xref: /qemu/hw/ppc/spapr.c (revision f32d4ab41c467c0e4ddcaff37e4f2af57bcaad8c)
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-common.h"
29 #include "qemu/datadir.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         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1022         size_t cb = 0;
1023         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 (boot_menu) {
1048             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
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         g_free(stdout_path);
1092         g_free(bootlist);
1093     }
1094 
1095     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1096 }
1097 
1098 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1099 {
1100     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1101      * KVM to work under pHyp with some guest co-operation */
1102     int hypervisor;
1103     uint8_t hypercall[16];
1104 
1105     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1106     /* indicate KVM hypercall interface */
1107     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1108     if (kvmppc_has_cap_fixup_hcalls()) {
1109         /*
1110          * Older KVM versions with older guest kernels were broken
1111          * with the magic page, don't allow the guest to map it.
1112          */
1113         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1114                                   sizeof(hypercall))) {
1115             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1116                              hypercall, sizeof(hypercall)));
1117         }
1118     }
1119 }
1120 
1121 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1122 {
1123     MachineState *machine = MACHINE(spapr);
1124     MachineClass *mc = MACHINE_GET_CLASS(machine);
1125     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1126     uint32_t root_drc_type_mask = 0;
1127     int ret;
1128     void *fdt;
1129     SpaprPhbState *phb;
1130     char *buf;
1131 
1132     fdt = g_malloc0(space);
1133     _FDT((fdt_create_empty_tree(fdt, space)));
1134 
1135     /* Root node */
1136     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1137     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1138     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1139 
1140     /* Guest UUID & Name*/
1141     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1142     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1143     if (qemu_uuid_set) {
1144         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1145     }
1146     g_free(buf);
1147 
1148     if (qemu_get_vm_name()) {
1149         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1150                                 qemu_get_vm_name()));
1151     }
1152 
1153     /* Host Model & Serial Number */
1154     if (spapr->host_model) {
1155         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1156     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1157         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1158         g_free(buf);
1159     }
1160 
1161     if (spapr->host_serial) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1163     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1164         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1165         g_free(buf);
1166     }
1167 
1168     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1169     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1170 
1171     /* /interrupt controller */
1172     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1173 
1174     ret = spapr_dt_memory(spapr, fdt);
1175     if (ret < 0) {
1176         error_report("couldn't setup memory nodes in fdt");
1177         exit(1);
1178     }
1179 
1180     /* /vdevice */
1181     spapr_dt_vdevice(spapr->vio_bus, fdt);
1182 
1183     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1184         ret = spapr_dt_rng(fdt);
1185         if (ret < 0) {
1186             error_report("could not set up rng device in the fdt");
1187             exit(1);
1188         }
1189     }
1190 
1191     QLIST_FOREACH(phb, &spapr->phbs, list) {
1192         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1193         if (ret < 0) {
1194             error_report("couldn't setup PCI devices in fdt");
1195             exit(1);
1196         }
1197     }
1198 
1199     spapr_dt_cpus(fdt, spapr);
1200 
1201     /* ibm,drc-indexes and friends */
1202     if (smc->dr_lmb_enabled) {
1203         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1204     }
1205     if (smc->dr_phb_enabled) {
1206         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1207     }
1208     if (mc->nvdimm_supported) {
1209         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1210     }
1211     if (root_drc_type_mask) {
1212         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1213     }
1214 
1215     if (mc->has_hotpluggable_cpus) {
1216         int offset = fdt_path_offset(fdt, "/cpus");
1217         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1218         if (ret < 0) {
1219             error_report("Couldn't set up CPU DR device tree properties");
1220             exit(1);
1221         }
1222     }
1223 
1224     /* /event-sources */
1225     spapr_dt_events(spapr, fdt);
1226 
1227     /* /rtas */
1228     spapr_dt_rtas(spapr, fdt);
1229 
1230     /* /chosen */
1231     spapr_dt_chosen(spapr, fdt, reset);
1232 
1233     /* /hypervisor */
1234     if (kvm_enabled()) {
1235         spapr_dt_hypervisor(spapr, fdt);
1236     }
1237 
1238     /* Build memory reserve map */
1239     if (reset) {
1240         if (spapr->kernel_size) {
1241             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1242                                   spapr->kernel_size)));
1243         }
1244         if (spapr->initrd_size) {
1245             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1246                                   spapr->initrd_size)));
1247         }
1248     }
1249 
1250     /* NVDIMM devices */
1251     if (mc->nvdimm_supported) {
1252         spapr_dt_persistent_memory(spapr, fdt);
1253     }
1254 
1255     return fdt;
1256 }
1257 
1258 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1259 {
1260     SpaprMachineState *spapr = opaque;
1261 
1262     return (addr & 0x0fffffff) + spapr->kernel_addr;
1263 }
1264 
1265 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1266                                     PowerPCCPU *cpu)
1267 {
1268     CPUPPCState *env = &cpu->env;
1269 
1270     /* The TCG path should also be holding the BQL at this point */
1271     g_assert(qemu_mutex_iothread_locked());
1272 
1273     if (msr_pr) {
1274         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1275         env->gpr[3] = H_PRIVILEGE;
1276     } else {
1277         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1278     }
1279 }
1280 
1281 struct LPCRSyncState {
1282     target_ulong value;
1283     target_ulong mask;
1284 };
1285 
1286 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1287 {
1288     struct LPCRSyncState *s = arg.host_ptr;
1289     PowerPCCPU *cpu = POWERPC_CPU(cs);
1290     CPUPPCState *env = &cpu->env;
1291     target_ulong lpcr;
1292 
1293     cpu_synchronize_state(cs);
1294     lpcr = env->spr[SPR_LPCR];
1295     lpcr &= ~s->mask;
1296     lpcr |= s->value;
1297     ppc_store_lpcr(cpu, lpcr);
1298 }
1299 
1300 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1301 {
1302     CPUState *cs;
1303     struct LPCRSyncState s = {
1304         .value = value,
1305         .mask = mask
1306     };
1307     CPU_FOREACH(cs) {
1308         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1309     }
1310 }
1311 
1312 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1313                            target_ulong lpid, ppc_v3_pate_t *entry)
1314 {
1315     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1316 
1317     assert(lpid == 0);
1318 
1319     /* Copy PATE1:GR into PATE0:HR */
1320     entry->dw0 = spapr->patb_entry & PATE0_HR;
1321     entry->dw1 = spapr->patb_entry;
1322 
1323     return true;
1324 }
1325 
1326 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1327 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1328 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1329 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1330 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1331 
1332 /*
1333  * Get the fd to access the kernel htab, re-opening it if necessary
1334  */
1335 static int get_htab_fd(SpaprMachineState *spapr)
1336 {
1337     Error *local_err = NULL;
1338 
1339     if (spapr->htab_fd >= 0) {
1340         return spapr->htab_fd;
1341     }
1342 
1343     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1344     if (spapr->htab_fd < 0) {
1345         error_report_err(local_err);
1346     }
1347 
1348     return spapr->htab_fd;
1349 }
1350 
1351 void close_htab_fd(SpaprMachineState *spapr)
1352 {
1353     if (spapr->htab_fd >= 0) {
1354         close(spapr->htab_fd);
1355     }
1356     spapr->htab_fd = -1;
1357 }
1358 
1359 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1360 {
1361     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1362 
1363     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1364 }
1365 
1366 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1367 {
1368     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1369 
1370     assert(kvm_enabled());
1371 
1372     if (!spapr->htab) {
1373         return 0;
1374     }
1375 
1376     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1377 }
1378 
1379 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1380                                                 hwaddr ptex, int n)
1381 {
1382     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1383     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1384 
1385     if (!spapr->htab) {
1386         /*
1387          * HTAB is controlled by KVM. Fetch into temporary buffer
1388          */
1389         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1390         kvmppc_read_hptes(hptes, ptex, n);
1391         return hptes;
1392     }
1393 
1394     /*
1395      * HTAB is controlled by QEMU. Just point to the internally
1396      * accessible PTEG.
1397      */
1398     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1399 }
1400 
1401 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1402                               const ppc_hash_pte64_t *hptes,
1403                               hwaddr ptex, int n)
1404 {
1405     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1406 
1407     if (!spapr->htab) {
1408         g_free((void *)hptes);
1409     }
1410 
1411     /* Nothing to do for qemu managed HPT */
1412 }
1413 
1414 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1415                       uint64_t pte0, uint64_t pte1)
1416 {
1417     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1418     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1419 
1420     if (!spapr->htab) {
1421         kvmppc_write_hpte(ptex, pte0, pte1);
1422     } else {
1423         if (pte0 & HPTE64_V_VALID) {
1424             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1425             /*
1426              * When setting valid, we write PTE1 first. This ensures
1427              * proper synchronization with the reading code in
1428              * ppc_hash64_pteg_search()
1429              */
1430             smp_wmb();
1431             stq_p(spapr->htab + offset, pte0);
1432         } else {
1433             stq_p(spapr->htab + offset, pte0);
1434             /*
1435              * When clearing it we set PTE0 first. This ensures proper
1436              * synchronization with the reading code in
1437              * ppc_hash64_pteg_search()
1438              */
1439             smp_wmb();
1440             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1441         }
1442     }
1443 }
1444 
1445 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1446                              uint64_t pte1)
1447 {
1448     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1449     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1450 
1451     if (!spapr->htab) {
1452         /* There should always be a hash table when this is called */
1453         error_report("spapr_hpte_set_c called with no hash table !");
1454         return;
1455     }
1456 
1457     /* The HW performs a non-atomic byte update */
1458     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1459 }
1460 
1461 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1462                              uint64_t pte1)
1463 {
1464     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1465     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1466 
1467     if (!spapr->htab) {
1468         /* There should always be a hash table when this is called */
1469         error_report("spapr_hpte_set_r called with no hash table !");
1470         return;
1471     }
1472 
1473     /* The HW performs a non-atomic byte update */
1474     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1475 }
1476 
1477 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1478 {
1479     int shift;
1480 
1481     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1482      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1483      * that's much more than is needed for Linux guests */
1484     shift = ctz64(pow2ceil(ramsize)) - 7;
1485     shift = MAX(shift, 18); /* Minimum architected size */
1486     shift = MIN(shift, 46); /* Maximum architected size */
1487     return shift;
1488 }
1489 
1490 void spapr_free_hpt(SpaprMachineState *spapr)
1491 {
1492     g_free(spapr->htab);
1493     spapr->htab = NULL;
1494     spapr->htab_shift = 0;
1495     close_htab_fd(spapr);
1496 }
1497 
1498 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1499 {
1500     ERRP_GUARD();
1501     long rc;
1502 
1503     /* Clean up any HPT info from a previous boot */
1504     spapr_free_hpt(spapr);
1505 
1506     rc = kvmppc_reset_htab(shift);
1507 
1508     if (rc == -EOPNOTSUPP) {
1509         error_setg(errp, "HPT not supported in nested guests");
1510         return -EOPNOTSUPP;
1511     }
1512 
1513     if (rc < 0) {
1514         /* kernel-side HPT needed, but couldn't allocate one */
1515         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1516                          shift);
1517         error_append_hint(errp, "Try smaller maxmem?\n");
1518         return -errno;
1519     } else if (rc > 0) {
1520         /* kernel-side HPT allocated */
1521         if (rc != shift) {
1522             error_setg(errp,
1523                        "Requested order %d HPT, but kernel allocated order %ld",
1524                        shift, rc);
1525             error_append_hint(errp, "Try smaller maxmem?\n");
1526             return -ENOSPC;
1527         }
1528 
1529         spapr->htab_shift = shift;
1530         spapr->htab = NULL;
1531     } else {
1532         /* kernel-side HPT not needed, allocate in userspace instead */
1533         size_t size = 1ULL << shift;
1534         int i;
1535 
1536         spapr->htab = qemu_memalign(size, size);
1537         memset(spapr->htab, 0, size);
1538         spapr->htab_shift = shift;
1539 
1540         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1541             DIRTY_HPTE(HPTE(spapr->htab, i));
1542         }
1543     }
1544     /* We're setting up a hash table, so that means we're not radix */
1545     spapr->patb_entry = 0;
1546     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1547     return 0;
1548 }
1549 
1550 void spapr_setup_hpt(SpaprMachineState *spapr)
1551 {
1552     int hpt_shift;
1553 
1554     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1555         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1556     } else {
1557         uint64_t current_ram_size;
1558 
1559         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1560         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1561     }
1562     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1563 
1564     if (kvm_enabled()) {
1565         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1566 
1567         /* Check our RMA fits in the possible VRMA */
1568         if (vrma_limit < spapr->rma_size) {
1569             error_report("Unable to create %" HWADDR_PRIu
1570                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1571                          spapr->rma_size / MiB, vrma_limit / MiB);
1572             exit(EXIT_FAILURE);
1573         }
1574     }
1575 }
1576 
1577 void spapr_check_mmu_mode(bool guest_radix)
1578 {
1579     if (guest_radix) {
1580         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1581             error_report("Guest requested unavailable MMU mode (radix).");
1582             exit(EXIT_FAILURE);
1583         }
1584     } else {
1585         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1586             && !kvmppc_has_cap_mmu_hash_v3()) {
1587             error_report("Guest requested unavailable MMU mode (hash).");
1588             exit(EXIT_FAILURE);
1589         }
1590     }
1591 }
1592 
1593 static void spapr_machine_reset(MachineState *machine)
1594 {
1595     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1596     PowerPCCPU *first_ppc_cpu;
1597     hwaddr fdt_addr;
1598     void *fdt;
1599     int rc;
1600 
1601     pef_kvm_reset(machine->cgs, &error_fatal);
1602     spapr_caps_apply(spapr);
1603 
1604     first_ppc_cpu = POWERPC_CPU(first_cpu);
1605     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1606         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1607                               spapr->max_compat_pvr)) {
1608         /*
1609          * If using KVM with radix mode available, VCPUs can be started
1610          * without a HPT because KVM will start them in radix mode.
1611          * Set the GR bit in PATE so that we know there is no HPT.
1612          */
1613         spapr->patb_entry = PATE1_GR;
1614         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1615     } else {
1616         spapr_setup_hpt(spapr);
1617     }
1618 
1619     qemu_devices_reset();
1620 
1621     spapr_ovec_cleanup(spapr->ov5_cas);
1622     spapr->ov5_cas = spapr_ovec_new();
1623 
1624     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1625 
1626     /*
1627      * This is fixing some of the default configuration of the XIVE
1628      * devices. To be called after the reset of the machine devices.
1629      */
1630     spapr_irq_reset(spapr, &error_fatal);
1631 
1632     /*
1633      * There is no CAS under qtest. Simulate one to please the code that
1634      * depends on spapr->ov5_cas. This is especially needed to test device
1635      * unplug, so we do that before resetting the DRCs.
1636      */
1637     if (qtest_enabled()) {
1638         spapr_ovec_cleanup(spapr->ov5_cas);
1639         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1640     }
1641 
1642     spapr_nvdimm_finish_flushes();
1643 
1644     /* DRC reset may cause a device to be unplugged. This will cause troubles
1645      * if this device is used by another device (eg, a running vhost backend
1646      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1647      * situations, we reset DRCs after all devices have been reset.
1648      */
1649     spapr_drc_reset_all(spapr);
1650 
1651     spapr_clear_pending_events(spapr);
1652 
1653     /*
1654      * We place the device tree just below either the top of the RMA,
1655      * or just below 2GB, whichever is lower, so that it can be
1656      * processed with 32-bit real mode code if necessary
1657      */
1658     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1659 
1660     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1661     if (spapr->vof) {
1662         spapr_vof_reset(spapr, fdt, &error_fatal);
1663         /*
1664          * Do not pack the FDT as the client may change properties.
1665          * VOF client does not expect the FDT so we do not load it to the VM.
1666          */
1667     } else {
1668         rc = fdt_pack(fdt);
1669         /* Should only fail if we've built a corrupted tree */
1670         assert(rc == 0);
1671 
1672         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1673                                   0, fdt_addr, 0);
1674         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1675     }
1676     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1677 
1678     g_free(spapr->fdt_blob);
1679     spapr->fdt_size = fdt_totalsize(fdt);
1680     spapr->fdt_initial_size = spapr->fdt_size;
1681     spapr->fdt_blob = fdt;
1682 
1683     /* Set up the entry state */
1684     first_ppc_cpu->env.gpr[5] = 0;
1685 
1686     spapr->fwnmi_system_reset_addr = -1;
1687     spapr->fwnmi_machine_check_addr = -1;
1688     spapr->fwnmi_machine_check_interlock = -1;
1689 
1690     /* Signal all vCPUs waiting on this condition */
1691     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1692 
1693     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1694 }
1695 
1696 static void spapr_create_nvram(SpaprMachineState *spapr)
1697 {
1698     DeviceState *dev = qdev_new("spapr-nvram");
1699     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1700 
1701     if (dinfo) {
1702         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1703                                 &error_fatal);
1704     }
1705 
1706     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1707 
1708     spapr->nvram = (struct SpaprNvram *)dev;
1709 }
1710 
1711 static void spapr_rtc_create(SpaprMachineState *spapr)
1712 {
1713     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1714                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1715                                        &error_fatal, NULL);
1716     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1717     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1718                               "date");
1719 }
1720 
1721 /* Returns whether we want to use VGA or not */
1722 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1723 {
1724     switch (vga_interface_type) {
1725     case VGA_NONE:
1726         return false;
1727     case VGA_DEVICE:
1728         return true;
1729     case VGA_STD:
1730     case VGA_VIRTIO:
1731     case VGA_CIRRUS:
1732         return pci_vga_init(pci_bus) != NULL;
1733     default:
1734         error_setg(errp,
1735                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1736         return false;
1737     }
1738 }
1739 
1740 static int spapr_pre_load(void *opaque)
1741 {
1742     int rc;
1743 
1744     rc = spapr_caps_pre_load(opaque);
1745     if (rc) {
1746         return rc;
1747     }
1748 
1749     return 0;
1750 }
1751 
1752 static int spapr_post_load(void *opaque, int version_id)
1753 {
1754     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1755     int err = 0;
1756 
1757     err = spapr_caps_post_migration(spapr);
1758     if (err) {
1759         return err;
1760     }
1761 
1762     /*
1763      * In earlier versions, there was no separate qdev for the PAPR
1764      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1765      * So when migrating from those versions, poke the incoming offset
1766      * value into the RTC device
1767      */
1768     if (version_id < 3) {
1769         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1770         if (err) {
1771             return err;
1772         }
1773     }
1774 
1775     if (kvm_enabled() && spapr->patb_entry) {
1776         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1777         bool radix = !!(spapr->patb_entry & PATE1_GR);
1778         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1779 
1780         /*
1781          * Update LPCR:HR and UPRT as they may not be set properly in
1782          * the stream
1783          */
1784         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1785                             LPCR_HR | LPCR_UPRT);
1786 
1787         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1788         if (err) {
1789             error_report("Process table config unsupported by the host");
1790             return -EINVAL;
1791         }
1792     }
1793 
1794     err = spapr_irq_post_load(spapr, version_id);
1795     if (err) {
1796         return err;
1797     }
1798 
1799     return err;
1800 }
1801 
1802 static int spapr_pre_save(void *opaque)
1803 {
1804     int rc;
1805 
1806     rc = spapr_caps_pre_save(opaque);
1807     if (rc) {
1808         return rc;
1809     }
1810 
1811     return 0;
1812 }
1813 
1814 static bool version_before_3(void *opaque, int version_id)
1815 {
1816     return version_id < 3;
1817 }
1818 
1819 static bool spapr_pending_events_needed(void *opaque)
1820 {
1821     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1822     return !QTAILQ_EMPTY(&spapr->pending_events);
1823 }
1824 
1825 static const VMStateDescription vmstate_spapr_event_entry = {
1826     .name = "spapr_event_log_entry",
1827     .version_id = 1,
1828     .minimum_version_id = 1,
1829     .fields = (VMStateField[]) {
1830         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1831         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1832         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1833                                      NULL, extended_length),
1834         VMSTATE_END_OF_LIST()
1835     },
1836 };
1837 
1838 static const VMStateDescription vmstate_spapr_pending_events = {
1839     .name = "spapr_pending_events",
1840     .version_id = 1,
1841     .minimum_version_id = 1,
1842     .needed = spapr_pending_events_needed,
1843     .fields = (VMStateField[]) {
1844         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1845                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1846         VMSTATE_END_OF_LIST()
1847     },
1848 };
1849 
1850 static bool spapr_ov5_cas_needed(void *opaque)
1851 {
1852     SpaprMachineState *spapr = opaque;
1853     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1854     bool cas_needed;
1855 
1856     /* Prior to the introduction of SpaprOptionVector, we had two option
1857      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1858      * Both of these options encode machine topology into the device-tree
1859      * in such a way that the now-booted OS should still be able to interact
1860      * appropriately with QEMU regardless of what options were actually
1861      * negotiatied on the source side.
1862      *
1863      * As such, we can avoid migrating the CAS-negotiated options if these
1864      * are the only options available on the current machine/platform.
1865      * Since these are the only options available for pseries-2.7 and
1866      * earlier, this allows us to maintain old->new/new->old migration
1867      * compatibility.
1868      *
1869      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1870      * via default pseries-2.8 machines and explicit command-line parameters.
1871      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1872      * of the actual CAS-negotiated values to continue working properly. For
1873      * example, availability of memory unplug depends on knowing whether
1874      * OV5_HP_EVT was negotiated via CAS.
1875      *
1876      * Thus, for any cases where the set of available CAS-negotiatable
1877      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1878      * include the CAS-negotiated options in the migration stream, unless
1879      * if they affect boot time behaviour only.
1880      */
1881     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1882     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1883     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1884 
1885     /* We need extra information if we have any bits outside the mask
1886      * defined above */
1887     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1888 
1889     spapr_ovec_cleanup(ov5_mask);
1890 
1891     return cas_needed;
1892 }
1893 
1894 static const VMStateDescription vmstate_spapr_ov5_cas = {
1895     .name = "spapr_option_vector_ov5_cas",
1896     .version_id = 1,
1897     .minimum_version_id = 1,
1898     .needed = spapr_ov5_cas_needed,
1899     .fields = (VMStateField[]) {
1900         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1901                                  vmstate_spapr_ovec, SpaprOptionVector),
1902         VMSTATE_END_OF_LIST()
1903     },
1904 };
1905 
1906 static bool spapr_patb_entry_needed(void *opaque)
1907 {
1908     SpaprMachineState *spapr = opaque;
1909 
1910     return !!spapr->patb_entry;
1911 }
1912 
1913 static const VMStateDescription vmstate_spapr_patb_entry = {
1914     .name = "spapr_patb_entry",
1915     .version_id = 1,
1916     .minimum_version_id = 1,
1917     .needed = spapr_patb_entry_needed,
1918     .fields = (VMStateField[]) {
1919         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1920         VMSTATE_END_OF_LIST()
1921     },
1922 };
1923 
1924 static bool spapr_irq_map_needed(void *opaque)
1925 {
1926     SpaprMachineState *spapr = opaque;
1927 
1928     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1929 }
1930 
1931 static const VMStateDescription vmstate_spapr_irq_map = {
1932     .name = "spapr_irq_map",
1933     .version_id = 1,
1934     .minimum_version_id = 1,
1935     .needed = spapr_irq_map_needed,
1936     .fields = (VMStateField[]) {
1937         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1938         VMSTATE_END_OF_LIST()
1939     },
1940 };
1941 
1942 static bool spapr_dtb_needed(void *opaque)
1943 {
1944     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1945 
1946     return smc->update_dt_enabled;
1947 }
1948 
1949 static int spapr_dtb_pre_load(void *opaque)
1950 {
1951     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1952 
1953     g_free(spapr->fdt_blob);
1954     spapr->fdt_blob = NULL;
1955     spapr->fdt_size = 0;
1956 
1957     return 0;
1958 }
1959 
1960 static const VMStateDescription vmstate_spapr_dtb = {
1961     .name = "spapr_dtb",
1962     .version_id = 1,
1963     .minimum_version_id = 1,
1964     .needed = spapr_dtb_needed,
1965     .pre_load = spapr_dtb_pre_load,
1966     .fields = (VMStateField[]) {
1967         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1968         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1969         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1970                                      fdt_size),
1971         VMSTATE_END_OF_LIST()
1972     },
1973 };
1974 
1975 static bool spapr_fwnmi_needed(void *opaque)
1976 {
1977     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1978 
1979     return spapr->fwnmi_machine_check_addr != -1;
1980 }
1981 
1982 static int spapr_fwnmi_pre_save(void *opaque)
1983 {
1984     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1985 
1986     /*
1987      * Check if machine check handling is in progress and print a
1988      * warning message.
1989      */
1990     if (spapr->fwnmi_machine_check_interlock != -1) {
1991         warn_report("A machine check is being handled during migration. The"
1992                 "handler may run and log hardware error on the destination");
1993     }
1994 
1995     return 0;
1996 }
1997 
1998 static const VMStateDescription vmstate_spapr_fwnmi = {
1999     .name = "spapr_fwnmi",
2000     .version_id = 1,
2001     .minimum_version_id = 1,
2002     .needed = spapr_fwnmi_needed,
2003     .pre_save = spapr_fwnmi_pre_save,
2004     .fields = (VMStateField[]) {
2005         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2006         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2007         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2008         VMSTATE_END_OF_LIST()
2009     },
2010 };
2011 
2012 static const VMStateDescription vmstate_spapr = {
2013     .name = "spapr",
2014     .version_id = 3,
2015     .minimum_version_id = 1,
2016     .pre_load = spapr_pre_load,
2017     .post_load = spapr_post_load,
2018     .pre_save = spapr_pre_save,
2019     .fields = (VMStateField[]) {
2020         /* used to be @next_irq */
2021         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2022 
2023         /* RTC offset */
2024         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2025 
2026         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2027         VMSTATE_END_OF_LIST()
2028     },
2029     .subsections = (const VMStateDescription*[]) {
2030         &vmstate_spapr_ov5_cas,
2031         &vmstate_spapr_patb_entry,
2032         &vmstate_spapr_pending_events,
2033         &vmstate_spapr_cap_htm,
2034         &vmstate_spapr_cap_vsx,
2035         &vmstate_spapr_cap_dfp,
2036         &vmstate_spapr_cap_cfpc,
2037         &vmstate_spapr_cap_sbbc,
2038         &vmstate_spapr_cap_ibs,
2039         &vmstate_spapr_cap_hpt_maxpagesize,
2040         &vmstate_spapr_irq_map,
2041         &vmstate_spapr_cap_nested_kvm_hv,
2042         &vmstate_spapr_dtb,
2043         &vmstate_spapr_cap_large_decr,
2044         &vmstate_spapr_cap_ccf_assist,
2045         &vmstate_spapr_cap_fwnmi,
2046         &vmstate_spapr_fwnmi,
2047         &vmstate_spapr_cap_rpt_invalidate,
2048         NULL
2049     }
2050 };
2051 
2052 static int htab_save_setup(QEMUFile *f, void *opaque)
2053 {
2054     SpaprMachineState *spapr = opaque;
2055 
2056     /* "Iteration" header */
2057     if (!spapr->htab_shift) {
2058         qemu_put_be32(f, -1);
2059     } else {
2060         qemu_put_be32(f, spapr->htab_shift);
2061     }
2062 
2063     if (spapr->htab) {
2064         spapr->htab_save_index = 0;
2065         spapr->htab_first_pass = true;
2066     } else {
2067         if (spapr->htab_shift) {
2068             assert(kvm_enabled());
2069         }
2070     }
2071 
2072 
2073     return 0;
2074 }
2075 
2076 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2077                             int chunkstart, int n_valid, int n_invalid)
2078 {
2079     qemu_put_be32(f, chunkstart);
2080     qemu_put_be16(f, n_valid);
2081     qemu_put_be16(f, n_invalid);
2082     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2083                     HASH_PTE_SIZE_64 * n_valid);
2084 }
2085 
2086 static void htab_save_end_marker(QEMUFile *f)
2087 {
2088     qemu_put_be32(f, 0);
2089     qemu_put_be16(f, 0);
2090     qemu_put_be16(f, 0);
2091 }
2092 
2093 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2094                                  int64_t max_ns)
2095 {
2096     bool has_timeout = max_ns != -1;
2097     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2098     int index = spapr->htab_save_index;
2099     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2100 
2101     assert(spapr->htab_first_pass);
2102 
2103     do {
2104         int chunkstart;
2105 
2106         /* Consume invalid HPTEs */
2107         while ((index < htabslots)
2108                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2109             CLEAN_HPTE(HPTE(spapr->htab, index));
2110             index++;
2111         }
2112 
2113         /* Consume valid HPTEs */
2114         chunkstart = index;
2115         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2116                && HPTE_VALID(HPTE(spapr->htab, index))) {
2117             CLEAN_HPTE(HPTE(spapr->htab, index));
2118             index++;
2119         }
2120 
2121         if (index > chunkstart) {
2122             int n_valid = index - chunkstart;
2123 
2124             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2125 
2126             if (has_timeout &&
2127                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2128                 break;
2129             }
2130         }
2131     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2132 
2133     if (index >= htabslots) {
2134         assert(index == htabslots);
2135         index = 0;
2136         spapr->htab_first_pass = false;
2137     }
2138     spapr->htab_save_index = index;
2139 }
2140 
2141 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2142                                 int64_t max_ns)
2143 {
2144     bool final = max_ns < 0;
2145     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2146     int examined = 0, sent = 0;
2147     int index = spapr->htab_save_index;
2148     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2149 
2150     assert(!spapr->htab_first_pass);
2151 
2152     do {
2153         int chunkstart, invalidstart;
2154 
2155         /* Consume non-dirty HPTEs */
2156         while ((index < htabslots)
2157                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2158             index++;
2159             examined++;
2160         }
2161 
2162         chunkstart = index;
2163         /* Consume valid dirty HPTEs */
2164         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2165                && HPTE_DIRTY(HPTE(spapr->htab, index))
2166                && HPTE_VALID(HPTE(spapr->htab, index))) {
2167             CLEAN_HPTE(HPTE(spapr->htab, index));
2168             index++;
2169             examined++;
2170         }
2171 
2172         invalidstart = index;
2173         /* Consume invalid dirty HPTEs */
2174         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2175                && HPTE_DIRTY(HPTE(spapr->htab, index))
2176                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2177             CLEAN_HPTE(HPTE(spapr->htab, index));
2178             index++;
2179             examined++;
2180         }
2181 
2182         if (index > chunkstart) {
2183             int n_valid = invalidstart - chunkstart;
2184             int n_invalid = index - invalidstart;
2185 
2186             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2187             sent += index - chunkstart;
2188 
2189             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2190                 break;
2191             }
2192         }
2193 
2194         if (examined >= htabslots) {
2195             break;
2196         }
2197 
2198         if (index >= htabslots) {
2199             assert(index == htabslots);
2200             index = 0;
2201         }
2202     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2203 
2204     if (index >= htabslots) {
2205         assert(index == htabslots);
2206         index = 0;
2207     }
2208 
2209     spapr->htab_save_index = index;
2210 
2211     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2212 }
2213 
2214 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2215 #define MAX_KVM_BUF_SIZE    2048
2216 
2217 static int htab_save_iterate(QEMUFile *f, void *opaque)
2218 {
2219     SpaprMachineState *spapr = opaque;
2220     int fd;
2221     int rc = 0;
2222 
2223     /* Iteration header */
2224     if (!spapr->htab_shift) {
2225         qemu_put_be32(f, -1);
2226         return 1;
2227     } else {
2228         qemu_put_be32(f, 0);
2229     }
2230 
2231     if (!spapr->htab) {
2232         assert(kvm_enabled());
2233 
2234         fd = get_htab_fd(spapr);
2235         if (fd < 0) {
2236             return fd;
2237         }
2238 
2239         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2240         if (rc < 0) {
2241             return rc;
2242         }
2243     } else  if (spapr->htab_first_pass) {
2244         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2245     } else {
2246         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2247     }
2248 
2249     htab_save_end_marker(f);
2250 
2251     return rc;
2252 }
2253 
2254 static int htab_save_complete(QEMUFile *f, void *opaque)
2255 {
2256     SpaprMachineState *spapr = opaque;
2257     int fd;
2258 
2259     /* Iteration header */
2260     if (!spapr->htab_shift) {
2261         qemu_put_be32(f, -1);
2262         return 0;
2263     } else {
2264         qemu_put_be32(f, 0);
2265     }
2266 
2267     if (!spapr->htab) {
2268         int rc;
2269 
2270         assert(kvm_enabled());
2271 
2272         fd = get_htab_fd(spapr);
2273         if (fd < 0) {
2274             return fd;
2275         }
2276 
2277         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2278         if (rc < 0) {
2279             return rc;
2280         }
2281     } else {
2282         if (spapr->htab_first_pass) {
2283             htab_save_first_pass(f, spapr, -1);
2284         }
2285         htab_save_later_pass(f, spapr, -1);
2286     }
2287 
2288     /* End marker */
2289     htab_save_end_marker(f);
2290 
2291     return 0;
2292 }
2293 
2294 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2295 {
2296     SpaprMachineState *spapr = opaque;
2297     uint32_t section_hdr;
2298     int fd = -1;
2299     Error *local_err = NULL;
2300 
2301     if (version_id < 1 || version_id > 1) {
2302         error_report("htab_load() bad version");
2303         return -EINVAL;
2304     }
2305 
2306     section_hdr = qemu_get_be32(f);
2307 
2308     if (section_hdr == -1) {
2309         spapr_free_hpt(spapr);
2310         return 0;
2311     }
2312 
2313     if (section_hdr) {
2314         int ret;
2315 
2316         /* First section gives the htab size */
2317         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2318         if (ret < 0) {
2319             error_report_err(local_err);
2320             return ret;
2321         }
2322         return 0;
2323     }
2324 
2325     if (!spapr->htab) {
2326         assert(kvm_enabled());
2327 
2328         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2329         if (fd < 0) {
2330             error_report_err(local_err);
2331             return fd;
2332         }
2333     }
2334 
2335     while (true) {
2336         uint32_t index;
2337         uint16_t n_valid, n_invalid;
2338 
2339         index = qemu_get_be32(f);
2340         n_valid = qemu_get_be16(f);
2341         n_invalid = qemu_get_be16(f);
2342 
2343         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2344             /* End of Stream */
2345             break;
2346         }
2347 
2348         if ((index + n_valid + n_invalid) >
2349             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2350             /* Bad index in stream */
2351             error_report(
2352                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2353                 index, n_valid, n_invalid, spapr->htab_shift);
2354             return -EINVAL;
2355         }
2356 
2357         if (spapr->htab) {
2358             if (n_valid) {
2359                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2360                                 HASH_PTE_SIZE_64 * n_valid);
2361             }
2362             if (n_invalid) {
2363                 memset(HPTE(spapr->htab, index + n_valid), 0,
2364                        HASH_PTE_SIZE_64 * n_invalid);
2365             }
2366         } else {
2367             int rc;
2368 
2369             assert(fd >= 0);
2370 
2371             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2372                                         &local_err);
2373             if (rc < 0) {
2374                 error_report_err(local_err);
2375                 return rc;
2376             }
2377         }
2378     }
2379 
2380     if (!spapr->htab) {
2381         assert(fd >= 0);
2382         close(fd);
2383     }
2384 
2385     return 0;
2386 }
2387 
2388 static void htab_save_cleanup(void *opaque)
2389 {
2390     SpaprMachineState *spapr = opaque;
2391 
2392     close_htab_fd(spapr);
2393 }
2394 
2395 static SaveVMHandlers savevm_htab_handlers = {
2396     .save_setup = htab_save_setup,
2397     .save_live_iterate = htab_save_iterate,
2398     .save_live_complete_precopy = htab_save_complete,
2399     .save_cleanup = htab_save_cleanup,
2400     .load_state = htab_load,
2401 };
2402 
2403 static void spapr_boot_set(void *opaque, const char *boot_device,
2404                            Error **errp)
2405 {
2406     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2407 
2408     g_free(spapr->boot_device);
2409     spapr->boot_device = g_strdup(boot_device);
2410 }
2411 
2412 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2413 {
2414     MachineState *machine = MACHINE(spapr);
2415     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2416     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2417     int i;
2418 
2419     for (i = 0; i < nr_lmbs; i++) {
2420         uint64_t addr;
2421 
2422         addr = i * lmb_size + machine->device_memory->base;
2423         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2424                                addr / lmb_size);
2425     }
2426 }
2427 
2428 /*
2429  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2430  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2431  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2432  */
2433 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2434 {
2435     int i;
2436 
2437     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2438         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2439                    " is not aligned to %" PRIu64 " MiB",
2440                    machine->ram_size,
2441                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2442         return;
2443     }
2444 
2445     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2446         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2447                    " is not aligned to %" PRIu64 " MiB",
2448                    machine->ram_size,
2449                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2450         return;
2451     }
2452 
2453     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2454         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2455             error_setg(errp,
2456                        "Node %d memory size 0x%" PRIx64
2457                        " is not aligned to %" PRIu64 " MiB",
2458                        i, machine->numa_state->nodes[i].node_mem,
2459                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2460             return;
2461         }
2462     }
2463 }
2464 
2465 /* find cpu slot in machine->possible_cpus by core_id */
2466 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2467 {
2468     int index = id / ms->smp.threads;
2469 
2470     if (index >= ms->possible_cpus->len) {
2471         return NULL;
2472     }
2473     if (idx) {
2474         *idx = index;
2475     }
2476     return &ms->possible_cpus->cpus[index];
2477 }
2478 
2479 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2480 {
2481     MachineState *ms = MACHINE(spapr);
2482     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2483     Error *local_err = NULL;
2484     bool vsmt_user = !!spapr->vsmt;
2485     int kvm_smt = kvmppc_smt_threads();
2486     int ret;
2487     unsigned int smp_threads = ms->smp.threads;
2488 
2489     if (!kvm_enabled() && (smp_threads > 1)) {
2490         error_setg(errp, "TCG cannot support more than 1 thread/core "
2491                    "on a pseries machine");
2492         return;
2493     }
2494     if (!is_power_of_2(smp_threads)) {
2495         error_setg(errp, "Cannot support %d threads/core on a pseries "
2496                    "machine because it must be a power of 2", smp_threads);
2497         return;
2498     }
2499 
2500     /* Detemine the VSMT mode to use: */
2501     if (vsmt_user) {
2502         if (spapr->vsmt < smp_threads) {
2503             error_setg(errp, "Cannot support VSMT mode %d"
2504                        " because it must be >= threads/core (%d)",
2505                        spapr->vsmt, smp_threads);
2506             return;
2507         }
2508         /* In this case, spapr->vsmt has been set by the command line */
2509     } else if (!smc->smp_threads_vsmt) {
2510         /*
2511          * Default VSMT value is tricky, because we need it to be as
2512          * consistent as possible (for migration), but this requires
2513          * changing it for at least some existing cases.  We pick 8 as
2514          * the value that we'd get with KVM on POWER8, the
2515          * overwhelmingly common case in production systems.
2516          */
2517         spapr->vsmt = MAX(8, smp_threads);
2518     } else {
2519         spapr->vsmt = smp_threads;
2520     }
2521 
2522     /* KVM: If necessary, set the SMT mode: */
2523     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2524         ret = kvmppc_set_smt_threads(spapr->vsmt);
2525         if (ret) {
2526             /* Looks like KVM isn't able to change VSMT mode */
2527             error_setg(&local_err,
2528                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2529                        spapr->vsmt, ret);
2530             /* We can live with that if the default one is big enough
2531              * for the number of threads, and a submultiple of the one
2532              * we want.  In this case we'll waste some vcpu ids, but
2533              * behaviour will be correct */
2534             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2535                 warn_report_err(local_err);
2536             } else {
2537                 if (!vsmt_user) {
2538                     error_append_hint(&local_err,
2539                                       "On PPC, a VM with %d threads/core"
2540                                       " on a host with %d threads/core"
2541                                       " requires the use of VSMT mode %d.\n",
2542                                       smp_threads, kvm_smt, spapr->vsmt);
2543                 }
2544                 kvmppc_error_append_smt_possible_hint(&local_err);
2545                 error_propagate(errp, local_err);
2546             }
2547         }
2548     }
2549     /* else TCG: nothing to do currently */
2550 }
2551 
2552 static void spapr_init_cpus(SpaprMachineState *spapr)
2553 {
2554     MachineState *machine = MACHINE(spapr);
2555     MachineClass *mc = MACHINE_GET_CLASS(machine);
2556     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2557     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2558     const CPUArchIdList *possible_cpus;
2559     unsigned int smp_cpus = machine->smp.cpus;
2560     unsigned int smp_threads = machine->smp.threads;
2561     unsigned int max_cpus = machine->smp.max_cpus;
2562     int boot_cores_nr = smp_cpus / smp_threads;
2563     int i;
2564 
2565     possible_cpus = mc->possible_cpu_arch_ids(machine);
2566     if (mc->has_hotpluggable_cpus) {
2567         if (smp_cpus % smp_threads) {
2568             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2569                          smp_cpus, smp_threads);
2570             exit(1);
2571         }
2572         if (max_cpus % smp_threads) {
2573             error_report("max_cpus (%u) must be multiple of threads (%u)",
2574                          max_cpus, smp_threads);
2575             exit(1);
2576         }
2577     } else {
2578         if (max_cpus != smp_cpus) {
2579             error_report("This machine version does not support CPU hotplug");
2580             exit(1);
2581         }
2582         boot_cores_nr = possible_cpus->len;
2583     }
2584 
2585     if (smc->pre_2_10_has_unused_icps) {
2586         int i;
2587 
2588         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2589             /* Dummy entries get deregistered when real ICPState objects
2590              * are registered during CPU core hotplug.
2591              */
2592             pre_2_10_vmstate_register_dummy_icp(i);
2593         }
2594     }
2595 
2596     for (i = 0; i < possible_cpus->len; i++) {
2597         int core_id = i * smp_threads;
2598 
2599         if (mc->has_hotpluggable_cpus) {
2600             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2601                                    spapr_vcpu_id(spapr, core_id));
2602         }
2603 
2604         if (i < boot_cores_nr) {
2605             Object *core  = object_new(type);
2606             int nr_threads = smp_threads;
2607 
2608             /* Handle the partially filled core for older machine types */
2609             if ((i + 1) * smp_threads >= smp_cpus) {
2610                 nr_threads = smp_cpus - i * smp_threads;
2611             }
2612 
2613             object_property_set_int(core, "nr-threads", nr_threads,
2614                                     &error_fatal);
2615             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2616                                     &error_fatal);
2617             qdev_realize(DEVICE(core), NULL, &error_fatal);
2618 
2619             object_unref(core);
2620         }
2621     }
2622 }
2623 
2624 static PCIHostState *spapr_create_default_phb(void)
2625 {
2626     DeviceState *dev;
2627 
2628     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2629     qdev_prop_set_uint32(dev, "index", 0);
2630     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2631 
2632     return PCI_HOST_BRIDGE(dev);
2633 }
2634 
2635 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2636 {
2637     MachineState *machine = MACHINE(spapr);
2638     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2639     hwaddr rma_size = machine->ram_size;
2640     hwaddr node0_size = spapr_node0_size(machine);
2641 
2642     /* RMA has to fit in the first NUMA node */
2643     rma_size = MIN(rma_size, node0_size);
2644 
2645     /*
2646      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2647      * never exceed that
2648      */
2649     rma_size = MIN(rma_size, 1 * TiB);
2650 
2651     /*
2652      * Clamp the RMA size based on machine type.  This is for
2653      * migration compatibility with older qemu versions, which limited
2654      * the RMA size for complicated and mostly bad reasons.
2655      */
2656     if (smc->rma_limit) {
2657         rma_size = MIN(rma_size, smc->rma_limit);
2658     }
2659 
2660     if (rma_size < MIN_RMA_SLOF) {
2661         error_setg(errp,
2662                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2663                    "ldMiB guest RMA (Real Mode Area memory)",
2664                    MIN_RMA_SLOF / MiB);
2665         return 0;
2666     }
2667 
2668     return rma_size;
2669 }
2670 
2671 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2672 {
2673     MachineState *machine = MACHINE(spapr);
2674     int i;
2675 
2676     for (i = 0; i < machine->ram_slots; i++) {
2677         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2678     }
2679 }
2680 
2681 /* pSeries LPAR / sPAPR hardware init */
2682 static void spapr_machine_init(MachineState *machine)
2683 {
2684     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2685     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2686     MachineClass *mc = MACHINE_GET_CLASS(machine);
2687     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2688     const char *bios_name = machine->firmware ?: bios_default;
2689     const char *kernel_filename = machine->kernel_filename;
2690     const char *initrd_filename = machine->initrd_filename;
2691     PCIHostState *phb;
2692     int i;
2693     MemoryRegion *sysmem = get_system_memory();
2694     long load_limit, fw_size;
2695     char *filename;
2696     Error *resize_hpt_err = NULL;
2697 
2698     /*
2699      * if Secure VM (PEF) support is configured, then initialize it
2700      */
2701     pef_kvm_init(machine->cgs, &error_fatal);
2702 
2703     msi_nonbroken = true;
2704 
2705     QLIST_INIT(&spapr->phbs);
2706     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2707 
2708     /* Determine capabilities to run with */
2709     spapr_caps_init(spapr);
2710 
2711     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2712     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2713         /*
2714          * If the user explicitly requested a mode we should either
2715          * supply it, or fail completely (which we do below).  But if
2716          * it's not set explicitly, we reset our mode to something
2717          * that works
2718          */
2719         if (resize_hpt_err) {
2720             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2721             error_free(resize_hpt_err);
2722             resize_hpt_err = NULL;
2723         } else {
2724             spapr->resize_hpt = smc->resize_hpt_default;
2725         }
2726     }
2727 
2728     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2729 
2730     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2731         /*
2732          * User requested HPT resize, but this host can't supply it.  Bail out
2733          */
2734         error_report_err(resize_hpt_err);
2735         exit(1);
2736     }
2737     error_free(resize_hpt_err);
2738 
2739     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2740 
2741     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2742     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2743 
2744     /*
2745      * VSMT must be set in order to be able to compute VCPU ids, ie to
2746      * call spapr_max_server_number() or spapr_vcpu_id().
2747      */
2748     spapr_set_vsmt_mode(spapr, &error_fatal);
2749 
2750     /* Set up Interrupt Controller before we create the VCPUs */
2751     spapr_irq_init(spapr, &error_fatal);
2752 
2753     /* Set up containers for ibm,client-architecture-support negotiated options
2754      */
2755     spapr->ov5 = spapr_ovec_new();
2756     spapr->ov5_cas = spapr_ovec_new();
2757 
2758     if (smc->dr_lmb_enabled) {
2759         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2760         spapr_validate_node_memory(machine, &error_fatal);
2761     }
2762 
2763     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2764 
2765     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2766     if (!smc->pre_6_2_numa_affinity) {
2767         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2768     }
2769 
2770     /* advertise support for dedicated HP event source to guests */
2771     if (spapr->use_hotplug_event_source) {
2772         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2773     }
2774 
2775     /* advertise support for HPT resizing */
2776     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2777         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2778     }
2779 
2780     /* advertise support for ibm,dyamic-memory-v2 */
2781     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2782 
2783     /* advertise XIVE on POWER9 machines */
2784     if (spapr->irq->xive) {
2785         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2786     }
2787 
2788     /* init CPUs */
2789     spapr_init_cpus(spapr);
2790 
2791     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2792 
2793     /* Init numa_assoc_array */
2794     spapr_numa_associativity_init(spapr, machine);
2795 
2796     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2797         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2798                               spapr->max_compat_pvr)) {
2799         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2800         /* KVM and TCG always allow GTSE with radix... */
2801         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2802     }
2803     /* ... but not with hash (currently). */
2804 
2805     if (kvm_enabled()) {
2806         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2807         kvmppc_enable_logical_ci_hcalls();
2808         kvmppc_enable_set_mode_hcall();
2809 
2810         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2811         kvmppc_enable_clear_ref_mod_hcalls();
2812 
2813         /* Enable H_PAGE_INIT */
2814         kvmppc_enable_h_page_init();
2815     }
2816 
2817     /* map RAM */
2818     memory_region_add_subregion(sysmem, 0, machine->ram);
2819 
2820     /* always allocate the device memory information */
2821     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2822 
2823     /* initialize hotplug memory address space */
2824     if (machine->ram_size < machine->maxram_size) {
2825         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2826         /*
2827          * Limit the number of hotpluggable memory slots to half the number
2828          * slots that KVM supports, leaving the other half for PCI and other
2829          * devices. However ensure that number of slots doesn't drop below 32.
2830          */
2831         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2832                            SPAPR_MAX_RAM_SLOTS;
2833 
2834         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2835             max_memslots = SPAPR_MAX_RAM_SLOTS;
2836         }
2837         if (machine->ram_slots > max_memslots) {
2838             error_report("Specified number of memory slots %"
2839                          PRIu64" exceeds max supported %d",
2840                          machine->ram_slots, max_memslots);
2841             exit(1);
2842         }
2843 
2844         machine->device_memory->base = ROUND_UP(machine->ram_size,
2845                                                 SPAPR_DEVICE_MEM_ALIGN);
2846         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2847                            "device-memory", device_mem_size);
2848         memory_region_add_subregion(sysmem, machine->device_memory->base,
2849                                     &machine->device_memory->mr);
2850     }
2851 
2852     if (smc->dr_lmb_enabled) {
2853         spapr_create_lmb_dr_connectors(spapr);
2854     }
2855 
2856     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2857         /* Create the error string for live migration blocker */
2858         error_setg(&spapr->fwnmi_migration_blocker,
2859             "A machine check is being handled during migration. The handler"
2860             "may run and log hardware error on the destination");
2861     }
2862 
2863     if (mc->nvdimm_supported) {
2864         spapr_create_nvdimm_dr_connectors(spapr);
2865     }
2866 
2867     /* Set up RTAS event infrastructure */
2868     spapr_events_init(spapr);
2869 
2870     /* Set up the RTC RTAS interfaces */
2871     spapr_rtc_create(spapr);
2872 
2873     /* Set up VIO bus */
2874     spapr->vio_bus = spapr_vio_bus_init();
2875 
2876     for (i = 0; serial_hd(i); i++) {
2877         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2878     }
2879 
2880     /* We always have at least the nvram device on VIO */
2881     spapr_create_nvram(spapr);
2882 
2883     /*
2884      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2885      * connectors (described in root DT node's "ibm,drc-types" property)
2886      * are pre-initialized here. additional child connectors (such as
2887      * connectors for a PHBs PCI slots) are added as needed during their
2888      * parent's realization.
2889      */
2890     if (smc->dr_phb_enabled) {
2891         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2892             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2893         }
2894     }
2895 
2896     /* Set up PCI */
2897     spapr_pci_rtas_init();
2898 
2899     phb = spapr_create_default_phb();
2900 
2901     for (i = 0; i < nb_nics; i++) {
2902         NICInfo *nd = &nd_table[i];
2903 
2904         if (!nd->model) {
2905             nd->model = g_strdup("spapr-vlan");
2906         }
2907 
2908         if (g_str_equal(nd->model, "spapr-vlan") ||
2909             g_str_equal(nd->model, "ibmveth")) {
2910             spapr_vlan_create(spapr->vio_bus, nd);
2911         } else {
2912             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2913         }
2914     }
2915 
2916     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2917         spapr_vscsi_create(spapr->vio_bus);
2918     }
2919 
2920     /* Graphics */
2921     if (spapr_vga_init(phb->bus, &error_fatal)) {
2922         spapr->has_graphics = true;
2923         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2924     }
2925 
2926     if (machine->usb) {
2927         if (smc->use_ohci_by_default) {
2928             pci_create_simple(phb->bus, -1, "pci-ohci");
2929         } else {
2930             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2931         }
2932 
2933         if (spapr->has_graphics) {
2934             USBBus *usb_bus = usb_bus_find(-1);
2935 
2936             usb_create_simple(usb_bus, "usb-kbd");
2937             usb_create_simple(usb_bus, "usb-mouse");
2938         }
2939     }
2940 
2941     if (kernel_filename) {
2942         spapr->kernel_size = load_elf(kernel_filename, NULL,
2943                                       translate_kernel_address, spapr,
2944                                       NULL, NULL, NULL, NULL, 1,
2945                                       PPC_ELF_MACHINE, 0, 0);
2946         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2947             spapr->kernel_size = load_elf(kernel_filename, NULL,
2948                                           translate_kernel_address, spapr,
2949                                           NULL, NULL, NULL, NULL, 0,
2950                                           PPC_ELF_MACHINE, 0, 0);
2951             spapr->kernel_le = spapr->kernel_size > 0;
2952         }
2953         if (spapr->kernel_size < 0) {
2954             error_report("error loading %s: %s", kernel_filename,
2955                          load_elf_strerror(spapr->kernel_size));
2956             exit(1);
2957         }
2958 
2959         /* load initrd */
2960         if (initrd_filename) {
2961             /* Try to locate the initrd in the gap between the kernel
2962              * and the firmware. Add a bit of space just in case
2963              */
2964             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2965                                   + 0x1ffff) & ~0xffff;
2966             spapr->initrd_size = load_image_targphys(initrd_filename,
2967                                                      spapr->initrd_base,
2968                                                      load_limit
2969                                                      - spapr->initrd_base);
2970             if (spapr->initrd_size < 0) {
2971                 error_report("could not load initial ram disk '%s'",
2972                              initrd_filename);
2973                 exit(1);
2974             }
2975         }
2976     }
2977 
2978     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2979     if (!filename) {
2980         error_report("Could not find LPAR firmware '%s'", bios_name);
2981         exit(1);
2982     }
2983     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2984     if (fw_size <= 0) {
2985         error_report("Could not load LPAR firmware '%s'", filename);
2986         exit(1);
2987     }
2988     g_free(filename);
2989 
2990     /* FIXME: Should register things through the MachineState's qdev
2991      * interface, this is a legacy from the sPAPREnvironment structure
2992      * which predated MachineState but had a similar function */
2993     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2994     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2995                          &savevm_htab_handlers, spapr);
2996 
2997     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2998 
2999     qemu_register_boot_set(spapr_boot_set, spapr);
3000 
3001     /*
3002      * Nothing needs to be done to resume a suspended guest because
3003      * suspending does not change the machine state, so no need for
3004      * a ->wakeup method.
3005      */
3006     qemu_register_wakeup_support();
3007 
3008     if (kvm_enabled()) {
3009         /* to stop and start vmclock */
3010         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3011                                          &spapr->tb);
3012 
3013         kvmppc_spapr_enable_inkernel_multitce();
3014     }
3015 
3016     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3017     if (spapr->vof) {
3018         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3019         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3020     }
3021 }
3022 
3023 #define DEFAULT_KVM_TYPE "auto"
3024 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3025 {
3026     /*
3027      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3028      * accomodate the 'HV' and 'PV' formats that exists in the
3029      * wild. The 'auto' mode is being introduced already as
3030      * lower-case, thus we don't need to bother checking for
3031      * "AUTO".
3032      */
3033     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3034         return 0;
3035     }
3036 
3037     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3038         return 1;
3039     }
3040 
3041     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3042         return 2;
3043     }
3044 
3045     error_report("Unknown kvm-type specified '%s'", vm_type);
3046     exit(1);
3047 }
3048 
3049 /*
3050  * Implementation of an interface to adjust firmware path
3051  * for the bootindex property handling.
3052  */
3053 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3054                                    DeviceState *dev)
3055 {
3056 #define CAST(type, obj, name) \
3057     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3058     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3059     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3060     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3061     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3062 
3063     if (d && bus) {
3064         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3065         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3066         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3067 
3068         if (spapr) {
3069             /*
3070              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3071              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3072              * 0x8000 | (target << 8) | (bus << 5) | lun
3073              * (see the "Logical unit addressing format" table in SAM5)
3074              */
3075             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3076             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3077                                    (uint64_t)id << 48);
3078         } else if (virtio) {
3079             /*
3080              * We use SRP luns of the form 01000000 | (target << 8) | lun
3081              * in the top 32 bits of the 64-bit LUN
3082              * Note: the quote above is from SLOF and it is wrong,
3083              * the actual binding is:
3084              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3085              */
3086             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3087             if (d->lun >= 256) {
3088                 /* Use the LUN "flat space addressing method" */
3089                 id |= 0x4000;
3090             }
3091             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3092                                    (uint64_t)id << 32);
3093         } else if (usb) {
3094             /*
3095              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3096              * in the top 32 bits of the 64-bit LUN
3097              */
3098             unsigned usb_port = atoi(usb->port->path);
3099             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3100             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3101                                    (uint64_t)id << 32);
3102         }
3103     }
3104 
3105     /*
3106      * SLOF probes the USB devices, and if it recognizes that the device is a
3107      * storage device, it changes its name to "storage" instead of "usb-host",
3108      * and additionally adds a child node for the SCSI LUN, so the correct
3109      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3110      */
3111     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3112         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3113         if (usb_device_is_scsi_storage(usbdev)) {
3114             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3115         }
3116     }
3117 
3118     if (phb) {
3119         /* Replace "pci" with "pci@800000020000000" */
3120         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3121     }
3122 
3123     if (vsc) {
3124         /* Same logic as virtio above */
3125         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3126         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3127     }
3128 
3129     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3130         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3131         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3132         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3133     }
3134 
3135     if (pcidev) {
3136         return spapr_pci_fw_dev_name(pcidev);
3137     }
3138 
3139     return NULL;
3140 }
3141 
3142 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3143 {
3144     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3145 
3146     return g_strdup(spapr->kvm_type);
3147 }
3148 
3149 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3150 {
3151     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3152 
3153     g_free(spapr->kvm_type);
3154     spapr->kvm_type = g_strdup(value);
3155 }
3156 
3157 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     return spapr->use_hotplug_event_source;
3162 }
3163 
3164 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3165                                             Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     spapr->use_hotplug_event_source = value;
3170 }
3171 
3172 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3173 {
3174     return true;
3175 }
3176 
3177 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3178 {
3179     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3180 
3181     switch (spapr->resize_hpt) {
3182     case SPAPR_RESIZE_HPT_DEFAULT:
3183         return g_strdup("default");
3184     case SPAPR_RESIZE_HPT_DISABLED:
3185         return g_strdup("disabled");
3186     case SPAPR_RESIZE_HPT_ENABLED:
3187         return g_strdup("enabled");
3188     case SPAPR_RESIZE_HPT_REQUIRED:
3189         return g_strdup("required");
3190     }
3191     g_assert_not_reached();
3192 }
3193 
3194 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3195 {
3196     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3197 
3198     if (strcmp(value, "default") == 0) {
3199         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3200     } else if (strcmp(value, "disabled") == 0) {
3201         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3202     } else if (strcmp(value, "enabled") == 0) {
3203         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3204     } else if (strcmp(value, "required") == 0) {
3205         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3206     } else {
3207         error_setg(errp, "Bad value for \"resize-hpt\" property");
3208     }
3209 }
3210 
3211 static bool spapr_get_vof(Object *obj, Error **errp)
3212 {
3213     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3214 
3215     return spapr->vof != NULL;
3216 }
3217 
3218 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3219 {
3220     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3221 
3222     if (spapr->vof) {
3223         vof_cleanup(spapr->vof);
3224         g_free(spapr->vof);
3225         spapr->vof = NULL;
3226     }
3227     if (!value) {
3228         return;
3229     }
3230     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3231 }
3232 
3233 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3234 {
3235     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3236 
3237     if (spapr->irq == &spapr_irq_xics_legacy) {
3238         return g_strdup("legacy");
3239     } else if (spapr->irq == &spapr_irq_xics) {
3240         return g_strdup("xics");
3241     } else if (spapr->irq == &spapr_irq_xive) {
3242         return g_strdup("xive");
3243     } else if (spapr->irq == &spapr_irq_dual) {
3244         return g_strdup("dual");
3245     }
3246     g_assert_not_reached();
3247 }
3248 
3249 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3250 {
3251     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3252 
3253     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3254         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3255         return;
3256     }
3257 
3258     /* The legacy IRQ backend can not be set */
3259     if (strcmp(value, "xics") == 0) {
3260         spapr->irq = &spapr_irq_xics;
3261     } else if (strcmp(value, "xive") == 0) {
3262         spapr->irq = &spapr_irq_xive;
3263     } else if (strcmp(value, "dual") == 0) {
3264         spapr->irq = &spapr_irq_dual;
3265     } else {
3266         error_setg(errp, "Bad value for \"ic-mode\" property");
3267     }
3268 }
3269 
3270 static char *spapr_get_host_model(Object *obj, Error **errp)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273 
3274     return g_strdup(spapr->host_model);
3275 }
3276 
3277 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3278 {
3279     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3280 
3281     g_free(spapr->host_model);
3282     spapr->host_model = g_strdup(value);
3283 }
3284 
3285 static char *spapr_get_host_serial(Object *obj, Error **errp)
3286 {
3287     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3288 
3289     return g_strdup(spapr->host_serial);
3290 }
3291 
3292 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3293 {
3294     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3295 
3296     g_free(spapr->host_serial);
3297     spapr->host_serial = g_strdup(value);
3298 }
3299 
3300 static void spapr_instance_init(Object *obj)
3301 {
3302     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3303     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3304     MachineState *ms = MACHINE(spapr);
3305     MachineClass *mc = MACHINE_GET_CLASS(ms);
3306 
3307     /*
3308      * NVDIMM support went live in 5.1 without considering that, in
3309      * other archs, the user needs to enable NVDIMM support with the
3310      * 'nvdimm' machine option and the default behavior is NVDIMM
3311      * support disabled. It is too late to roll back to the standard
3312      * behavior without breaking 5.1 guests.
3313      */
3314     if (mc->nvdimm_supported) {
3315         ms->nvdimms_state->is_enabled = true;
3316     }
3317 
3318     spapr->htab_fd = -1;
3319     spapr->use_hotplug_event_source = true;
3320     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3321     object_property_add_str(obj, "kvm-type",
3322                             spapr_get_kvm_type, spapr_set_kvm_type);
3323     object_property_set_description(obj, "kvm-type",
3324                                     "Specifies the KVM virtualization mode (auto,"
3325                                     " hv, pr). Defaults to 'auto'. This mode will use"
3326                                     " any available KVM module loaded in the host,"
3327                                     " where kvm_hv takes precedence if both kvm_hv and"
3328                                     " kvm_pr are loaded.");
3329     object_property_add_bool(obj, "modern-hotplug-events",
3330                             spapr_get_modern_hotplug_events,
3331                             spapr_set_modern_hotplug_events);
3332     object_property_set_description(obj, "modern-hotplug-events",
3333                                     "Use dedicated hotplug event mechanism in"
3334                                     " place of standard EPOW events when possible"
3335                                     " (required for memory hot-unplug support)");
3336     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3337                             "Maximum permitted CPU compatibility mode");
3338 
3339     object_property_add_str(obj, "resize-hpt",
3340                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3341     object_property_set_description(obj, "resize-hpt",
3342                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3343     object_property_add_uint32_ptr(obj, "vsmt",
3344                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3345     object_property_set_description(obj, "vsmt",
3346                                     "Virtual SMT: KVM behaves as if this were"
3347                                     " the host's SMT mode");
3348 
3349     object_property_add_bool(obj, "vfio-no-msix-emulation",
3350                              spapr_get_msix_emulation, NULL);
3351 
3352     object_property_add_uint64_ptr(obj, "kernel-addr",
3353                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3354     object_property_set_description(obj, "kernel-addr",
3355                                     stringify(KERNEL_LOAD_ADDR)
3356                                     " for -kernel is the default");
3357     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3358 
3359     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3360     object_property_set_description(obj, "x-vof",
3361                                     "Enable Virtual Open Firmware (experimental)");
3362 
3363     /* The machine class defines the default interrupt controller mode */
3364     spapr->irq = smc->irq;
3365     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3366                             spapr_set_ic_mode);
3367     object_property_set_description(obj, "ic-mode",
3368                  "Specifies the interrupt controller mode (xics, xive, dual)");
3369 
3370     object_property_add_str(obj, "host-model",
3371         spapr_get_host_model, spapr_set_host_model);
3372     object_property_set_description(obj, "host-model",
3373         "Host model to advertise in guest device tree");
3374     object_property_add_str(obj, "host-serial",
3375         spapr_get_host_serial, spapr_set_host_serial);
3376     object_property_set_description(obj, "host-serial",
3377         "Host serial number to advertise in guest device tree");
3378 }
3379 
3380 static void spapr_machine_finalizefn(Object *obj)
3381 {
3382     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3383 
3384     g_free(spapr->kvm_type);
3385 }
3386 
3387 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3388 {
3389     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3390     PowerPCCPU *cpu = POWERPC_CPU(cs);
3391     CPUPPCState *env = &cpu->env;
3392 
3393     cpu_synchronize_state(cs);
3394     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3395     if (spapr->fwnmi_system_reset_addr != -1) {
3396         uint64_t rtas_addr, addr;
3397 
3398         /* get rtas addr from fdt */
3399         rtas_addr = spapr_get_rtas_addr();
3400         if (!rtas_addr) {
3401             qemu_system_guest_panicked(NULL);
3402             return;
3403         }
3404 
3405         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3406         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3407         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3408         env->gpr[3] = addr;
3409     }
3410     ppc_cpu_do_system_reset(cs);
3411     if (spapr->fwnmi_system_reset_addr != -1) {
3412         env->nip = spapr->fwnmi_system_reset_addr;
3413     }
3414 }
3415 
3416 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3417 {
3418     CPUState *cs;
3419 
3420     CPU_FOREACH(cs) {
3421         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3422     }
3423 }
3424 
3425 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3426                           void *fdt, int *fdt_start_offset, Error **errp)
3427 {
3428     uint64_t addr;
3429     uint32_t node;
3430 
3431     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3432     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3433                                     &error_abort);
3434     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3435                                              SPAPR_MEMORY_BLOCK_SIZE);
3436     return 0;
3437 }
3438 
3439 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3440                            bool dedicated_hp_event_source)
3441 {
3442     SpaprDrc *drc;
3443     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3444     int i;
3445     uint64_t addr = addr_start;
3446     bool hotplugged = spapr_drc_hotplugged(dev);
3447 
3448     for (i = 0; i < nr_lmbs; i++) {
3449         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3450                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3451         g_assert(drc);
3452 
3453         /*
3454          * memory_device_get_free_addr() provided a range of free addresses
3455          * that doesn't overlap with any existing mapping at pre-plug. The
3456          * corresponding LMB DRCs are thus assumed to be all attachable.
3457          */
3458         spapr_drc_attach(drc, dev);
3459         if (!hotplugged) {
3460             spapr_drc_reset(drc);
3461         }
3462         addr += SPAPR_MEMORY_BLOCK_SIZE;
3463     }
3464     /* send hotplug notification to the
3465      * guest only in case of hotplugged memory
3466      */
3467     if (hotplugged) {
3468         if (dedicated_hp_event_source) {
3469             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3470                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3471             g_assert(drc);
3472             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3473                                                    nr_lmbs,
3474                                                    spapr_drc_index(drc));
3475         } else {
3476             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3477                                            nr_lmbs);
3478         }
3479     }
3480 }
3481 
3482 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3483 {
3484     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3485     PCDIMMDevice *dimm = PC_DIMM(dev);
3486     uint64_t size, addr;
3487     int64_t slot;
3488     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3489 
3490     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3491 
3492     pc_dimm_plug(dimm, MACHINE(ms));
3493 
3494     if (!is_nvdimm) {
3495         addr = object_property_get_uint(OBJECT(dimm),
3496                                         PC_DIMM_ADDR_PROP, &error_abort);
3497         spapr_add_lmbs(dev, addr, size,
3498                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3499     } else {
3500         slot = object_property_get_int(OBJECT(dimm),
3501                                        PC_DIMM_SLOT_PROP, &error_abort);
3502         /* We should have valid slot number at this point */
3503         g_assert(slot >= 0);
3504         spapr_add_nvdimm(dev, slot);
3505     }
3506 }
3507 
3508 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3509                                   Error **errp)
3510 {
3511     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3512     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3513     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3514     PCDIMMDevice *dimm = PC_DIMM(dev);
3515     Error *local_err = NULL;
3516     uint64_t size;
3517     Object *memdev;
3518     hwaddr pagesize;
3519 
3520     if (!smc->dr_lmb_enabled) {
3521         error_setg(errp, "Memory hotplug not supported for this machine");
3522         return;
3523     }
3524 
3525     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3526     if (local_err) {
3527         error_propagate(errp, local_err);
3528         return;
3529     }
3530 
3531     if (is_nvdimm) {
3532         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3533             return;
3534         }
3535     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3536         error_setg(errp, "Hotplugged memory size must be a multiple of "
3537                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3538         return;
3539     }
3540 
3541     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3542                                       &error_abort);
3543     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3544     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3545         return;
3546     }
3547 
3548     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3549 }
3550 
3551 struct SpaprDimmState {
3552     PCDIMMDevice *dimm;
3553     uint32_t nr_lmbs;
3554     QTAILQ_ENTRY(SpaprDimmState) next;
3555 };
3556 
3557 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3558                                                        PCDIMMDevice *dimm)
3559 {
3560     SpaprDimmState *dimm_state = NULL;
3561 
3562     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3563         if (dimm_state->dimm == dimm) {
3564             break;
3565         }
3566     }
3567     return dimm_state;
3568 }
3569 
3570 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3571                                                       uint32_t nr_lmbs,
3572                                                       PCDIMMDevice *dimm)
3573 {
3574     SpaprDimmState *ds = NULL;
3575 
3576     /*
3577      * If this request is for a DIMM whose removal had failed earlier
3578      * (due to guest's refusal to remove the LMBs), we would have this
3579      * dimm already in the pending_dimm_unplugs list. In that
3580      * case don't add again.
3581      */
3582     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3583     if (!ds) {
3584         ds = g_malloc0(sizeof(SpaprDimmState));
3585         ds->nr_lmbs = nr_lmbs;
3586         ds->dimm = dimm;
3587         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3588     }
3589     return ds;
3590 }
3591 
3592 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3593                                               SpaprDimmState *dimm_state)
3594 {
3595     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3596     g_free(dimm_state);
3597 }
3598 
3599 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3600                                                         PCDIMMDevice *dimm)
3601 {
3602     SpaprDrc *drc;
3603     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3604                                                   &error_abort);
3605     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3606     uint32_t avail_lmbs = 0;
3607     uint64_t addr_start, addr;
3608     int i;
3609 
3610     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3611                                           &error_abort);
3612 
3613     addr = addr_start;
3614     for (i = 0; i < nr_lmbs; i++) {
3615         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3616                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3617         g_assert(drc);
3618         if (drc->dev) {
3619             avail_lmbs++;
3620         }
3621         addr += SPAPR_MEMORY_BLOCK_SIZE;
3622     }
3623 
3624     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3625 }
3626 
3627 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3628 {
3629     SpaprDimmState *ds;
3630     PCDIMMDevice *dimm;
3631     SpaprDrc *drc;
3632     uint32_t nr_lmbs;
3633     uint64_t size, addr_start, addr;
3634     g_autofree char *qapi_error = NULL;
3635     int i;
3636 
3637     if (!dev) {
3638         return;
3639     }
3640 
3641     dimm = PC_DIMM(dev);
3642     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3643 
3644     /*
3645      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3646      * unplug state, but one of its DRC is marked as unplug_requested.
3647      * This is bad and weird enough to g_assert() out.
3648      */
3649     g_assert(ds);
3650 
3651     spapr_pending_dimm_unplugs_remove(spapr, ds);
3652 
3653     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3654     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3655 
3656     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3657                                           &error_abort);
3658 
3659     addr = addr_start;
3660     for (i = 0; i < nr_lmbs; i++) {
3661         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3662                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3663         g_assert(drc);
3664 
3665         drc->unplug_requested = false;
3666         addr += SPAPR_MEMORY_BLOCK_SIZE;
3667     }
3668 
3669     /*
3670      * Tell QAPI that something happened and the memory
3671      * hotunplug wasn't successful. Keep sending
3672      * MEM_UNPLUG_ERROR even while sending
3673      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3674      * MEM_UNPLUG_ERROR is due.
3675      */
3676     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3677                                  "for device %s", dev->id);
3678 
3679     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3680 
3681     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3682                                               dev->canonical_path);
3683 }
3684 
3685 /* Callback to be called during DRC release. */
3686 void spapr_lmb_release(DeviceState *dev)
3687 {
3688     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3689     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3690     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3691 
3692     /* This information will get lost if a migration occurs
3693      * during the unplug process. In this case recover it. */
3694     if (ds == NULL) {
3695         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3696         g_assert(ds);
3697         /* The DRC being examined by the caller at least must be counted */
3698         g_assert(ds->nr_lmbs);
3699     }
3700 
3701     if (--ds->nr_lmbs) {
3702         return;
3703     }
3704 
3705     /*
3706      * Now that all the LMBs have been removed by the guest, call the
3707      * unplug handler chain. This can never fail.
3708      */
3709     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3710     object_unparent(OBJECT(dev));
3711 }
3712 
3713 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3714 {
3715     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3716     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3717 
3718     /* We really shouldn't get this far without anything to unplug */
3719     g_assert(ds);
3720 
3721     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3722     qdev_unrealize(dev);
3723     spapr_pending_dimm_unplugs_remove(spapr, ds);
3724 }
3725 
3726 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3727                                         DeviceState *dev, Error **errp)
3728 {
3729     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3730     PCDIMMDevice *dimm = PC_DIMM(dev);
3731     uint32_t nr_lmbs;
3732     uint64_t size, addr_start, addr;
3733     int i;
3734     SpaprDrc *drc;
3735 
3736     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3737         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3738         return;
3739     }
3740 
3741     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3742     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3743 
3744     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3745                                           &error_abort);
3746 
3747     /*
3748      * An existing pending dimm state for this DIMM means that there is an
3749      * unplug operation in progress, waiting for the spapr_lmb_release
3750      * callback to complete the job (BQL can't cover that far). In this case,
3751      * bail out to avoid detaching DRCs that were already released.
3752      */
3753     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3754         error_setg(errp, "Memory unplug already in progress for device %s",
3755                    dev->id);
3756         return;
3757     }
3758 
3759     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3760 
3761     addr = addr_start;
3762     for (i = 0; i < nr_lmbs; i++) {
3763         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3764                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3765         g_assert(drc);
3766 
3767         spapr_drc_unplug_request(drc);
3768         addr += SPAPR_MEMORY_BLOCK_SIZE;
3769     }
3770 
3771     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3772                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3773     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3774                                               nr_lmbs, spapr_drc_index(drc));
3775 }
3776 
3777 /* Callback to be called during DRC release. */
3778 void spapr_core_release(DeviceState *dev)
3779 {
3780     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3781 
3782     /* Call the unplug handler chain. This can never fail. */
3783     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3784     object_unparent(OBJECT(dev));
3785 }
3786 
3787 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3788 {
3789     MachineState *ms = MACHINE(hotplug_dev);
3790     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3791     CPUCore *cc = CPU_CORE(dev);
3792     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3793 
3794     if (smc->pre_2_10_has_unused_icps) {
3795         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3796         int i;
3797 
3798         for (i = 0; i < cc->nr_threads; i++) {
3799             CPUState *cs = CPU(sc->threads[i]);
3800 
3801             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3802         }
3803     }
3804 
3805     assert(core_slot);
3806     core_slot->cpu = NULL;
3807     qdev_unrealize(dev);
3808 }
3809 
3810 static
3811 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3812                                Error **errp)
3813 {
3814     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3815     int index;
3816     SpaprDrc *drc;
3817     CPUCore *cc = CPU_CORE(dev);
3818 
3819     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3820         error_setg(errp, "Unable to find CPU core with core-id: %d",
3821                    cc->core_id);
3822         return;
3823     }
3824     if (index == 0) {
3825         error_setg(errp, "Boot CPU core may not be unplugged");
3826         return;
3827     }
3828 
3829     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3830                           spapr_vcpu_id(spapr, cc->core_id));
3831     g_assert(drc);
3832 
3833     if (!spapr_drc_unplug_requested(drc)) {
3834         spapr_drc_unplug_request(drc);
3835     }
3836 
3837     /*
3838      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3839      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3840      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3841      * attempt (e.g. the kernel will refuse to remove the last online
3842      * CPU), we will never attempt it again because unplug_requested
3843      * will still be 'true' in that case.
3844      */
3845     spapr_hotplug_req_remove_by_index(drc);
3846 }
3847 
3848 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3849                            void *fdt, int *fdt_start_offset, Error **errp)
3850 {
3851     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3852     CPUState *cs = CPU(core->threads[0]);
3853     PowerPCCPU *cpu = POWERPC_CPU(cs);
3854     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3855     int id = spapr_get_vcpu_id(cpu);
3856     g_autofree char *nodename = NULL;
3857     int offset;
3858 
3859     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3860     offset = fdt_add_subnode(fdt, 0, nodename);
3861 
3862     spapr_dt_cpu(cs, fdt, offset, spapr);
3863 
3864     /*
3865      * spapr_dt_cpu() does not fill the 'name' property in the
3866      * CPU node. The function is called during boot process, before
3867      * and after CAS, and overwriting the 'name' property written
3868      * by SLOF is not allowed.
3869      *
3870      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3871      * CPUs more compatible with the coldplugged ones, which have
3872      * the 'name' property. Linux Kernel also relies on this
3873      * property to identify CPU nodes.
3874      */
3875     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3876 
3877     *fdt_start_offset = offset;
3878     return 0;
3879 }
3880 
3881 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3882 {
3883     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3884     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3885     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3886     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3887     CPUCore *cc = CPU_CORE(dev);
3888     CPUState *cs;
3889     SpaprDrc *drc;
3890     CPUArchId *core_slot;
3891     int index;
3892     bool hotplugged = spapr_drc_hotplugged(dev);
3893     int i;
3894 
3895     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3896     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3897 
3898     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3899                           spapr_vcpu_id(spapr, cc->core_id));
3900 
3901     g_assert(drc || !mc->has_hotpluggable_cpus);
3902 
3903     if (drc) {
3904         /*
3905          * spapr_core_pre_plug() already buys us this is a brand new
3906          * core being plugged into a free slot. Nothing should already
3907          * be attached to the corresponding DRC.
3908          */
3909         spapr_drc_attach(drc, dev);
3910 
3911         if (hotplugged) {
3912             /*
3913              * Send hotplug notification interrupt to the guest only
3914              * in case of hotplugged CPUs.
3915              */
3916             spapr_hotplug_req_add_by_index(drc);
3917         } else {
3918             spapr_drc_reset(drc);
3919         }
3920     }
3921 
3922     core_slot->cpu = OBJECT(dev);
3923 
3924     /*
3925      * Set compatibility mode to match the boot CPU, which was either set
3926      * by the machine reset code or by CAS. This really shouldn't fail at
3927      * this point.
3928      */
3929     if (hotplugged) {
3930         for (i = 0; i < cc->nr_threads; i++) {
3931             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3932                            &error_abort);
3933         }
3934     }
3935 
3936     if (smc->pre_2_10_has_unused_icps) {
3937         for (i = 0; i < cc->nr_threads; i++) {
3938             cs = CPU(core->threads[i]);
3939             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3940         }
3941     }
3942 }
3943 
3944 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3945                                 Error **errp)
3946 {
3947     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3948     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3949     CPUCore *cc = CPU_CORE(dev);
3950     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3951     const char *type = object_get_typename(OBJECT(dev));
3952     CPUArchId *core_slot;
3953     int index;
3954     unsigned int smp_threads = machine->smp.threads;
3955 
3956     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3957         error_setg(errp, "CPU hotplug not supported for this machine");
3958         return;
3959     }
3960 
3961     if (strcmp(base_core_type, type)) {
3962         error_setg(errp, "CPU core type should be %s", base_core_type);
3963         return;
3964     }
3965 
3966     if (cc->core_id % smp_threads) {
3967         error_setg(errp, "invalid core id %d", cc->core_id);
3968         return;
3969     }
3970 
3971     /*
3972      * In general we should have homogeneous threads-per-core, but old
3973      * (pre hotplug support) machine types allow the last core to have
3974      * reduced threads as a compatibility hack for when we allowed
3975      * total vcpus not a multiple of threads-per-core.
3976      */
3977     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3978         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3979                    smp_threads);
3980         return;
3981     }
3982 
3983     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3984     if (!core_slot) {
3985         error_setg(errp, "core id %d out of range", cc->core_id);
3986         return;
3987     }
3988 
3989     if (core_slot->cpu) {
3990         error_setg(errp, "core %d already populated", cc->core_id);
3991         return;
3992     }
3993 
3994     numa_cpu_pre_plug(core_slot, dev, errp);
3995 }
3996 
3997 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3998                           void *fdt, int *fdt_start_offset, Error **errp)
3999 {
4000     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4001     int intc_phandle;
4002 
4003     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4004     if (intc_phandle <= 0) {
4005         return -1;
4006     }
4007 
4008     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4009         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4010         return -1;
4011     }
4012 
4013     /* generally SLOF creates these, for hotplug it's up to QEMU */
4014     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4015 
4016     return 0;
4017 }
4018 
4019 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4020                                Error **errp)
4021 {
4022     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4023     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4024     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4025     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4026     SpaprDrc *drc;
4027 
4028     if (dev->hotplugged && !smc->dr_phb_enabled) {
4029         error_setg(errp, "PHB hotplug not supported for this machine");
4030         return false;
4031     }
4032 
4033     if (sphb->index == (uint32_t)-1) {
4034         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4035         return false;
4036     }
4037 
4038     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4039     if (drc && drc->dev) {
4040         error_setg(errp, "PHB %d already attached", sphb->index);
4041         return false;
4042     }
4043 
4044     /*
4045      * This will check that sphb->index doesn't exceed the maximum number of
4046      * PHBs for the current machine type.
4047      */
4048     return
4049         smc->phb_placement(spapr, sphb->index,
4050                            &sphb->buid, &sphb->io_win_addr,
4051                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4052                            windows_supported, sphb->dma_liobn,
4053                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4054                            errp);
4055 }
4056 
4057 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4058 {
4059     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4060     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4061     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4062     SpaprDrc *drc;
4063     bool hotplugged = spapr_drc_hotplugged(dev);
4064 
4065     if (!smc->dr_phb_enabled) {
4066         return;
4067     }
4068 
4069     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4070     /* hotplug hooks should check it's enabled before getting this far */
4071     assert(drc);
4072 
4073     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4074     spapr_drc_attach(drc, dev);
4075 
4076     if (hotplugged) {
4077         spapr_hotplug_req_add_by_index(drc);
4078     } else {
4079         spapr_drc_reset(drc);
4080     }
4081 }
4082 
4083 void spapr_phb_release(DeviceState *dev)
4084 {
4085     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4086 
4087     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4088     object_unparent(OBJECT(dev));
4089 }
4090 
4091 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4092 {
4093     qdev_unrealize(dev);
4094 }
4095 
4096 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4097                                      DeviceState *dev, Error **errp)
4098 {
4099     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4100     SpaprDrc *drc;
4101 
4102     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4103     assert(drc);
4104 
4105     if (!spapr_drc_unplug_requested(drc)) {
4106         spapr_drc_unplug_request(drc);
4107         spapr_hotplug_req_remove_by_index(drc);
4108     } else {
4109         error_setg(errp,
4110                    "PCI Host Bridge unplug already in progress for device %s",
4111                    dev->id);
4112     }
4113 }
4114 
4115 static
4116 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4117                               Error **errp)
4118 {
4119     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4120 
4121     if (spapr->tpm_proxy != NULL) {
4122         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4123         return false;
4124     }
4125 
4126     return true;
4127 }
4128 
4129 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4130 {
4131     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4132     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4133 
4134     /* Already checked in spapr_tpm_proxy_pre_plug() */
4135     g_assert(spapr->tpm_proxy == NULL);
4136 
4137     spapr->tpm_proxy = tpm_proxy;
4138 }
4139 
4140 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4141 {
4142     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4143 
4144     qdev_unrealize(dev);
4145     object_unparent(OBJECT(dev));
4146     spapr->tpm_proxy = NULL;
4147 }
4148 
4149 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4150                                       DeviceState *dev, Error **errp)
4151 {
4152     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4153         spapr_memory_plug(hotplug_dev, dev);
4154     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4155         spapr_core_plug(hotplug_dev, dev);
4156     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4157         spapr_phb_plug(hotplug_dev, dev);
4158     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4159         spapr_tpm_proxy_plug(hotplug_dev, dev);
4160     }
4161 }
4162 
4163 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4164                                         DeviceState *dev, Error **errp)
4165 {
4166     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4167         spapr_memory_unplug(hotplug_dev, dev);
4168     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4169         spapr_core_unplug(hotplug_dev, dev);
4170     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4171         spapr_phb_unplug(hotplug_dev, dev);
4172     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4173         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4174     }
4175 }
4176 
4177 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4178 {
4179     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4180         /*
4181          * CAS will process all pending unplug requests.
4182          *
4183          * HACK: a guest could theoretically have cleared all bits in OV5,
4184          * but none of the guests we care for do.
4185          */
4186         spapr_ovec_empty(spapr->ov5_cas);
4187 }
4188 
4189 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4190                                                 DeviceState *dev, Error **errp)
4191 {
4192     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4193     MachineClass *mc = MACHINE_GET_CLASS(sms);
4194     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4195 
4196     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4197         if (spapr_memory_hot_unplug_supported(sms)) {
4198             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4199         } else {
4200             error_setg(errp, "Memory hot unplug not supported for this guest");
4201         }
4202     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4203         if (!mc->has_hotpluggable_cpus) {
4204             error_setg(errp, "CPU hot unplug not supported on this machine");
4205             return;
4206         }
4207         spapr_core_unplug_request(hotplug_dev, dev, errp);
4208     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4209         if (!smc->dr_phb_enabled) {
4210             error_setg(errp, "PHB hot unplug not supported on this machine");
4211             return;
4212         }
4213         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4214     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4215         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4216     }
4217 }
4218 
4219 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4220                                           DeviceState *dev, Error **errp)
4221 {
4222     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4223         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4224     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4225         spapr_core_pre_plug(hotplug_dev, dev, errp);
4226     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4227         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4228     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4229         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4230     }
4231 }
4232 
4233 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4234                                                  DeviceState *dev)
4235 {
4236     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4237         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4238         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4239         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4240         return HOTPLUG_HANDLER(machine);
4241     }
4242     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4243         PCIDevice *pcidev = PCI_DEVICE(dev);
4244         PCIBus *root = pci_device_root_bus(pcidev);
4245         SpaprPhbState *phb =
4246             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4247                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4248 
4249         if (phb) {
4250             return HOTPLUG_HANDLER(phb);
4251         }
4252     }
4253     return NULL;
4254 }
4255 
4256 static CpuInstanceProperties
4257 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4258 {
4259     CPUArchId *core_slot;
4260     MachineClass *mc = MACHINE_GET_CLASS(machine);
4261 
4262     /* make sure possible_cpu are intialized */
4263     mc->possible_cpu_arch_ids(machine);
4264     /* get CPU core slot containing thread that matches cpu_index */
4265     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4266     assert(core_slot);
4267     return core_slot->props;
4268 }
4269 
4270 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4271 {
4272     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4273 }
4274 
4275 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4276 {
4277     int i;
4278     unsigned int smp_threads = machine->smp.threads;
4279     unsigned int smp_cpus = machine->smp.cpus;
4280     const char *core_type;
4281     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4282     MachineClass *mc = MACHINE_GET_CLASS(machine);
4283 
4284     if (!mc->has_hotpluggable_cpus) {
4285         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4286     }
4287     if (machine->possible_cpus) {
4288         assert(machine->possible_cpus->len == spapr_max_cores);
4289         return machine->possible_cpus;
4290     }
4291 
4292     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4293     if (!core_type) {
4294         error_report("Unable to find sPAPR CPU Core definition");
4295         exit(1);
4296     }
4297 
4298     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4299                              sizeof(CPUArchId) * spapr_max_cores);
4300     machine->possible_cpus->len = spapr_max_cores;
4301     for (i = 0; i < machine->possible_cpus->len; i++) {
4302         int core_id = i * smp_threads;
4303 
4304         machine->possible_cpus->cpus[i].type = core_type;
4305         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4306         machine->possible_cpus->cpus[i].arch_id = core_id;
4307         machine->possible_cpus->cpus[i].props.has_core_id = true;
4308         machine->possible_cpus->cpus[i].props.core_id = core_id;
4309     }
4310     return machine->possible_cpus;
4311 }
4312 
4313 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4314                                 uint64_t *buid, hwaddr *pio,
4315                                 hwaddr *mmio32, hwaddr *mmio64,
4316                                 unsigned n_dma, uint32_t *liobns,
4317                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4318 {
4319     /*
4320      * New-style PHB window placement.
4321      *
4322      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4323      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4324      * windows.
4325      *
4326      * Some guest kernels can't work with MMIO windows above 1<<46
4327      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4328      *
4329      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4330      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4331      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4332      * 1TiB 64-bit MMIO windows for each PHB.
4333      */
4334     const uint64_t base_buid = 0x800000020000000ULL;
4335     int i;
4336 
4337     /* Sanity check natural alignments */
4338     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4339     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4340     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4341     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4342     /* Sanity check bounds */
4343     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4344                       SPAPR_PCI_MEM32_WIN_SIZE);
4345     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4346                       SPAPR_PCI_MEM64_WIN_SIZE);
4347 
4348     if (index >= SPAPR_MAX_PHBS) {
4349         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4350                    SPAPR_MAX_PHBS - 1);
4351         return false;
4352     }
4353 
4354     *buid = base_buid + index;
4355     for (i = 0; i < n_dma; ++i) {
4356         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4357     }
4358 
4359     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4360     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4361     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4362 
4363     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4364     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4365     return true;
4366 }
4367 
4368 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4369 {
4370     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4371 
4372     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4373 }
4374 
4375 static void spapr_ics_resend(XICSFabric *dev)
4376 {
4377     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4378 
4379     ics_resend(spapr->ics);
4380 }
4381 
4382 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4383 {
4384     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4385 
4386     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4387 }
4388 
4389 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4390                                  Monitor *mon)
4391 {
4392     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4393 
4394     spapr_irq_print_info(spapr, mon);
4395     monitor_printf(mon, "irqchip: %s\n",
4396                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4397 }
4398 
4399 /*
4400  * This is a XIVE only operation
4401  */
4402 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4403                            uint8_t nvt_blk, uint32_t nvt_idx,
4404                            bool cam_ignore, uint8_t priority,
4405                            uint32_t logic_serv, XiveTCTXMatch *match)
4406 {
4407     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4408     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4409     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4410     int count;
4411 
4412     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4413                            priority, logic_serv, match);
4414     if (count < 0) {
4415         return count;
4416     }
4417 
4418     /*
4419      * When we implement the save and restore of the thread interrupt
4420      * contexts in the enter/exit CPU handlers of the machine and the
4421      * escalations in QEMU, we should be able to handle non dispatched
4422      * vCPUs.
4423      *
4424      * Until this is done, the sPAPR machine should find at least one
4425      * matching context always.
4426      */
4427     if (count == 0) {
4428         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4429                       nvt_blk, nvt_idx);
4430     }
4431 
4432     return count;
4433 }
4434 
4435 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4436 {
4437     return cpu->vcpu_id;
4438 }
4439 
4440 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4441 {
4442     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4443     MachineState *ms = MACHINE(spapr);
4444     int vcpu_id;
4445 
4446     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4447 
4448     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4449         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4450         error_append_hint(errp, "Adjust the number of cpus to %d "
4451                           "or try to raise the number of threads per core\n",
4452                           vcpu_id * ms->smp.threads / spapr->vsmt);
4453         return false;
4454     }
4455 
4456     cpu->vcpu_id = vcpu_id;
4457     return true;
4458 }
4459 
4460 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4461 {
4462     CPUState *cs;
4463 
4464     CPU_FOREACH(cs) {
4465         PowerPCCPU *cpu = POWERPC_CPU(cs);
4466 
4467         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4468             return cpu;
4469         }
4470     }
4471 
4472     return NULL;
4473 }
4474 
4475 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4476 {
4477     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4478 
4479     /* These are only called by TCG, KVM maintains dispatch state */
4480 
4481     spapr_cpu->prod = false;
4482     if (spapr_cpu->vpa_addr) {
4483         CPUState *cs = CPU(cpu);
4484         uint32_t dispatch;
4485 
4486         dispatch = ldl_be_phys(cs->as,
4487                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4488         dispatch++;
4489         if ((dispatch & 1) != 0) {
4490             qemu_log_mask(LOG_GUEST_ERROR,
4491                           "VPA: incorrect dispatch counter value for "
4492                           "dispatched partition %u, correcting.\n", dispatch);
4493             dispatch++;
4494         }
4495         stl_be_phys(cs->as,
4496                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4497     }
4498 }
4499 
4500 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4501 {
4502     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4503 
4504     if (spapr_cpu->vpa_addr) {
4505         CPUState *cs = CPU(cpu);
4506         uint32_t dispatch;
4507 
4508         dispatch = ldl_be_phys(cs->as,
4509                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4510         dispatch++;
4511         if ((dispatch & 1) != 1) {
4512             qemu_log_mask(LOG_GUEST_ERROR,
4513                           "VPA: incorrect dispatch counter value for "
4514                           "preempted partition %u, correcting.\n", dispatch);
4515             dispatch++;
4516         }
4517         stl_be_phys(cs->as,
4518                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4519     }
4520 }
4521 
4522 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4523 {
4524     MachineClass *mc = MACHINE_CLASS(oc);
4525     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4526     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4527     NMIClass *nc = NMI_CLASS(oc);
4528     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4529     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4530     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4531     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4532     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4533     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4534 
4535     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4536     mc->ignore_boot_device_suffixes = true;
4537 
4538     /*
4539      * We set up the default / latest behaviour here.  The class_init
4540      * functions for the specific versioned machine types can override
4541      * these details for backwards compatibility
4542      */
4543     mc->init = spapr_machine_init;
4544     mc->reset = spapr_machine_reset;
4545     mc->block_default_type = IF_SCSI;
4546 
4547     /*
4548      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4549      * should be limited by the host capability instead of hardcoded.
4550      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4551      * guests are welcome to have as many CPUs as the host are capable
4552      * of emulate.
4553      */
4554     mc->max_cpus = INT32_MAX;
4555 
4556     mc->no_parallel = 1;
4557     mc->default_boot_order = "";
4558     mc->default_ram_size = 512 * MiB;
4559     mc->default_ram_id = "ppc_spapr.ram";
4560     mc->default_display = "std";
4561     mc->kvm_type = spapr_kvm_type;
4562     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4563     mc->pci_allow_0_address = true;
4564     assert(!mc->get_hotplug_handler);
4565     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4566     hc->pre_plug = spapr_machine_device_pre_plug;
4567     hc->plug = spapr_machine_device_plug;
4568     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4569     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4570     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4571     hc->unplug_request = spapr_machine_device_unplug_request;
4572     hc->unplug = spapr_machine_device_unplug;
4573 
4574     smc->dr_lmb_enabled = true;
4575     smc->update_dt_enabled = true;
4576     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4577     mc->has_hotpluggable_cpus = true;
4578     mc->nvdimm_supported = true;
4579     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4580     fwc->get_dev_path = spapr_get_fw_dev_path;
4581     nc->nmi_monitor_handler = spapr_nmi;
4582     smc->phb_placement = spapr_phb_placement;
4583     vhc->hypercall = emulate_spapr_hypercall;
4584     vhc->hpt_mask = spapr_hpt_mask;
4585     vhc->map_hptes = spapr_map_hptes;
4586     vhc->unmap_hptes = spapr_unmap_hptes;
4587     vhc->hpte_set_c = spapr_hpte_set_c;
4588     vhc->hpte_set_r = spapr_hpte_set_r;
4589     vhc->get_pate = spapr_get_pate;
4590     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4591     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4592     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4593     xic->ics_get = spapr_ics_get;
4594     xic->ics_resend = spapr_ics_resend;
4595     xic->icp_get = spapr_icp_get;
4596     ispc->print_info = spapr_pic_print_info;
4597     /* Force NUMA node memory size to be a multiple of
4598      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4599      * in which LMBs are represented and hot-added
4600      */
4601     mc->numa_mem_align_shift = 28;
4602     mc->auto_enable_numa = true;
4603 
4604     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4605     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4606     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4607     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4608     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4609     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4610     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4611     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4612     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4613     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4614     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4615     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4616     spapr_caps_add_properties(smc);
4617     smc->irq = &spapr_irq_dual;
4618     smc->dr_phb_enabled = true;
4619     smc->linux_pci_probe = true;
4620     smc->smp_threads_vsmt = true;
4621     smc->nr_xirqs = SPAPR_NR_XIRQS;
4622     xfc->match_nvt = spapr_match_nvt;
4623     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4624     vmc->quiesce = spapr_vof_quiesce;
4625     vmc->setprop = spapr_vof_setprop;
4626 }
4627 
4628 static const TypeInfo spapr_machine_info = {
4629     .name          = TYPE_SPAPR_MACHINE,
4630     .parent        = TYPE_MACHINE,
4631     .abstract      = true,
4632     .instance_size = sizeof(SpaprMachineState),
4633     .instance_init = spapr_instance_init,
4634     .instance_finalize = spapr_machine_finalizefn,
4635     .class_size    = sizeof(SpaprMachineClass),
4636     .class_init    = spapr_machine_class_init,
4637     .interfaces = (InterfaceInfo[]) {
4638         { TYPE_FW_PATH_PROVIDER },
4639         { TYPE_NMI },
4640         { TYPE_HOTPLUG_HANDLER },
4641         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4642         { TYPE_XICS_FABRIC },
4643         { TYPE_INTERRUPT_STATS_PROVIDER },
4644         { TYPE_XIVE_FABRIC },
4645         { TYPE_VOF_MACHINE_IF },
4646         { }
4647     },
4648 };
4649 
4650 static void spapr_machine_latest_class_options(MachineClass *mc)
4651 {
4652     mc->alias = "pseries";
4653     mc->is_default = true;
4654 }
4655 
4656 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4657     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4658                                                     void *data)      \
4659     {                                                                \
4660         MachineClass *mc = MACHINE_CLASS(oc);                        \
4661         spapr_machine_##suffix##_class_options(mc);                  \
4662         if (latest) {                                                \
4663             spapr_machine_latest_class_options(mc);                  \
4664         }                                                            \
4665     }                                                                \
4666     static const TypeInfo spapr_machine_##suffix##_info = {          \
4667         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4668         .parent = TYPE_SPAPR_MACHINE,                                \
4669         .class_init = spapr_machine_##suffix##_class_init,           \
4670     };                                                               \
4671     static void spapr_machine_register_##suffix(void)                \
4672     {                                                                \
4673         type_register(&spapr_machine_##suffix##_info);               \
4674     }                                                                \
4675     type_init(spapr_machine_register_##suffix)
4676 
4677 /*
4678  * pseries-7.0
4679  */
4680 static void spapr_machine_7_0_class_options(MachineClass *mc)
4681 {
4682     /* Defaults for the latest behaviour inherited from the base class */
4683 }
4684 
4685 DEFINE_SPAPR_MACHINE(7_0, "7.0", true);
4686 
4687 /*
4688  * pseries-6.2
4689  */
4690 static void spapr_machine_6_2_class_options(MachineClass *mc)
4691 {
4692     spapr_machine_7_0_class_options(mc);
4693     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4694 }
4695 
4696 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4697 
4698 /*
4699  * pseries-6.1
4700  */
4701 static void spapr_machine_6_1_class_options(MachineClass *mc)
4702 {
4703     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4704 
4705     spapr_machine_6_2_class_options(mc);
4706     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4707     smc->pre_6_2_numa_affinity = true;
4708     mc->smp_props.prefer_sockets = true;
4709 }
4710 
4711 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4712 
4713 /*
4714  * pseries-6.0
4715  */
4716 static void spapr_machine_6_0_class_options(MachineClass *mc)
4717 {
4718     spapr_machine_6_1_class_options(mc);
4719     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4720 }
4721 
4722 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4723 
4724 /*
4725  * pseries-5.2
4726  */
4727 static void spapr_machine_5_2_class_options(MachineClass *mc)
4728 {
4729     spapr_machine_6_0_class_options(mc);
4730     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4731 }
4732 
4733 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4734 
4735 /*
4736  * pseries-5.1
4737  */
4738 static void spapr_machine_5_1_class_options(MachineClass *mc)
4739 {
4740     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4741 
4742     spapr_machine_5_2_class_options(mc);
4743     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4744     smc->pre_5_2_numa_associativity = true;
4745 }
4746 
4747 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4748 
4749 /*
4750  * pseries-5.0
4751  */
4752 static void spapr_machine_5_0_class_options(MachineClass *mc)
4753 {
4754     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4755     static GlobalProperty compat[] = {
4756         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4757     };
4758 
4759     spapr_machine_5_1_class_options(mc);
4760     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4761     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4762     mc->numa_mem_supported = true;
4763     smc->pre_5_1_assoc_refpoints = true;
4764 }
4765 
4766 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4767 
4768 /*
4769  * pseries-4.2
4770  */
4771 static void spapr_machine_4_2_class_options(MachineClass *mc)
4772 {
4773     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4774 
4775     spapr_machine_5_0_class_options(mc);
4776     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4777     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4778     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4779     smc->rma_limit = 16 * GiB;
4780     mc->nvdimm_supported = false;
4781 }
4782 
4783 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4784 
4785 /*
4786  * pseries-4.1
4787  */
4788 static void spapr_machine_4_1_class_options(MachineClass *mc)
4789 {
4790     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4791     static GlobalProperty compat[] = {
4792         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4793         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4794     };
4795 
4796     spapr_machine_4_2_class_options(mc);
4797     smc->linux_pci_probe = false;
4798     smc->smp_threads_vsmt = false;
4799     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4800     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4801 }
4802 
4803 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4804 
4805 /*
4806  * pseries-4.0
4807  */
4808 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4809                               uint64_t *buid, hwaddr *pio,
4810                               hwaddr *mmio32, hwaddr *mmio64,
4811                               unsigned n_dma, uint32_t *liobns,
4812                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4813 {
4814     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4815                              liobns, nv2gpa, nv2atsd, errp)) {
4816         return false;
4817     }
4818 
4819     *nv2gpa = 0;
4820     *nv2atsd = 0;
4821     return true;
4822 }
4823 static void spapr_machine_4_0_class_options(MachineClass *mc)
4824 {
4825     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4826 
4827     spapr_machine_4_1_class_options(mc);
4828     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4829     smc->phb_placement = phb_placement_4_0;
4830     smc->irq = &spapr_irq_xics;
4831     smc->pre_4_1_migration = true;
4832 }
4833 
4834 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4835 
4836 /*
4837  * pseries-3.1
4838  */
4839 static void spapr_machine_3_1_class_options(MachineClass *mc)
4840 {
4841     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4842 
4843     spapr_machine_4_0_class_options(mc);
4844     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4845 
4846     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4847     smc->update_dt_enabled = false;
4848     smc->dr_phb_enabled = false;
4849     smc->broken_host_serial_model = true;
4850     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4851     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4852     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4853     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4854 }
4855 
4856 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4857 
4858 /*
4859  * pseries-3.0
4860  */
4861 
4862 static void spapr_machine_3_0_class_options(MachineClass *mc)
4863 {
4864     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4865 
4866     spapr_machine_3_1_class_options(mc);
4867     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4868 
4869     smc->legacy_irq_allocation = true;
4870     smc->nr_xirqs = 0x400;
4871     smc->irq = &spapr_irq_xics_legacy;
4872 }
4873 
4874 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4875 
4876 /*
4877  * pseries-2.12
4878  */
4879 static void spapr_machine_2_12_class_options(MachineClass *mc)
4880 {
4881     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4882     static GlobalProperty compat[] = {
4883         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4884         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4885     };
4886 
4887     spapr_machine_3_0_class_options(mc);
4888     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4889     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4890 
4891     /* We depend on kvm_enabled() to choose a default value for the
4892      * hpt-max-page-size capability. Of course we can't do it here
4893      * because this is too early and the HW accelerator isn't initialzed
4894      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4895      */
4896     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4897 }
4898 
4899 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4900 
4901 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4902 {
4903     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4904 
4905     spapr_machine_2_12_class_options(mc);
4906     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4907     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4908     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4909 }
4910 
4911 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4912 
4913 /*
4914  * pseries-2.11
4915  */
4916 
4917 static void spapr_machine_2_11_class_options(MachineClass *mc)
4918 {
4919     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4920 
4921     spapr_machine_2_12_class_options(mc);
4922     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4923     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4924 }
4925 
4926 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4927 
4928 /*
4929  * pseries-2.10
4930  */
4931 
4932 static void spapr_machine_2_10_class_options(MachineClass *mc)
4933 {
4934     spapr_machine_2_11_class_options(mc);
4935     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4936 }
4937 
4938 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4939 
4940 /*
4941  * pseries-2.9
4942  */
4943 
4944 static void spapr_machine_2_9_class_options(MachineClass *mc)
4945 {
4946     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4947     static GlobalProperty compat[] = {
4948         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4949     };
4950 
4951     spapr_machine_2_10_class_options(mc);
4952     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4953     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4954     smc->pre_2_10_has_unused_icps = true;
4955     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4956 }
4957 
4958 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4959 
4960 /*
4961  * pseries-2.8
4962  */
4963 
4964 static void spapr_machine_2_8_class_options(MachineClass *mc)
4965 {
4966     static GlobalProperty compat[] = {
4967         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4968     };
4969 
4970     spapr_machine_2_9_class_options(mc);
4971     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4972     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4973     mc->numa_mem_align_shift = 23;
4974 }
4975 
4976 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4977 
4978 /*
4979  * pseries-2.7
4980  */
4981 
4982 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4983                               uint64_t *buid, hwaddr *pio,
4984                               hwaddr *mmio32, hwaddr *mmio64,
4985                               unsigned n_dma, uint32_t *liobns,
4986                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4987 {
4988     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4989     const uint64_t base_buid = 0x800000020000000ULL;
4990     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4991     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4992     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4993     const uint32_t max_index = 255;
4994     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4995 
4996     uint64_t ram_top = MACHINE(spapr)->ram_size;
4997     hwaddr phb0_base, phb_base;
4998     int i;
4999 
5000     /* Do we have device memory? */
5001     if (MACHINE(spapr)->maxram_size > ram_top) {
5002         /* Can't just use maxram_size, because there may be an
5003          * alignment gap between normal and device memory regions
5004          */
5005         ram_top = MACHINE(spapr)->device_memory->base +
5006             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5007     }
5008 
5009     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5010 
5011     if (index > max_index) {
5012         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5013                    max_index);
5014         return false;
5015     }
5016 
5017     *buid = base_buid + index;
5018     for (i = 0; i < n_dma; ++i) {
5019         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5020     }
5021 
5022     phb_base = phb0_base + index * phb_spacing;
5023     *pio = phb_base + pio_offset;
5024     *mmio32 = phb_base + mmio_offset;
5025     /*
5026      * We don't set the 64-bit MMIO window, relying on the PHB's
5027      * fallback behaviour of automatically splitting a large "32-bit"
5028      * window into contiguous 32-bit and 64-bit windows
5029      */
5030 
5031     *nv2gpa = 0;
5032     *nv2atsd = 0;
5033     return true;
5034 }
5035 
5036 static void spapr_machine_2_7_class_options(MachineClass *mc)
5037 {
5038     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5039     static GlobalProperty compat[] = {
5040         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5041         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5042         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5043         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5044     };
5045 
5046     spapr_machine_2_8_class_options(mc);
5047     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5048     mc->default_machine_opts = "modern-hotplug-events=off";
5049     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5050     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5051     smc->phb_placement = phb_placement_2_7;
5052 }
5053 
5054 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5055 
5056 /*
5057  * pseries-2.6
5058  */
5059 
5060 static void spapr_machine_2_6_class_options(MachineClass *mc)
5061 {
5062     static GlobalProperty compat[] = {
5063         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5064     };
5065 
5066     spapr_machine_2_7_class_options(mc);
5067     mc->has_hotpluggable_cpus = false;
5068     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5069     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5070 }
5071 
5072 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5073 
5074 /*
5075  * pseries-2.5
5076  */
5077 
5078 static void spapr_machine_2_5_class_options(MachineClass *mc)
5079 {
5080     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5081     static GlobalProperty compat[] = {
5082         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5083     };
5084 
5085     spapr_machine_2_6_class_options(mc);
5086     smc->use_ohci_by_default = true;
5087     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5088     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5089 }
5090 
5091 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5092 
5093 /*
5094  * pseries-2.4
5095  */
5096 
5097 static void spapr_machine_2_4_class_options(MachineClass *mc)
5098 {
5099     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5100 
5101     spapr_machine_2_5_class_options(mc);
5102     smc->dr_lmb_enabled = false;
5103     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5104 }
5105 
5106 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5107 
5108 /*
5109  * pseries-2.3
5110  */
5111 
5112 static void spapr_machine_2_3_class_options(MachineClass *mc)
5113 {
5114     static GlobalProperty compat[] = {
5115         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5116     };
5117     spapr_machine_2_4_class_options(mc);
5118     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5119     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5120 }
5121 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5122 
5123 /*
5124  * pseries-2.2
5125  */
5126 
5127 static void spapr_machine_2_2_class_options(MachineClass *mc)
5128 {
5129     static GlobalProperty compat[] = {
5130         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5131     };
5132 
5133     spapr_machine_2_3_class_options(mc);
5134     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5135     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5136     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5137 }
5138 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5139 
5140 /*
5141  * pseries-2.1
5142  */
5143 
5144 static void spapr_machine_2_1_class_options(MachineClass *mc)
5145 {
5146     spapr_machine_2_2_class_options(mc);
5147     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5148 }
5149 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5150 
5151 static void spapr_machine_register_types(void)
5152 {
5153     type_register_static(&spapr_machine_info);
5154 }
5155 
5156 type_init(spapr_machine_register_types)
5157