smc->dr_phb_enabled) {
2923         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2924             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2925         }
2926     }
2927 
2928     /* Set up PCI */
2929     spapr_pci_rtas_init();
2930 
2931     phb = spapr_create_default_phb();
2932 
2933     for (i = 0; i < nb_nics; i++) {
2934         NICInfo *nd = &nd_table[i];
2935 
2936         if (!nd->model) {
2937             nd->model = g_strdup("spapr-vlan");
2938         }
2939 
2940         if (g_str_equal(nd->model, "spapr-vlan") ||
2941             g_str_equal(nd->model, "ibmveth")) {
2942             spapr_vlan_create(spapr->vio_bus, nd);
2943         } else {
2944             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2945         }
2946     }
2947 
2948     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2949         spapr_vscsi_create(spapr->vio_bus);
2950     }
2951 
2952     /* Graphics */
2953     if (spapr_vga_init(phb->bus, &error_fatal)) {
2954         spapr->has_graphics = true;
2955         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2956     }
2957 
2958     if (machine->usb) {
2959         if (smc->use_ohci_by_default) {
2960             pci_create_simple(phb->bus, -1, "pci-ohci");
2961         } else {
2962             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2963         }
2964 
2965         if (spapr->has_graphics) {
2966             USBBus *usb_bus = usb_bus_find(-1);
2967 
2968             usb_create_simple(usb_bus, "usb-kbd");
2969             usb_create_simple(usb_bus, "usb-mouse");
2970         }
2971     }
2972 
2973     if (kernel_filename) {
2974         spapr->kernel_size = load_elf(kernel_filename, NULL,
2975                                       translate_kernel_address, spapr,
2976                                       NULL, NULL, NULL, NULL, 1,
2977                                       PPC_ELF_MACHINE, 0, 0);
2978         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2979             spapr->kernel_size = load_elf(kernel_filename, NULL,
2980                                           translate_kernel_address, spapr,
2981                                           NULL, NULL, NULL, NULL, 0,
2982                                           PPC_ELF_MACHINE, 0, 0);
2983             spapr->kernel_le = spapr->kernel_size > 0;
2984         }
2985         if (spapr->kernel_size < 0) {
2986             error_report("error loading %s: %s", kernel_filename,
2987                          load_elf_strerror(spapr->kernel_size));
2988             exit(1);
2989         }
2990 
2991         /* load initrd */
2992         if (initrd_filename) {
2993             /* Try to locate the initrd in the gap between the kernel
2994              * and the firmware. Add a bit of space just in case
2995              */
2996             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2997                                   + 0x1ffff) & ~0xffff;
2998             spapr->initrd_size = load_image_targphys(initrd_filename,
2999                                                      spapr->initrd_base,
3000                                                      load_limit
3001                                                      - spapr->initrd_base);
3002             if (spapr->initrd_size < 0) {
3003                 error_report("could not load initial ram disk '%s'",
3004                              initrd_filename);
3005                 exit(1);
3006             }
3007         }
3008     }
3009 
3010     /* FIXME: Should register things through the MachineState's qdev
3011      * interface, this is a legacy from the sPAPREnvironment structure
3012      * which predated MachineState but had a similar function */
3013     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3014     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3015                          &savevm_htab_handlers, spapr);
3016 
3017     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3018 
3019     qemu_register_boot_set(spapr_boot_set, spapr);
3020 
3021     /*
3022      * Nothing needs to be done to resume a suspended guest because
3023      * suspending does not change the machine state, so no need for
3024      * a ->wakeup method.
3025      */
3026     qemu_register_wakeup_support();
3027 
3028     if (kvm_enabled()) {
3029         /* to stop and start vmclock */
3030         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3031                                          &spapr->tb);
3032 
3033         kvmppc_spapr_enable_inkernel_multitce();
3034     }
3035 
3036     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3037     if (spapr->vof) {
3038         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3039         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3040     }
3041 }
3042 
3043 #define DEFAULT_KVM_TYPE "auto"
3044 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3045 {
3046     /*
3047      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3048      * accomodate the 'HV' and 'PV' formats that exists in the
3049      * wild. The 'auto' mode is being introduced already as
3050      * lower-case, thus we don't need to bother checking for
3051      * "AUTO".
3052      */
3053     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3054         return 0;
3055     }
3056 
3057     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3058         return 1;
3059     }
3060 
3061     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3062         return 2;
3063     }
3064 
3065     error_report("Unknown kvm-type specified '%s'", vm_type);
3066     exit(1);
3067 }
3068 
3069 /*
3070  * Implementation of an interface to adjust firmware path
3071  * for the bootindex property handling.
3072  */
3073 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3074                                    DeviceState *dev)
3075 {
3076 #define CAST(type, obj, name) \
3077     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3078     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3079     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3080     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3081     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3082 
3083     if (d && bus) {
3084         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3085         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3086         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3087 
3088         if (spapr) {
3089             /*
3090              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3091              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3092              * 0x8000 | (target << 8) | (bus << 5) | lun
3093              * (see the "Logical unit addressing format" table in SAM5)
3094              */
3095             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3096             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3097                                    (uint64_t)id << 48);
3098         } else if (virtio) {
3099             /*
3100              * We use SRP luns of the form 01000000 | (target << 8) | lun
3101              * in the top 32 bits of the 64-bit LUN
3102              * Note: the quote above is from SLOF and it is wrong,
3103              * the actual binding is:
3104              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3105              */
3106             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3107             if (d->lun >= 256) {
3108                 /* Use the LUN "flat space addressing method" */
3109                 id |= 0x4000;
3110             }
3111             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3112                                    (uint64_t)id << 32);
3113         } else if (usb) {
3114             /*
3115              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3116              * in the top 32 bits of the 64-bit LUN
3117              */
3118             unsigned usb_port = atoi(usb->port->path);
3119             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3120             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3121                                    (uint64_t)id << 32);
3122         }
3123     }
3124 
3125     /*
3126      * SLOF probes the USB devices, and if it recognizes that the device is a
3127      * storage device, it changes its name to "storage" instead of "usb-host",
3128      * and additionally adds a child node for the SCSI LUN, so the correct
3129      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3130      */
3131     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3132         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3133         if (usb_device_is_scsi_storage(usbdev)) {
3134             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3135         }
3136     }
3137 
3138     if (phb) {
3139         /* Replace "pci" with "pci@800000020000000" */
3140         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3141     }
3142 
3143     if (vsc) {
3144         /* Same logic as virtio above */
3145         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3146         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3147     }
3148 
3149     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3150         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3151         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3152         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3153     }
3154 
3155     if (pcidev) {
3156         return spapr_pci_fw_dev_name(pcidev);
3157     }
3158 
3159     return NULL;
3160 }
3161 
3162 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3163 {
3164     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3165 
3166     return g_strdup(spapr->kvm_type);
3167 }
3168 
3169 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3170 {
3171     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3172 
3173     g_free(spapr->kvm_type);
3174     spapr->kvm_type = g_strdup(value);
3175 }
3176 
3177 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3178 {
3179     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3180 
3181     return spapr->use_hotplug_event_source;
3182 }
3183 
3184 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3185                                             Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     spapr->use_hotplug_event_source = value;
3190 }
3191 
3192 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3193 {
3194     return true;
3195 }
3196 
3197 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3198 {
3199     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3200 
3201     switch (spapr->resize_hpt) {
3202     case SPAPR_RESIZE_HPT_DEFAULT:
3203         return g_strdup("default");
3204     case SPAPR_RESIZE_HPT_DISABLED:
3205         return g_strdup("disabled");
3206     case SPAPR_RESIZE_HPT_ENABLED:
3207         return g_strdup("enabled");
3208     case SPAPR_RESIZE_HPT_REQUIRED:
3209         return g_strdup("required");
3210     }
3211     g_assert_not_reached();
3212 }
3213 
3214 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3215 {
3216     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3217 
3218     if (strcmp(value, "default") == 0) {
3219         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3220     } else if (strcmp(value, "disabled") == 0) {
3221         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3222     } else if (strcmp(value, "enabled") == 0) {
3223         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3224     } else if (strcmp(value, "required") == 0) {
3225         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3226     } else {
3227         error_setg(errp, "Bad value for \"resize-hpt\" property");
3228     }
3229 }
3230 
3231 static bool spapr_get_vof(Object *obj, Error **errp)
3232 {
3233     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3234 
3235     return spapr->vof != NULL;
3236 }
3237 
3238 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3239 {
3240     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3241 
3242     if (spapr->vof) {
3243         vof_cleanup(spapr->vof);
3244         g_free(spapr->vof);
3245         spapr->vof = NULL;
3246     }
3247     if (!value) {
3248         return;
3249     }
3250     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3251 }
3252 
3253 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3254 {
3255     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3256 
3257     if (spapr->irq == &spapr_irq_xics_legacy) {
3258         return g_strdup("legacy");
3259     } else if (spapr->irq == &spapr_irq_xics) {
3260         return g_strdup("xics");
3261     } else if (spapr->irq == &spapr_irq_xive) {
3262         return g_strdup("xive");
3263     } else if (spapr->irq == &spapr_irq_dual) {
3264         return g_strdup("dual");
3265     }
3266     g_assert_not_reached();
3267 }
3268 
3269 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3270 {
3271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272 
3273     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3274         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3275         return;
3276     }
3277 
3278     /* The legacy IRQ backend can not be set */
3279     if (strcmp(value, "xics") == 0) {
3280         spapr->irq = &spapr_irq_xics;
3281     } else if (strcmp(value, "xive") == 0) {
3282         spapr->irq = &spapr_irq_xive;
3283     } else if (strcmp(value, "dual") == 0) {
3284         spapr->irq = &spapr_irq_dual;
3285     } else {
3286         error_setg(errp, "Bad value for \"ic-mode\" property");
3287     }
3288 }
3289 
3290 static char *spapr_get_host_model(Object *obj, Error **errp)
3291 {
3292     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3293 
3294     return g_strdup(spapr->host_model);
3295 }
3296 
3297 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3298 {
3299     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3300 
3301     g_free(spapr->host_model);
3302     spapr->host_model = g_strdup(value);
3303 }
3304 
3305 static char *spapr_get_host_serial(Object *obj, Error **errp)
3306 {
3307     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3308 
3309     return g_strdup(spapr->host_serial);
3310 }
3311 
3312 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3313 {
3314     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3315 
3316     g_free(spapr->host_serial);
3317     spapr->host_serial = g_strdup(value);
3318 }
3319 
3320 static void spapr_instance_init(Object *obj)
3321 {
3322     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3323     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3324     MachineState *ms = MACHINE(spapr);
3325     MachineClass *mc = MACHINE_GET_CLASS(ms);
3326 
3327     /*
3328      * NVDIMM support went live in 5.1 without considering that, in
3329      * other archs, the user needs to enable NVDIMM support with the
3330      * 'nvdimm' machine option and the default behavior is NVDIMM
3331      * support disabled. It is too late to roll back to the standard
3332      * behavior without breaking 5.1 guests.
3333      */
3334     if (mc->nvdimm_supported) {
3335         ms->nvdimms_state->is_enabled = true;
3336     }
3337 
3338     spapr->htab_fd = -1;
3339     spapr->use_hotplug_event_source = true;
3340     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3341     object_property_add_str(obj, "kvm-type",
3342                             spapr_get_kvm_type, spapr_set_kvm_type);
3343     object_property_set_description(obj, "kvm-type",
3344                                     "Specifies the KVM virtualization mode (auto,"
3345                                     " hv, pr). Defaults to 'auto'. This mode will use"
3346                                     " any available KVM module loaded in the host,"
3347                                     " where kvm_hv takes precedence if both kvm_hv and"
3348                                     " kvm_pr are loaded.");
3349     object_property_add_bool(obj, "modern-hotplug-events",
3350                             spapr_get_modern_hotplug_events,
3351                             spapr_set_modern_hotplug_events);
3352     object_property_set_description(obj, "modern-hotplug-events",
3353                                     "Use dedicated hotplug event mechanism in"
3354                                     " place of standard EPOW events when possible"
3355                                     " (required for memory hot-unplug support)");
3356     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3357                             "Maximum permitted CPU compatibility mode");
3358 
3359     object_property_add_str(obj, "resize-hpt",
3360                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3361     object_property_set_description(obj, "resize-hpt",
3362                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3363     object_property_add_uint32_ptr(obj, "vsmt",
3364                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3365     object_property_set_description(obj, "vsmt",
3366                                     "Virtual SMT: KVM behaves as if this were"
3367                                     " the host's SMT mode");
3368 
3369     object_property_add_bool(obj, "vfio-no-msix-emulation",
3370                              spapr_get_msix_emulation, NULL);
3371 
3372     object_property_add_uint64_ptr(obj, "kernel-addr",
3373                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3374     object_property_set_description(obj, "kernel-addr",
3375                                     stringify(KERNEL_LOAD_ADDR)
3376                                     " for -kernel is the default");
3377     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3378 
3379     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3380     object_property_set_description(obj, "x-vof",
3381                                     "Enable Virtual Open Firmware (experimental)");
3382 
3383     /* The machine class defines the default interrupt controller mode */
3384     spapr->irq = smc->irq;
3385     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3386                             spapr_set_ic_mode);
3387     object_property_set_description(obj, "ic-mode",
3388                  "Specifies the interrupt controller mode (xics, xive, dual)");
3389 
3390     object_property_add_str(obj, "host-model",
3391         spapr_get_host_model, spapr_set_host_model);
3392     object_property_set_description(obj, "host-model",
3393         "Host model to advertise in guest device tree");
3394     object_property_add_str(obj, "host-serial",
3395         spapr_get_host_serial, spapr_set_host_serial);
3396     object_property_set_description(obj, "host-serial",
3397         "Host serial number to advertise in guest device tree");
3398 }
3399 
3400 static void spapr_machine_finalizefn(Object *obj)
3401 {
3402     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3403 
3404     g_free(spapr->kvm_type);
3405 }
3406 
3407 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3408 {
3409     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3410     PowerPCCPU *cpu = POWERPC_CPU(cs);
3411     CPUPPCState *env = &cpu->env;
3412 
3413     cpu_synchronize_state(cs);
3414     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3415     if (spapr->fwnmi_system_reset_addr != -1) {
3416         uint64_t rtas_addr, addr;
3417 
3418         /* get rtas addr from fdt */
3419         rtas_addr = spapr_get_rtas_addr();
3420         if (!rtas_addr) {
3421             qemu_system_guest_panicked(NULL);
3422             return;
3423         }
3424 
3425         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3426         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3427         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3428         env->gpr[3] = addr;
3429     }
3430     ppc_cpu_do_system_reset(cs);
3431     if (spapr->fwnmi_system_reset_addr != -1) {
3432         env->nip = spapr->fwnmi_system_reset_addr;
3433     }
3434 }
3435 
3436 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3437 {
3438     CPUState *cs;
3439 
3440     CPU_FOREACH(cs) {
3441         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3442     }
3443 }
3444 
3445 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3446                           void *fdt, int *fdt_start_offset, Error **errp)
3447 {
3448     uint64_t addr;
3449     uint32_t node;
3450 
3451     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3452     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3453                                     &error_abort);
3454     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3455                                              SPAPR_MEMORY_BLOCK_SIZE);
3456     return 0;
3457 }
3458 
3459 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3460                            bool dedicated_hp_event_source)
3461 {
3462     SpaprDrc *drc;
3463     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3464     int i;
3465     uint64_t addr = addr_start;
3466     bool hotplugged = spapr_drc_hotplugged(dev);
3467 
3468     for (i = 0; i < nr_lmbs; i++) {
3469         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3470                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3471         g_assert(drc);
3472 
3473         /*
3474          * memory_device_get_free_addr() provided a range of free addresses
3475          * that doesn't overlap with any existing mapping at pre-plug. The
3476          * corresponding LMB DRCs are thus assumed to be all attachable.
3477          */
3478         spapr_drc_attach(drc, dev);
3479         if (!hotplugged) {
3480             spapr_drc_reset(drc);
3481         }
3482         addr += SPAPR_MEMORY_BLOCK_SIZE;
3483     }
3484     /* send hotplug notification to the
3485      * guest only in case of hotplugged memory
3486      */
3487     if (hotplugged) {
3488         if (dedicated_hp_event_source) {
3489             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3490                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3491             g_assert(drc);
3492             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3493                                                    nr_lmbs,
3494                                                    spapr_drc_index(drc));
3495         } else {
3496             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3497                                            nr_lmbs);
3498         }
3499     }
3500 }
3501 
3502 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3503 {
3504     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3505     PCDIMMDevice *dimm = PC_DIMM(dev);
3506     uint64_t size, addr;
3507     int64_t slot;
3508     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3509 
3510     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3511 
3512     pc_dimm_plug(dimm, MACHINE(ms));
3513 
3514     if (!is_nvdimm) {
3515         addr = object_property_get_uint(OBJECT(dimm),
3516                                         PC_DIMM_ADDR_PROP, &error_abort);
3517         spapr_add_lmbs(dev, addr, size,
3518                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3519     } else {
3520         slot = object_property_get_int(OBJECT(dimm),
3521                                        PC_DIMM_SLOT_PROP, &error_abort);
3522         /* We should have valid slot number at this point */
3523         g_assert(slot >= 0);
3524         spapr_add_nvdimm(dev, slot);
3525     }
3526 }
3527 
3528 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3529                                   Error **errp)
3530 {
3531     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3532     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3533     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3534     PCDIMMDevice *dimm = PC_DIMM(dev);
3535     Error *local_err = NULL;
3536     uint64_t size;
3537     Object *memdev;
3538     hwaddr pagesize;
3539 
3540     if (!smc->dr_lmb_enabled) {
3541         error_setg(errp, "Memory hotplug not supported for this machine");
3542         return;
3543     }
3544 
3545     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3546     if (local_err) {
3547         error_propagate(errp, local_err);
3548         return;
3549     }
3550 
3551     if (is_nvdimm) {
3552         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3553             return;
3554         }
3555     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3556         error_setg(errp, "Hotplugged memory size must be a multiple of "
3557                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3558         return;
3559     }
3560 
3561     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3562                                       &error_abort);
3563     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3564     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3565         return;
3566     }
3567 
3568     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3569 }
3570 
3571 struct SpaprDimmState {
3572     PCDIMMDevice *dimm;
3573     uint32_t nr_lmbs;
3574     QTAILQ_ENTRY(SpaprDimmState) next;
3575 };
3576 
3577 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3578                                                        PCDIMMDevice *dimm)
3579 {
3580     SpaprDimmState *dimm_state = NULL;
3581 
3582     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3583         if (dimm_state->dimm == dimm) {
3584             break;
3585         }
3586     }
3587     return dimm_state;
3588 }
3589 
3590 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3591                                                       uint32_t nr_lmbs,
3592                                                       PCDIMMDevice *dimm)
3593 {
3594     SpaprDimmState *ds = NULL;
3595 
3596     /*
3597      * If this request is for a DIMM whose removal had failed earlier
3598      * (due to guest's refusal to remove the LMBs), we would have this
3599      * dimm already in the pending_dimm_unplugs list. In that
3600      * case don't add again.
3601      */
3602     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3603     if (!ds) {
3604         ds = g_new0(SpaprDimmState, 1);
3605         ds->nr_lmbs = nr_lmbs;
3606         ds->dimm = dimm;
3607         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3608     }
3609     return ds;
3610 }
3611 
3612 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3613                                               SpaprDimmState *dimm_state)
3614 {
3615     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3616     g_free(dimm_state);
3617 }
3618 
3619 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3620                                                         PCDIMMDevice *dimm)
3621 {
3622     SpaprDrc *drc;
3623     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3624                                                   &error_abort);
3625     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3626     uint32_t avail_lmbs = 0;
3627     uint64_t addr_start, addr;
3628     int i;
3629 
3630     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3631                                           &error_abort);
3632 
3633     addr = addr_start;
3634     for (i = 0; i < nr_lmbs; i++) {
3635         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3636                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3637         g_assert(drc);
3638         if (drc->dev) {
3639             avail_lmbs++;
3640         }
3641         addr += SPAPR_MEMORY_BLOCK_SIZE;
3642     }
3643 
3644     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3645 }
3646 
3647 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3648 {
3649     SpaprDimmState *ds;
3650     PCDIMMDevice *dimm;
3651     SpaprDrc *drc;
3652     uint32_t nr_lmbs;
3653     uint64_t size, addr_start, addr;
3654     g_autofree char *qapi_error = NULL;
3655     int i;
3656 
3657     if (!dev) {
3658         return;
3659     }
3660 
3661     dimm = PC_DIMM(dev);
3662     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3663 
3664     /*
3665      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3666      * unplug state, but one of its DRC is marked as unplug_requested.
3667      * This is bad and weird enough to g_assert() out.
3668      */
3669     g_assert(ds);
3670 
3671     spapr_pending_dimm_unplugs_remove(spapr, ds);
3672 
3673     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3674     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3675 
3676     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3677                                           &error_abort);
3678 
3679     addr = addr_start;
3680     for (i = 0; i < nr_lmbs; i++) {
3681         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3682                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3683         g_assert(drc);
3684 
3685         drc->unplug_requested = false;
3686         addr += SPAPR_MEMORY_BLOCK_SIZE;
3687     }
3688 
3689     /*
3690      * Tell QAPI that something happened and the memory
3691      * hotunplug wasn't successful. Keep sending
3692      * MEM_UNPLUG_ERROR even while sending
3693      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3694      * MEM_UNPLUG_ERROR is due.
3695      */
3696     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3697                                  "for device %s", dev->id);
3698 
3699     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3700 
3701     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3702                                               dev->canonical_path);
3703 }
3704 
3705 /* Callback to be called during DRC release. */
3706 void spapr_lmb_release(DeviceState *dev)
3707 {
3708     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3709     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3710     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3711 
3712     /* This information will get lost if a migration occurs
3713      * during the unplug process. In this case recover it. */
3714     if (ds == NULL) {
3715         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3716         g_assert(ds);
3717         /* The DRC being examined by the caller at least must be counted */
3718         g_assert(ds->nr_lmbs);
3719     }
3720 
3721     if (--ds->nr_lmbs) {
3722         return;
3723     }
3724 
3725     /*
3726      * Now that all the LMBs have been removed by the guest, call the
3727      * unplug handler chain. This can never fail.
3728      */
3729     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3730     object_unparent(OBJECT(dev));
3731 }
3732 
3733 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3734 {
3735     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3736     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3737 
3738     /* We really shouldn't get this far without anything to unplug */
3739     g_assert(ds);
3740 
3741     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3742     qdev_unrealize(dev);
3743     spapr_pending_dimm_unplugs_remove(spapr, ds);
3744 }
3745 
3746 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3747                                         DeviceState *dev, Error **errp)
3748 {
3749     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3750     PCDIMMDevice *dimm = PC_DIMM(dev);
3751     uint32_t nr_lmbs;
3752     uint64_t size, addr_start, addr;
3753     int i;
3754     SpaprDrc *drc;
3755 
3756     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3757         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3758         return;
3759     }
3760 
3761     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3762     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3763 
3764     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3765                                           &error_abort);
3766 
3767     /*
3768      * An existing pending dimm state for this DIMM means that there is an
3769      * unplug operation in progress, waiting for the spapr_lmb_release
3770      * callback to complete the job (BQL can't cover that far). In this case,
3771      * bail out to avoid detaching DRCs that were already released.
3772      */
3773     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3774         error_setg(errp, "Memory unplug already in progress for device %s",
3775                    dev->id);
3776         return;
3777     }
3778 
3779     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
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         spapr_drc_unplug_request(drc);
3788         addr += SPAPR_MEMORY_BLOCK_SIZE;
3789     }
3790 
3791     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3792                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3793     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3794                                               nr_lmbs, spapr_drc_index(drc));
3795 }
3796 
3797 /* Callback to be called during DRC release. */
3798 void spapr_core_release(DeviceState *dev)
3799 {
3800     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3801 
3802     /* Call the unplug handler chain. This can never fail. */
3803     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3804     object_unparent(OBJECT(dev));
3805 }
3806 
3807 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3808 {
3809     MachineState *ms = MACHINE(hotplug_dev);
3810     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3811     CPUCore *cc = CPU_CORE(dev);
3812     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3813 
3814     if (smc->pre_2_10_has_unused_icps) {
3815         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3816         int i;
3817 
3818         for (i = 0; i < cc->nr_threads; i++) {
3819             CPUState *cs = CPU(sc->threads[i]);
3820 
3821             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3822         }
3823     }
3824 
3825     assert(core_slot);
3826     core_slot->cpu = NULL;
3827     qdev_unrealize(dev);
3828 }
3829 
3830 static
3831 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3832                                Error **errp)
3833 {
3834     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3835     int index;
3836     SpaprDrc *drc;
3837     CPUCore *cc = CPU_CORE(dev);
3838 
3839     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3840         error_setg(errp, "Unable to find CPU core with core-id: %d",
3841                    cc->core_id);
3842         return;
3843     }
3844     if (index == 0) {
3845         error_setg(errp, "Boot CPU core may not be unplugged");
3846         return;
3847     }
3848 
3849     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3850                           spapr_vcpu_id(spapr, cc->core_id));
3851     g_assert(drc);
3852 
3853     if (!spapr_drc_unplug_requested(drc)) {
3854         spapr_drc_unplug_request(drc);
3855     }
3856 
3857     /*
3858      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3859      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3860      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3861      * attempt (e.g. the kernel will refuse to remove the last online
3862      * CPU), we will never attempt it again because unplug_requested
3863      * will still be 'true' in that case.
3864      */
3865     spapr_hotplug_req_remove_by_index(drc);
3866 }
3867 
3868 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3869                            void *fdt, int *fdt_start_offset, Error **errp)
3870 {
3871     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3872     CPUState *cs = CPU(core->threads[0]);
3873     PowerPCCPU *cpu = POWERPC_CPU(cs);
3874     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3875     int id = spapr_get_vcpu_id(cpu);
3876     g_autofree char *nodename = NULL;
3877     int offset;
3878 
3879     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3880     offset = fdt_add_subnode(fdt, 0, nodename);
3881 
3882     spapr_dt_cpu(cs, fdt, offset, spapr);
3883 
3884     /*
3885      * spapr_dt_cpu() does not fill the 'name' property in the
3886      * CPU node. The function is called during boot process, before
3887      * and after CAS, and overwriting the 'name' property written
3888      * by SLOF is not allowed.
3889      *
3890      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3891      * CPUs more compatible with the coldplugged ones, which have
3892      * the 'name' property. Linux Kernel also relies on this
3893      * property to identify CPU nodes.
3894      */
3895     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3896 
3897     *fdt_start_offset = offset;
3898     return 0;
3899 }
3900 
3901 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3902 {
3903     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3904     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3905     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3906     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3907     CPUCore *cc = CPU_CORE(dev);
3908     CPUState *cs;
3909     SpaprDrc *drc;
3910     CPUArchId *core_slot;
3911     int index;
3912     bool hotplugged = spapr_drc_hotplugged(dev);
3913     int i;
3914 
3915     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3916     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3917 
3918     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3919                           spapr_vcpu_id(spapr, cc->core_id));
3920 
3921     g_assert(drc || !mc->has_hotpluggable_cpus);
3922 
3923     if (drc) {
3924         /*
3925          * spapr_core_pre_plug() already buys us this is a brand new
3926          * core being plugged into a free slot. Nothing should already
3927          * be attached to the corresponding DRC.
3928          */
3929         spapr_drc_attach(drc, dev);
3930 
3931         if (hotplugged) {
3932             /*
3933              * Send hotplug notification interrupt to the guest only
3934              * in case of hotplugged CPUs.
3935              */
3936             spapr_hotplug_req_add_by_index(drc);
3937         } else {
3938             spapr_drc_reset(drc);
3939         }
3940     }
3941 
3942     core_slot->cpu = OBJECT(dev);
3943 
3944     /*
3945      * Set compatibility mode to match the boot CPU, which was either set
3946      * by the machine reset code or by CAS. This really shouldn't fail at
3947      * this point.
3948      */
3949     if (hotplugged) {
3950         for (i = 0; i < cc->nr_threads; i++) {
3951             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3952                            &error_abort);
3953         }
3954     }
3955 
3956     if (smc->pre_2_10_has_unused_icps) {
3957         for (i = 0; i < cc->nr_threads; i++) {
3958             cs = CPU(core->threads[i]);
3959             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3960         }
3961     }
3962 }
3963 
3964 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3965                                 Error **errp)
3966 {
3967     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3968     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3969     CPUCore *cc = CPU_CORE(dev);
3970     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3971     const char *type = object_get_typename(OBJECT(dev));
3972     CPUArchId *core_slot;
3973     int index;
3974     unsigned int smp_threads = machine->smp.threads;
3975 
3976     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3977         error_setg(errp, "CPU hotplug not supported for this machine");
3978         return;
3979     }
3980 
3981     if (strcmp(base_core_type, type)) {
3982         error_setg(errp, "CPU core type should be %s", base_core_type);
3983         return;
3984     }
3985 
3986     if (cc->core_id % smp_threads) {
3987         error_setg(errp, "invalid core id %d", cc->core_id);
3988         return;
3989     }
3990 
3991     /*
3992      * In general we should have homogeneous threads-per-core, but old
3993      * (pre hotplug support) machine types allow the last core to have
3994      * reduced threads as a compatibility hack for when we allowed
3995      * total vcpus not a multiple of threads-per-core.
3996      */
3997     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3998         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3999                    smp_threads);
4000         return;
4001     }
4002 
4003     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4004     if (!core_slot) {
4005         error_setg(errp, "core id %d out of range", cc->core_id);
4006         return;
4007     }
4008 
4009     if (core_slot->cpu) {
4010         error_setg(errp, "core %d already populated", cc->core_id);
4011         return;
4012     }
4013 
4014     numa_cpu_pre_plug(core_slot, dev, errp);
4015 }
4016 
4017 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4018                           void *fdt, int *fdt_start_offset, Error **errp)
4019 {
4020     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4021     int intc_phandle;
4022 
4023     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4024     if (intc_phandle <= 0) {
4025         return -1;
4026     }
4027 
4028     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4029         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4030         return -1;
4031     }
4032 
4033     /* generally SLOF creates these, for hotplug it's up to QEMU */
4034     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4035 
4036     return 0;
4037 }
4038 
4039 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4040                                Error **errp)
4041 {
4042     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4043     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4044     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4045     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4046     SpaprDrc *drc;
4047 
4048     if (dev->hotplugged && !smc->dr_phb_enabled) {
4049         error_setg(errp, "PHB hotplug not supported for this machine");
4050         return false;
4051     }
4052 
4053     if (sphb->index == (uint32_t)-1) {
4054         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4055         return false;
4056     }
4057 
4058     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4059     if (drc && drc->dev) {
4060         error_setg(errp, "PHB %d already attached", sphb->index);
4061         return false;
4062     }
4063 
4064     /*
4065      * This will check that sphb->index doesn't exceed the maximum number of
4066      * PHBs for the current machine type.
4067      */
4068     return
4069         smc->phb_placement(spapr, sphb->index,
4070                            &sphb->buid, &sphb->io_win_addr,
4071                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4072                            windows_supported, sphb->dma_liobn,
4073                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4074                            errp);
4075 }
4076 
4077 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4078 {
4079     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4080     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4081     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4082     SpaprDrc *drc;
4083     bool hotplugged = spapr_drc_hotplugged(dev);
4084 
4085     if (!smc->dr_phb_enabled) {
4086         return;
4087     }
4088 
4089     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4090     /* hotplug hooks should check it's enabled before getting this far */
4091     assert(drc);
4092 
4093     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4094     spapr_drc_attach(drc, dev);
4095 
4096     if (hotplugged) {
4097         spapr_hotplug_req_add_by_index(drc);
4098     } else {
4099         spapr_drc_reset(drc);
4100     }
4101 }
4102 
4103 void spapr_phb_release(DeviceState *dev)
4104 {
4105     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4106 
4107     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4108     object_unparent(OBJECT(dev));
4109 }
4110 
4111 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4112 {
4113     qdev_unrealize(dev);
4114 }
4115 
4116 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4117                                      DeviceState *dev, Error **errp)
4118 {
4119     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4120     SpaprDrc *drc;
4121 
4122     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4123     assert(drc);
4124 
4125     if (!spapr_drc_unplug_requested(drc)) {
4126         spapr_drc_unplug_request(drc);
4127         spapr_hotplug_req_remove_by_index(drc);
4128     } else {
4129         error_setg(errp,
4130                    "PCI Host Bridge unplug already in progress for device %s",
4131                    dev->id);
4132     }
4133 }
4134 
4135 static
4136 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4137                               Error **errp)
4138 {
4139     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4140 
4141     if (spapr->tpm_proxy != NULL) {
4142         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4143         return false;
4144     }
4145 
4146     return true;
4147 }
4148 
4149 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4150 {
4151     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4152     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4153 
4154     /* Already checked in spapr_tpm_proxy_pre_plug() */
4155     g_assert(spapr->tpm_proxy == NULL);
4156 
4157     spapr->tpm_proxy = tpm_proxy;
4158 }
4159 
4160 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4161 {
4162     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4163 
4164     qdev_unrealize(dev);
4165     object_unparent(OBJECT(dev));
4166     spapr->tpm_proxy = NULL;
4167 }
4168 
4169 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4170                                       DeviceState *dev, Error **errp)
4171 {
4172     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4173         spapr_memory_plug(hotplug_dev, dev);
4174     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4175         spapr_core_plug(hotplug_dev, dev);
4176     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4177         spapr_phb_plug(hotplug_dev, dev);
4178     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4179         spapr_tpm_proxy_plug(hotplug_dev, dev);
4180     }
4181 }
4182 
4183 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4184                                         DeviceState *dev, Error **errp)
4185 {
4186     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4187         spapr_memory_unplug(hotplug_dev, dev);
4188     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4189         spapr_core_unplug(hotplug_dev, dev);
4190     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4191         spapr_phb_unplug(hotplug_dev, dev);
4192     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4193         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4194     }
4195 }
4196 
4197 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4198 {
4199     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4200         /*
4201          * CAS will process all pending unplug requests.
4202          *
4203          * HACK: a guest could theoretically have cleared all bits in OV5,
4204          * but none of the guests we care for do.
4205          */
4206         spapr_ovec_empty(spapr->ov5_cas);
4207 }
4208 
4209 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4210                                                 DeviceState *dev, Error **errp)
4211 {
4212     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4213     MachineClass *mc = MACHINE_GET_CLASS(sms);
4214     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4215 
4216     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4217         if (spapr_memory_hot_unplug_supported(sms)) {
4218             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4219         } else {
4220             error_setg(errp, "Memory hot unplug not supported for this guest");
4221         }
4222     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4223         if (!mc->has_hotpluggable_cpus) {
4224             error_setg(errp, "CPU hot unplug not supported on this machine");
4225             return;
4226         }
4227         spapr_core_unplug_request(hotplug_dev, dev, errp);
4228     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4229         if (!smc->dr_phb_enabled) {
4230             error_setg(errp, "PHB hot unplug not supported on this machine");
4231             return;
4232         }
4233         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4234     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4235         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4236     }
4237 }
4238 
4239 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4240                                           DeviceState *dev, Error **errp)
4241 {
4242     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4243         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4244     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4245         spapr_core_pre_plug(hotplug_dev, dev, errp);
4246     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4247         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4248     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4249         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4250     }
4251 }
4252 
4253 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4254                                                  DeviceState *dev)
4255 {
4256     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4257         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4258         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4259         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4260         return HOTPLUG_HANDLER(machine);
4261     }
4262     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4263         PCIDevice *pcidev = PCI_DEVICE(dev);
4264         PCIBus *root = pci_device_root_bus(pcidev);
4265         SpaprPhbState *phb =
4266             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4267                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4268 
4269         if (phb) {
4270             return HOTPLUG_HANDLER(phb);
4271         }
4272     }
4273     return NULL;
4274 }
4275 
4276 static CpuInstanceProperties
4277 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4278 {
4279     CPUArchId *core_slot;
4280     MachineClass *mc = MACHINE_GET_CLASS(machine);
4281 
4282     /* make sure possible_cpu are intialized */
4283     mc->possible_cpu_arch_ids(machine);
4284     /* get CPU core slot containing thread that matches cpu_index */
4285     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4286     assert(core_slot);
4287     return core_slot->props;
4288 }
4289 
4290 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4291 {
4292     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4293 }
4294 
4295 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4296 {
4297     int i;
4298     unsigned int smp_threads = machine->smp.threads;
4299     unsigned int smp_cpus = machine->smp.cpus;
4300     const char *core_type;
4301     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4302     MachineClass *mc = MACHINE_GET_CLASS(machine);
4303 
4304     if (!mc->has_hotpluggable_cpus) {
4305         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4306     }
4307     if (machine->possible_cpus) {
4308         assert(machine->possible_cpus->len == spapr_max_cores);
4309         return machine->possible_cpus;
4310     }
4311 
4312     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4313     if (!core_type) {
4314         error_report("Unable to find sPAPR CPU Core definition");
4315         exit(1);
4316     }
4317 
4318     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4319                              sizeof(CPUArchId) * spapr_max_cores);
4320     machine->possible_cpus->len = spapr_max_cores;
4321     for (i = 0; i < machine->possible_cpus->len; i++) {
4322         int core_id = i * smp_threads;
4323 
4324         machine->possible_cpus->cpus[i].type = core_type;
4325         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4326         machine->possible_cpus->cpus[i].arch_id = core_id;
4327         machine->possible_cpus->cpus[i].props.has_core_id = true;
4328         machine->possible_cpus->cpus[i].props.core_id = core_id;
4329     }
4330     return machine->possible_cpus;
4331 }
4332 
4333 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4334                                 uint64_t *buid, hwaddr *pio,
4335                                 hwaddr *mmio32, hwaddr *mmio64,
4336                                 unsigned n_dma, uint32_t *liobns,
4337                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4338 {
4339     /*
4340      * New-style PHB window placement.
4341      *
4342      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4343      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4344      * windows.
4345      *
4346      * Some guest kernels can't work with MMIO windows above 1<<46
4347      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4348      *
4349      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4350      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4351      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4352      * 1TiB 64-bit MMIO windows for each PHB.
4353      */
4354     const uint64_t base_buid = 0x800000020000000ULL;
4355     int i;
4356 
4357     /* Sanity check natural alignments */
4358     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4359     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4360     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4361     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4362     /* Sanity check bounds */
4363     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4364                       SPAPR_PCI_MEM32_WIN_SIZE);
4365     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4366                       SPAPR_PCI_MEM64_WIN_SIZE);
4367 
4368     if (index >= SPAPR_MAX_PHBS) {
4369         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4370                    SPAPR_MAX_PHBS - 1);
4371         return false;
4372     }
4373 
4374     *buid = base_buid + index;
4375     for (i = 0; i < n_dma; ++i) {
4376         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4377     }
4378 
4379     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4380     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4381     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4382 
4383     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4384     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4385     return true;
4386 }
4387 
4388 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4389 {
4390     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4391 
4392     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4393 }
4394 
4395 static void spapr_ics_resend(XICSFabric *dev)
4396 {
4397     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4398 
4399     ics_resend(spapr->ics);
4400 }
4401 
4402 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4403 {
4404     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4405 
4406     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4407 }
4408 
4409 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4410                                  Monitor *mon)
4411 {
4412     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4413 
4414     spapr_irq_print_info(spapr, mon);
4415     monitor_printf(mon, "irqchip: %s\n",
4416                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4417 }
4418 
4419 /*
4420  * This is a XIVE only operation
4421  */
4422 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4423                            uint8_t nvt_blk, uint32_t nvt_idx,
4424                            bool cam_ignore, uint8_t priority,
4425                            uint32_t logic_serv, XiveTCTXMatch *match)
4426 {
4427     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4428     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4429     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4430     int count;
4431 
4432     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4433                            priority, logic_serv, match);
4434     if (count < 0) {
4435         return count;
4436     }
4437 
4438     /*
4439      * When we implement the save and restore of the thread interrupt
4440      * contexts in the enter/exit CPU handlers of the machine and the
4441      * escalations in QEMU, we should be able to handle non dispatched
4442      * vCPUs.
4443      *
4444      * Until this is done, the sPAPR machine should find at least one
4445      * matching context always.
4446      */
4447     if (count == 0) {
4448         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4449                       nvt_blk, nvt_idx);
4450     }
4451 
4452     return count;
4453 }
4454 
4455 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4456 {
4457     return cpu->vcpu_id;
4458 }
4459 
4460 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4461 {
4462     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4463     MachineState *ms = MACHINE(spapr);
4464     int vcpu_id;
4465 
4466     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4467 
4468     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4469         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4470         error_append_hint(errp, "Adjust the number of cpus to %d "
4471                           "or try to raise the number of threads per core\n",
4472                           vcpu_id * ms->smp.threads / spapr->vsmt);
4473         return false;
4474     }
4475 
4476     cpu->vcpu_id = vcpu_id;
4477     return true;
4478 }
4479 
4480 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4481 {
4482     CPUState *cs;
4483 
4484     CPU_FOREACH(cs) {
4485         PowerPCCPU *cpu = POWERPC_CPU(cs);
4486 
4487         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4488             return cpu;
4489         }
4490     }
4491 
4492     return NULL;
4493 }
4494 
4495 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4496 {
4497     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4498 
4499     return spapr_cpu->in_nested;
4500 }
4501 
4502 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4503 {
4504     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4505 
4506     /* These are only called by TCG, KVM maintains dispatch state */
4507 
4508     spapr_cpu->prod = false;
4509     if (spapr_cpu->vpa_addr) {
4510         CPUState *cs = CPU(cpu);
4511         uint32_t dispatch;
4512 
4513         dispatch = ldl_be_phys(cs->as,
4514                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4515         dispatch++;
4516         if ((dispatch & 1) != 0) {
4517             qemu_log_mask(LOG_GUEST_ERROR,
4518                           "VPA: incorrect dispatch counter value for "
4519                           "dispatched partition %u, correcting.\n", dispatch);
4520             dispatch++;
4521         }
4522         stl_be_phys(cs->as,
4523                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4524     }
4525 }
4526 
4527 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4528 {
4529     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4530 
4531     if (spapr_cpu->vpa_addr) {
4532         CPUState *cs = CPU(cpu);
4533         uint32_t dispatch;
4534 
4535         dispatch = ldl_be_phys(cs->as,
4536                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4537         dispatch++;
4538         if ((dispatch & 1) != 1) {
4539             qemu_log_mask(LOG_GUEST_ERROR,
4540                           "VPA: incorrect dispatch counter value for "
4541                           "preempted partition %u, correcting.\n", dispatch);
4542             dispatch++;
4543         }
4544         stl_be_phys(cs->as,
4545                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4546     }
4547 }
4548 
4549 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4550 {
4551     MachineClass *mc = MACHINE_CLASS(oc);
4552     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4553     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4554     NMIClass *nc = NMI_CLASS(oc);
4555     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4556     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4557     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4558     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4559     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4560     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4561 
4562     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4563     mc->ignore_boot_device_suffixes = true;
4564 
4565     /*
4566      * We set up the default / latest behaviour here.  The class_init
4567      * functions for the specific versioned machine types can override
4568      * these details for backwards compatibility
4569      */
4570     mc->init = spapr_machine_init;
4571     mc->reset = spapr_machine_reset;
4572     mc->block_default_type = IF_SCSI;
4573 
4574     /*
4575      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4576      * should be limited by the host capability instead of hardcoded.
4577      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4578      * guests are welcome to have as many CPUs as the host are capable
4579      * of emulate.
4580      */
4581     mc->max_cpus = INT32_MAX;
4582 
4583     mc->no_parallel = 1;
4584     mc->default_boot_order = "";
4585     mc->default_ram_size = 512 * MiB;
4586     mc->default_ram_id = "ppc_spapr.ram";
4587     mc->default_display = "std";
4588     mc->kvm_type = spapr_kvm_type;
4589     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4590     mc->pci_allow_0_address = true;
4591     assert(!mc->get_hotplug_handler);
4592     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4593     hc->pre_plug = spapr_machine_device_pre_plug;
4594     hc->plug = spapr_machine_device_plug;
4595     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4596     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4597     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4598     hc->unplug_request = spapr_machine_device_unplug_request;
4599     hc->unplug = spapr_machine_device_unplug;
4600 
4601     smc->dr_lmb_enabled = true;
4602     smc->update_dt_enabled = true;
4603     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4604     mc->has_hotpluggable_cpus = true;
4605     mc->nvdimm_supported = true;
4606     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4607     fwc->get_dev_path = spapr_get_fw_dev_path;
4608     nc->nmi_monitor_handler = spapr_nmi;
4609     smc->phb_placement = spapr_phb_placement;
4610     vhc->cpu_in_nested = spapr_cpu_in_nested;
4611     vhc->deliver_hv_excp = spapr_exit_nested;
4612     vhc->hypercall = emulate_spapr_hypercall;
4613     vhc->hpt_mask = spapr_hpt_mask;
4614     vhc->map_hptes = spapr_map_hptes;
4615     vhc->unmap_hptes = spapr_unmap_hptes;
4616     vhc->hpte_set_c = spapr_hpte_set_c;
4617     vhc->hpte_set_r = spapr_hpte_set_r;
4618     vhc->get_pate = spapr_get_pate;
4619     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4620     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4621     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4622     xic->ics_get = spapr_ics_get;
4623     xic->ics_resend = spapr_ics_resend;
4624     xic->icp_get = spapr_icp_get;
4625     ispc->print_info = spapr_pic_print_info;
4626     /* Force NUMA node memory size to be a multiple of
4627      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4628      * in which LMBs are represented and hot-added
4629      */
4630     mc->numa_mem_align_shift = 28;
4631     mc->auto_enable_numa = true;
4632 
4633     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4634     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4635     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4636     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4637     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4638     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4639     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4640     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4641     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4642     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4643     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4644     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4645     spapr_caps_add_properties(smc);
4646     smc->irq = &spapr_irq_dual;
4647     smc->dr_phb_enabled = true;
4648     smc->linux_pci_probe = true;
4649     smc->smp_threads_vsmt = true;
4650     smc->nr_xirqs = SPAPR_NR_XIRQS;
4651     xfc->match_nvt = spapr_match_nvt;
4652     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4653     vmc->quiesce = spapr_vof_quiesce;
4654     vmc->setprop = spapr_vof_setprop;
4655 }
4656 
4657 static const TypeInfo spapr_machine_info = {
4658     .name          = TYPE_SPAPR_MACHINE,
4659     .parent        = TYPE_MACHINE,
4660     .abstract      = true,
4661     .instance_size = sizeof(SpaprMachineState),
4662     .instance_init = spapr_instance_init,
4663     .instance_finalize = spapr_machine_finalizefn,
4664     .class_size    = sizeof(SpaprMachineClass),
4665     .class_init    = spapr_machine_class_init,
4666     .interfaces = (InterfaceInfo[]) {
4667         { TYPE_FW_PATH_PROVIDER },
4668         { TYPE_NMI },
4669         { TYPE_HOTPLUG_HANDLER },
4670         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4671         { TYPE_XICS_FABRIC },
4672         { TYPE_INTERRUPT_STATS_PROVIDER },
4673         { TYPE_XIVE_FABRIC },
4674         { TYPE_VOF_MACHINE_IF },
4675         { }
4676     },
4677 };
4678 
4679 static void spapr_machine_latest_class_options(MachineClass *mc)
4680 {
4681     mc->alias = "pseries";
4682     mc->is_default = true;
4683 }
4684 
4685 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4686     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4687                                                     void *data)      \
4688     {                                                                \
4689         MachineClass *mc = MACHINE_CLASS(oc);                        \
4690         spapr_machine_##suffix##_class_options(mc);                  \
4691         if (latest) {                                                \
4692             spapr_machine_latest_class_options(mc);                  \
4693         }                                                            \
4694     }                                                                \
4695     static const TypeInfo spapr_machine_##suffix##_info = {          \
4696         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4697         .parent = TYPE_SPAPR_MACHINE,                                \
4698         .class_init = spapr_machine_##suffix##_class_init,           \
4699     };                                                               \
4700     static void spapr_machine_register_##suffix(void)                \
4701     {                                                                \
4702         type_register(&spapr_machine_##suffix##_info);               \
4703     }                                                                \
4704     type_init(spapr_machine_register_##suffix)
4705 
4706 /*
4707  * pseries-7.1
4708  */
4709 static void spapr_machine_7_1_class_options(MachineClass *mc)
4710 {
4711     /* Defaults for the latest behaviour inherited from the base class */
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(7_1, "7.1", true);
4715 
4716 /*
4717  * pseries-7.0
4718  */
4719 static void spapr_machine_7_0_class_options(MachineClass *mc)
4720 {
4721     spapr_machine_7_1_class_options(mc);
4722     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4723 }
4724 
4725 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4726 
4727 /*
4728  * pseries-6.2
4729  */
4730 static void spapr_machine_6_2_class_options(MachineClass *mc)
4731 {
4732     spapr_machine_7_0_class_options(mc);
4733     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4734 }
4735 
4736 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4737 
4738 /*
4739  * pseries-6.1
4740  */
4741 static void spapr_machine_6_1_class_options(MachineClass *mc)
4742 {
4743     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4744 
4745     spapr_machine_6_2_class_options(mc);
4746     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4747     smc->pre_6_2_numa_affinity = true;
4748     mc->smp_props.prefer_sockets = true;
4749 }
4750 
4751 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4752 
4753 /*
4754  * pseries-6.0
4755  */
4756 static void spapr_machine_6_0_class_options(MachineClass *mc)
4757 {
4758     spapr_machine_6_1_class_options(mc);
4759     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4760 }
4761 
4762 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4763 
4764 /*
4765  * pseries-5.2
4766  */
4767 static void spapr_machine_5_2_class_options(MachineClass *mc)
4768 {
4769     spapr_machine_6_0_class_options(mc);
4770     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4771 }
4772 
4773 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4774 
4775 /*
4776  * pseries-5.1
4777  */
4778 static void spapr_machine_5_1_class_options(MachineClass *mc)
4779 {
4780     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4781 
4782     spapr_machine_5_2_class_options(mc);
4783     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4784     smc->pre_5_2_numa_associativity = true;
4785 }
4786 
4787 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4788 
4789 /*
4790  * pseries-5.0
4791  */
4792 static void spapr_machine_5_0_class_options(MachineClass *mc)
4793 {
4794     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4795     static GlobalProperty compat[] = {
4796         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4797     };
4798 
4799     spapr_machine_5_1_class_options(mc);
4800     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4801     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4802     mc->numa_mem_supported = true;
4803     smc->pre_5_1_assoc_refpoints = true;
4804 }
4805 
4806 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4807 
4808 /*
4809  * pseries-4.2
4810  */
4811 static void spapr_machine_4_2_class_options(MachineClass *mc)
4812 {
4813     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4814 
4815     spapr_machine_5_0_class_options(mc);
4816     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4817     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4818     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4819     smc->rma_limit = 16 * GiB;
4820     mc->nvdimm_supported = false;
4821 }
4822 
4823 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4824 
4825 /*
4826  * pseries-4.1
4827  */
4828 static void spapr_machine_4_1_class_options(MachineClass *mc)
4829 {
4830     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4831     static GlobalProperty compat[] = {
4832         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4833         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4834     };
4835 
4836     spapr_machine_4_2_class_options(mc);
4837     smc->linux_pci_probe = false;
4838     smc->smp_threads_vsmt = false;
4839     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4840     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4841 }
4842 
4843 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4844 
4845 /*
4846  * pseries-4.0
4847  */
4848 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4849                               uint64_t *buid, hwaddr *pio,
4850                               hwaddr *mmio32, hwaddr *mmio64,
4851                               unsigned n_dma, uint32_t *liobns,
4852                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4853 {
4854     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4855                              liobns, nv2gpa, nv2atsd, errp)) {
4856         return false;
4857     }
4858 
4859     *nv2gpa = 0;
4860     *nv2atsd = 0;
4861     return true;
4862 }
4863 static void spapr_machine_4_0_class_options(MachineClass *mc)
4864 {
4865     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4866 
4867     spapr_machine_4_1_class_options(mc);
4868     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4869     smc->phb_placement = phb_placement_4_0;
4870     smc->irq = &spapr_irq_xics;
4871     smc->pre_4_1_migration = true;
4872 }
4873 
4874 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4875 
4876 /*
4877  * pseries-3.1
4878  */
4879 static void spapr_machine_3_1_class_options(MachineClass *mc)
4880 {
4881     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4882 
4883     spapr_machine_4_0_class_options(mc);
4884     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4885 
4886     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4887     smc->update_dt_enabled = false;
4888     smc->dr_phb_enabled = false;
4889     smc->broken_host_serial_model = true;
4890     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4891     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4892     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4893     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4894 }
4895 
4896 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4897 
4898 /*
4899  * pseries-3.0
4900  */
4901 
4902 static void spapr_machine_3_0_class_options(MachineClass *mc)
4903 {
4904     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4905 
4906     spapr_machine_3_1_class_options(mc);
4907     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4908 
4909     smc->legacy_irq_allocation = true;
4910     smc->nr_xirqs = 0x400;
4911     smc->irq = &spapr_irq_xics_legacy;
4912 }
4913 
4914 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4915 
4916 /*
4917  * pseries-2.12
4918  */
4919 static void spapr_machine_2_12_class_options(MachineClass *mc)
4920 {
4921     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4922     static GlobalProperty compat[] = {
4923         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4924         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4925     };
4926 
4927     spapr_machine_3_0_class_options(mc);
4928     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4929     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4930 
4931     /* We depend on kvm_enabled() to choose a default value for the
4932      * hpt-max-page-size capability. Of course we can't do it here
4933      * because this is too early and the HW accelerator isn't initialzed
4934      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4935      */
4936     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4937 }
4938 
4939 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4940 
4941 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4942 {
4943     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4944 
4945     spapr_machine_2_12_class_options(mc);
4946     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4947     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4948     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4949 }
4950 
4951 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4952 
4953 /*
4954  * pseries-2.11
4955  */
4956 
4957 static void spapr_machine_2_11_class_options(MachineClass *mc)
4958 {
4959     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4960 
4961     spapr_machine_2_12_class_options(mc);
4962     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4963     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4964 }
4965 
4966 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4967 
4968 /*
4969  * pseries-2.10
4970  */
4971 
4972 static void spapr_machine_2_10_class_options(MachineClass *mc)
4973 {
4974     spapr_machine_2_11_class_options(mc);
4975     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4976 }
4977 
4978 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4979 
4980 /*
4981  * pseries-2.9
4982  */
4983 
4984 static void spapr_machine_2_9_class_options(MachineClass *mc)
4985 {
4986     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4987     static GlobalProperty compat[] = {
4988         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4989     };
4990 
4991     spapr_machine_2_10_class_options(mc);
4992     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4993     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4994     smc->pre_2_10_has_unused_icps = true;
4995     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4996 }
4997 
4998 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4999 
5000 /*
5001  * pseries-2.8
5002  */
5003 
5004 static void spapr_machine_2_8_class_options(MachineClass *mc)
5005 {
5006     static GlobalProperty compat[] = {
5007         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5008     };
5009 
5010     spapr_machine_2_9_class_options(mc);
5011     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5012     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5013     mc->numa_mem_align_shift = 23;
5014 }
5015 
5016 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5017 
5018 /*
5019  * pseries-2.7
5020  */
5021 
5022 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5023                               uint64_t *buid, hwaddr *pio,
5024                               hwaddr *mmio32, hwaddr *mmio64,
5025                               unsigned n_dma, uint32_t *liobns,
5026                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
5027 {
5028     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5029     const uint64_t base_buid = 0x800000020000000ULL;
5030     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5031     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5032     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5033     const uint32_t max_index = 255;
5034     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5035 
5036     uint64_t ram_top = MACHINE(spapr)->ram_size;
5037     hwaddr phb0_base, phb_base;
5038     int i;
5039 
5040     /* Do we have device memory? */
5041     if (MACHINE(spapr)->maxram_size > ram_top) {
5042         /* Can't just use maxram_size, because there may be an
5043          * alignment gap between normal and device memory regions
5044          */
5045         ram_top = MACHINE(spapr)->device_memory->base +
5046             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5047     }
5048 
5049     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5050 
5051     if (index > max_index) {
5052         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5053                    max_index);
5054         return false;
5055     }
5056 
5057     *buid = base_buid + index;
5058     for (i = 0; i < n_dma; ++i) {
5059         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5060     }
5061 
5062     phb_base = phb0_base + index * phb_spacing;
5063     *pio = phb_base + pio_offset;
5064     *mmio32 = phb_base + mmio_offset;
5065     /*
5066      * We don't set the 64-bit MMIO window, relying on the PHB's
5067      * fallback behaviour of automatically splitting a large "32-bit"
5068      * window into contiguous 32-bit and 64-bit windows
5069      */
5070 
5071     *nv2gpa = 0;
5072     *nv2atsd = 0;
5073     return true;
5074 }
5075 
5076 static void spapr_machine_2_7_class_options(MachineClass *mc)
5077 {
5078     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5079     static GlobalProperty compat[] = {
5080         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5081         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5082         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5083         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5084     };
5085 
5086     spapr_machine_2_8_class_options(mc);
5087     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5088     mc->default_machine_opts = "modern-hotplug-events=off";
5089     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5090     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5091     smc->phb_placement = phb_placement_2_7;
5092 }
5093 
5094 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5095 
5096 /*
5097  * pseries-2.6
5098  */
5099 
5100 static void spapr_machine_2_6_class_options(MachineClass *mc)
5101 {
5102     static GlobalProperty compat[] = {
5103         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5104     };
5105 
5106     spapr_machine_2_7_class_options(mc);
5107     mc->has_hotpluggable_cpus = false;
5108     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5109     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5110 }
5111 
5112 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5113 
5114 /*
5115  * pseries-2.5
5116  */
5117 
5118 static void spapr_machine_2_5_class_options(MachineClass *mc)
5119 {
5120     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5121     static GlobalProperty compat[] = {
5122         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5123     };
5124 
5125     spapr_machine_2_6_class_options(mc);
5126     smc->use_ohci_by_default = true;
5127     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5128     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5129 }
5130 
5131 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5132 
5133 /*
5134  * pseries-2.4
5135  */
5136 
5137 static void spapr_machine_2_4_class_options(MachineClass *mc)
5138 {
5139     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5140 
5141     spapr_machine_2_5_class_options(mc);
5142     smc->dr_lmb_enabled = false;
5143     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5144 }
5145 
5146 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5147 
5148 /*
5149  * pseries-2.3
5150  */
5151 
5152 static void spapr_machine_2_3_class_options(MachineClass *mc)
5153 {
5154     static GlobalProperty compat[] = {
5155         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5156     };
5157     spapr_machine_2_4_class_options(mc);
5158     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5159     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5160 }
5161 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5162 
5163 /*
5164  * pseries-2.2
5165  */
5166 
5167 static void spapr_machine_2_2_class_options(MachineClass *mc)
5168 {
5169     static GlobalProperty compat[] = {
5170         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5171     };
5172 
5173     spapr_machine_2_3_class_options(mc);
5174     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5175     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5176     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5177 }
5178 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5179 
5180 /*
5181  * pseries-2.1
5182  */
5183 
5184 static void spapr_machine_2_1_class_options(MachineClass *mc)
5185 {
5186     spapr_machine_2_2_class_options(mc);
5187     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5188 }
5189 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5190 
5191 static void spapr_machine_register_types(void)
5192 {
5193     type_register_static(&spapr_machine_info);
5194 }
5195 
5196 type_init(spapr_machine_register_types)
5197