xref: /openbmc/qemu/hw/ppc/spapr.c (revision 1eee9950264867232a8771e60f3955b84acf538f)
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 "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 #include "hw/ppc/spapr_numa.h"
85 
86 #include "monitor/monitor.h"
87 
88 #include <libfdt.h>
89 
90 /* SLOF memory layout:
91  *
92  * SLOF raw image loaded at 0, copies its romfs right below the flat
93  * device-tree, then position SLOF itself 31M below that
94  *
95  * So we set FW_OVERHEAD to 40MB which should account for all of that
96  * and more
97  *
98  * We load our kernel at 4M, leaving space for SLOF initial image
99  */
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 static void spapr_dt_pa_features(SpaprMachineState *spapr,
221                                  PowerPCCPU *cpu,
222                                  void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
318                                 hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
344 {
345     MemoryDeviceInfoList *info;
346 
347     for (info = list; info; info = info->next) {
348         MemoryDeviceInfo *value = info->value;
349 
350         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
351             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
352 
353             if (addr >= pcdimm_info->addr &&
354                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
355                 return pcdimm_info->node;
356             }
357         }
358     }
359 
360     return -1;
361 }
362 
363 struct sPAPRDrconfCellV2 {
364      uint32_t seq_lmbs;
365      uint64_t base_addr;
366      uint32_t drc_index;
367      uint32_t aa_index;
368      uint32_t flags;
369 } QEMU_PACKED;
370 
371 typedef struct DrconfCellQueue {
372     struct sPAPRDrconfCellV2 cell;
373     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
374 } DrconfCellQueue;
375 
376 static DrconfCellQueue *
377 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
378                       uint32_t drc_index, uint32_t aa_index,
379                       uint32_t flags)
380 {
381     DrconfCellQueue *elem;
382 
383     elem = g_malloc0(sizeof(*elem));
384     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
385     elem->cell.base_addr = cpu_to_be64(base_addr);
386     elem->cell.drc_index = cpu_to_be32(drc_index);
387     elem->cell.aa_index = cpu_to_be32(aa_index);
388     elem->cell.flags = cpu_to_be32(flags);
389 
390     return elem;
391 }
392 
393 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
394                                       int offset, MemoryDeviceInfoList *dimms)
395 {
396     MachineState *machine = MACHINE(spapr);
397     uint8_t *int_buf, *cur_index;
398     int ret;
399     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
400     uint64_t addr, cur_addr, size;
401     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
402     uint64_t mem_end = machine->device_memory->base +
403                        memory_region_size(&machine->device_memory->mr);
404     uint32_t node, buf_len, nr_entries = 0;
405     SpaprDrc *drc;
406     DrconfCellQueue *elem, *next;
407     MemoryDeviceInfoList *info;
408     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
409         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
410 
411     /* Entry to cover RAM and the gap area */
412     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
413                                  SPAPR_LMB_FLAGS_RESERVED |
414                                  SPAPR_LMB_FLAGS_DRC_INVALID);
415     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
416     nr_entries++;
417 
418     cur_addr = machine->device_memory->base;
419     for (info = dimms; info; info = info->next) {
420         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
421 
422         addr = di->addr;
423         size = di->size;
424         node = di->node;
425 
426         /*
427          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
428          * area is marked hotpluggable in the next iteration for the bigger
429          * chunk including the NVDIMM occupied area.
430          */
431         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
432             continue;
433 
434         /* Entry for hot-pluggable area */
435         if (cur_addr < addr) {
436             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
437             g_assert(drc);
438             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
439                                          cur_addr, spapr_drc_index(drc), -1, 0);
440             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
441             nr_entries++;
442         }
443 
444         /* Entry for DIMM */
445         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
446         g_assert(drc);
447         elem = spapr_get_drconf_cell(size / lmb_size, addr,
448                                      spapr_drc_index(drc), node,
449                                      (SPAPR_LMB_FLAGS_ASSIGNED |
450                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
451         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
452         nr_entries++;
453         cur_addr = addr + size;
454     }
455 
456     /* Entry for remaining hotpluggable area */
457     if (cur_addr < mem_end) {
458         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
459         g_assert(drc);
460         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
461                                      cur_addr, spapr_drc_index(drc), -1, 0);
462         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
463         nr_entries++;
464     }
465 
466     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
467     int_buf = cur_index = g_malloc0(buf_len);
468     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
469     cur_index += sizeof(nr_entries);
470 
471     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
472         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
473         cur_index += sizeof(elem->cell);
474         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
475         g_free(elem);
476     }
477 
478     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
479     g_free(int_buf);
480     if (ret < 0) {
481         return -1;
482     }
483     return 0;
484 }
485 
486 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
487                                    int offset, MemoryDeviceInfoList *dimms)
488 {
489     MachineState *machine = MACHINE(spapr);
490     int i, ret;
491     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
492     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
493     uint32_t nr_lmbs = (machine->device_memory->base +
494                        memory_region_size(&machine->device_memory->mr)) /
495                        lmb_size;
496     uint32_t *int_buf, *cur_index, buf_len;
497 
498     /*
499      * Allocate enough buffer size to fit in ibm,dynamic-memory
500      */
501     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
502     cur_index = int_buf = g_malloc0(buf_len);
503     int_buf[0] = cpu_to_be32(nr_lmbs);
504     cur_index++;
505     for (i = 0; i < nr_lmbs; i++) {
506         uint64_t addr = i * lmb_size;
507         uint32_t *dynamic_memory = cur_index;
508 
509         if (i >= device_lmb_start) {
510             SpaprDrc *drc;
511 
512             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
513             g_assert(drc);
514 
515             dynamic_memory[0] = cpu_to_be32(addr >> 32);
516             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
517             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
518             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
519             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
520             if (memory_region_present(get_system_memory(), addr)) {
521                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
522             } else {
523                 dynamic_memory[5] = cpu_to_be32(0);
524             }
525         } else {
526             /*
527              * LMB information for RMA, boot time RAM and gap b/n RAM and
528              * device memory region -- all these are marked as reserved
529              * and as having no valid DRC.
530              */
531             dynamic_memory[0] = cpu_to_be32(addr >> 32);
532             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
533             dynamic_memory[2] = cpu_to_be32(0);
534             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
535             dynamic_memory[4] = cpu_to_be32(-1);
536             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
537                                             SPAPR_LMB_FLAGS_DRC_INVALID);
538         }
539 
540         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
541     }
542     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
543     g_free(int_buf);
544     if (ret < 0) {
545         return -1;
546     }
547     return 0;
548 }
549 
550 /*
551  * Adds ibm,dynamic-reconfiguration-memory node.
552  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
553  * of this device tree node.
554  */
555 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
556                                                    void *fdt)
557 {
558     MachineState *machine = MACHINE(spapr);
559     int nb_numa_nodes = machine->numa_state->num_nodes;
560     int ret, i, offset;
561     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
562     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
563                                 cpu_to_be32(lmb_size & 0xffffffff)};
564     uint32_t *int_buf, *cur_index, buf_len;
565     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
566     MemoryDeviceInfoList *dimms = NULL;
567 
568     /*
569      * Don't create the node if there is no device memory
570      */
571     if (machine->ram_size == machine->maxram_size) {
572         return 0;
573     }
574 
575     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
576 
577     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
578                     sizeof(prop_lmb_size));
579     if (ret < 0) {
580         return ret;
581     }
582 
583     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
584     if (ret < 0) {
585         return ret;
586     }
587 
588     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
589     if (ret < 0) {
590         return ret;
591     }
592 
593     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
594     dimms = qmp_memory_device_list();
595     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
596         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
597     } else {
598         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
599     }
600     qapi_free_MemoryDeviceInfoList(dimms);
601 
602     if (ret < 0) {
603         return ret;
604     }
605 
606     /* ibm,associativity-lookup-arrays */
607     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
608     cur_index = int_buf = g_malloc0(buf_len);
609     int_buf[0] = cpu_to_be32(nr_nodes);
610     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
611     cur_index += 2;
612     for (i = 0; i < nr_nodes; i++) {
613         uint32_t associativity[] = {
614             cpu_to_be32(0x0),
615             cpu_to_be32(0x0),
616             cpu_to_be32(0x0),
617             cpu_to_be32(i)
618         };
619         memcpy(cur_index, associativity, sizeof(associativity));
620         cur_index += 4;
621     }
622     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
623             (cur_index - int_buf) * sizeof(uint32_t));
624     g_free(int_buf);
625 
626     return ret;
627 }
628 
629 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
630 {
631     MachineState *machine = MACHINE(spapr);
632     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
633     hwaddr mem_start, node_size;
634     int i, nb_nodes = machine->numa_state->num_nodes;
635     NodeInfo *nodes = machine->numa_state->nodes;
636 
637     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
638         if (!nodes[i].node_mem) {
639             continue;
640         }
641         if (mem_start >= machine->ram_size) {
642             node_size = 0;
643         } else {
644             node_size = nodes[i].node_mem;
645             if (node_size > machine->ram_size - mem_start) {
646                 node_size = machine->ram_size - mem_start;
647             }
648         }
649         if (!mem_start) {
650             /* spapr_machine_init() checks for rma_size <= node0_size
651              * already */
652             spapr_dt_memory_node(fdt, i, 0, spapr->rma_size);
653             mem_start += spapr->rma_size;
654             node_size -= spapr->rma_size;
655         }
656         for ( ; node_size; ) {
657             hwaddr sizetmp = pow2floor(node_size);
658 
659             /* mem_start != 0 here */
660             if (ctzl(mem_start) < ctzl(sizetmp)) {
661                 sizetmp = 1ULL << ctzl(mem_start);
662             }
663 
664             spapr_dt_memory_node(fdt, i, mem_start, sizetmp);
665             node_size -= sizetmp;
666             mem_start += sizetmp;
667         }
668     }
669 
670     /* Generate ibm,dynamic-reconfiguration-memory node if required */
671     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
672         int ret;
673 
674         g_assert(smc->dr_lmb_enabled);
675         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
676         if (ret) {
677             return ret;
678         }
679     }
680 
681     return 0;
682 }
683 
684 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
685                          SpaprMachineState *spapr)
686 {
687     MachineState *ms = MACHINE(spapr);
688     PowerPCCPU *cpu = POWERPC_CPU(cs);
689     CPUPPCState *env = &cpu->env;
690     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
691     int index = spapr_get_vcpu_id(cpu);
692     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
693                        0xffffffff, 0xffffffff};
694     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
695         : SPAPR_TIMEBASE_FREQ;
696     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
697     uint32_t page_sizes_prop[64];
698     size_t page_sizes_prop_size;
699     unsigned int smp_threads = ms->smp.threads;
700     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
701     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
702     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
703     SpaprDrc *drc;
704     int drc_index;
705     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
706     int i;
707 
708     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
709     if (drc) {
710         drc_index = spapr_drc_index(drc);
711         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
712     }
713 
714     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
715     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
716 
717     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
718     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
719                            env->dcache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
721                            env->dcache_line_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
723                            env->icache_line_size)));
724     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
725                            env->icache_line_size)));
726 
727     if (pcc->l1_dcache_size) {
728         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
729                                pcc->l1_dcache_size)));
730     } else {
731         warn_report("Unknown L1 dcache size for cpu");
732     }
733     if (pcc->l1_icache_size) {
734         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
735                                pcc->l1_icache_size)));
736     } else {
737         warn_report("Unknown L1 icache size for cpu");
738     }
739 
740     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
741     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
742     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
743     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
744     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
745     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
746 
747     if (env->spr_cb[SPR_PURR].oea_read) {
748         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
749     }
750     if (env->spr_cb[SPR_SPURR].oea_read) {
751         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
752     }
753 
754     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
755         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
756                           segs, sizeof(segs))));
757     }
758 
759     /* Advertise VSX (vector extensions) if available
760      *   1               == VMX / Altivec available
761      *   2               == VSX available
762      *
763      * Only CPUs for which we create core types in spapr_cpu_core.c
764      * are possible, and all of those have VMX */
765     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
766         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
767     } else {
768         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
769     }
770 
771     /* Advertise DFP (Decimal Floating Point) if available
772      *   0 / no property == no DFP
773      *   1               == DFP available */
774     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
775         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
776     }
777 
778     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
779                                                       sizeof(page_sizes_prop));
780     if (page_sizes_prop_size) {
781         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
782                           page_sizes_prop, page_sizes_prop_size)));
783     }
784 
785     spapr_dt_pa_features(spapr, cpu, fdt, offset);
786 
787     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
788                            cs->cpu_index / vcpus_per_socket)));
789 
790     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
791                       pft_size_prop, sizeof(pft_size_prop))));
792 
793     if (ms->numa_state->num_nodes > 1) {
794         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
795     }
796 
797     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
798 
799     if (pcc->radix_page_info) {
800         for (i = 0; i < pcc->radix_page_info->count; i++) {
801             radix_AP_encodings[i] =
802                 cpu_to_be32(pcc->radix_page_info->entries[i]);
803         }
804         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
805                           radix_AP_encodings,
806                           pcc->radix_page_info->count *
807                           sizeof(radix_AP_encodings[0]))));
808     }
809 
810     /*
811      * We set this property to let the guest know that it can use the large
812      * decrementer and its width in bits.
813      */
814     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
815         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
816                               pcc->lrg_decr_bits)));
817 }
818 
819 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
820 {
821     CPUState **rev;
822     CPUState *cs;
823     int n_cpus;
824     int cpus_offset;
825     char *nodename;
826     int i;
827 
828     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
829     _FDT(cpus_offset);
830     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
831     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
832 
833     /*
834      * We walk the CPUs in reverse order to ensure that CPU DT nodes
835      * created by fdt_add_subnode() end up in the right order in FDT
836      * for the guest kernel the enumerate the CPUs correctly.
837      *
838      * The CPU list cannot be traversed in reverse order, so we need
839      * to do extra work.
840      */
841     n_cpus = 0;
842     rev = NULL;
843     CPU_FOREACH(cs) {
844         rev = g_renew(CPUState *, rev, n_cpus + 1);
845         rev[n_cpus++] = cs;
846     }
847 
848     for (i = n_cpus - 1; i >= 0; i--) {
849         CPUState *cs = rev[i];
850         PowerPCCPU *cpu = POWERPC_CPU(cs);
851         int index = spapr_get_vcpu_id(cpu);
852         DeviceClass *dc = DEVICE_GET_CLASS(cs);
853         int offset;
854 
855         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
856             continue;
857         }
858 
859         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
860         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
861         g_free(nodename);
862         _FDT(offset);
863         spapr_dt_cpu(cs, fdt, offset, spapr);
864     }
865 
866     g_free(rev);
867 }
868 
869 static int spapr_dt_rng(void *fdt)
870 {
871     int node;
872     int ret;
873 
874     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
875     if (node <= 0) {
876         return -1;
877     }
878     ret = fdt_setprop_string(fdt, node, "device_type",
879                              "ibm,platform-facilities");
880     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
881     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
882 
883     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
884     if (node <= 0) {
885         return -1;
886     }
887     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
888 
889     return ret ? -1 : 0;
890 }
891 
892 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
893 {
894     MachineState *ms = MACHINE(spapr);
895     int rtas;
896     GString *hypertas = g_string_sized_new(256);
897     GString *qemu_hypertas = g_string_sized_new(256);
898     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
899         memory_region_size(&MACHINE(spapr)->device_memory->mr);
900     uint32_t lrdr_capacity[] = {
901         cpu_to_be32(max_device_addr >> 32),
902         cpu_to_be32(max_device_addr & 0xffffffff),
903         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
904         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
905         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
906     };
907 
908     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
909 
910     /* hypertas */
911     add_str(hypertas, "hcall-pft");
912     add_str(hypertas, "hcall-term");
913     add_str(hypertas, "hcall-dabr");
914     add_str(hypertas, "hcall-interrupt");
915     add_str(hypertas, "hcall-tce");
916     add_str(hypertas, "hcall-vio");
917     add_str(hypertas, "hcall-splpar");
918     add_str(hypertas, "hcall-join");
919     add_str(hypertas, "hcall-bulk");
920     add_str(hypertas, "hcall-set-mode");
921     add_str(hypertas, "hcall-sprg0");
922     add_str(hypertas, "hcall-copy");
923     add_str(hypertas, "hcall-debug");
924     add_str(hypertas, "hcall-vphn");
925     add_str(qemu_hypertas, "hcall-memop1");
926 
927     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
928         add_str(hypertas, "hcall-multi-tce");
929     }
930 
931     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
932         add_str(hypertas, "hcall-hpt-resize");
933     }
934 
935     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
936                      hypertas->str, hypertas->len));
937     g_string_free(hypertas, TRUE);
938     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
939                      qemu_hypertas->str, qemu_hypertas->len));
940     g_string_free(qemu_hypertas, TRUE);
941 
942     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
943 
944     /*
945      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
946      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
947      *
948      * The system reset requirements are driven by existing Linux and PowerVM
949      * implementation which (contrary to PAPR) saves r3 in the error log
950      * structure like machine check, so Linux expects to find the saved r3
951      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
952      * does not look at the error value).
953      *
954      * System reset interrupts are not subject to interlock like machine
955      * check, so this memory area could be corrupted if the sreset is
956      * interrupted by a machine check (or vice versa) if it was shared. To
957      * prevent this, system reset uses per-CPU areas for the sreset save
958      * area. A system reset that interrupts a system reset handler could
959      * still overwrite this area, but Linux doesn't try to recover in that
960      * case anyway.
961      *
962      * The extra 8 bytes is required because Linux's FWNMI error log check
963      * is off-by-one.
964      */
965     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
966 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
967     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
968                           RTAS_ERROR_LOG_MAX));
969     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
970                           RTAS_EVENT_SCAN_RATE));
971 
972     g_assert(msi_nonbroken);
973     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
974 
975     /*
976      * According to PAPR, rtas ibm,os-term does not guarantee a return
977      * back to the guest cpu.
978      *
979      * While an additional ibm,extended-os-term property indicates
980      * that rtas call return will always occur. Set this property.
981      */
982     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
983 
984     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
985                      lrdr_capacity, sizeof(lrdr_capacity)));
986 
987     spapr_dt_rtas_tokens(fdt, rtas);
988 }
989 
990 /*
991  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
992  * and the XIVE features that the guest may request and thus the valid
993  * values for bytes 23..26 of option vector 5:
994  */
995 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
996                                           int chosen)
997 {
998     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
999 
1000     char val[2 * 4] = {
1001         23, 0x00, /* XICS / XIVE mode */
1002         24, 0x00, /* Hash/Radix, filled in below. */
1003         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1004         26, 0x40, /* Radix options: GTSE == yes. */
1005     };
1006 
1007     if (spapr->irq->xics && spapr->irq->xive) {
1008         val[1] = SPAPR_OV5_XIVE_BOTH;
1009     } else if (spapr->irq->xive) {
1010         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1011     } else {
1012         assert(spapr->irq->xics);
1013         val[1] = SPAPR_OV5_XIVE_LEGACY;
1014     }
1015 
1016     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1017                           first_ppc_cpu->compat_pvr)) {
1018         /*
1019          * If we're in a pre POWER9 compat mode then the guest should
1020          * do hash and use the legacy interrupt mode
1021          */
1022         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1023         val[3] = 0x00; /* Hash */
1024     } else if (kvm_enabled()) {
1025         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1026             val[3] = 0x80; /* OV5_MMU_BOTH */
1027         } else if (kvmppc_has_cap_mmu_radix()) {
1028             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1029         } else {
1030             val[3] = 0x00; /* Hash */
1031         }
1032     } else {
1033         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1034         val[3] = 0xC0;
1035     }
1036     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1037                      val, sizeof(val)));
1038 }
1039 
1040 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1041 {
1042     MachineState *machine = MACHINE(spapr);
1043     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1044     int chosen;
1045 
1046     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1047 
1048     if (reset) {
1049         const char *boot_device = machine->boot_order;
1050         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1051         size_t cb = 0;
1052         char *bootlist = get_boot_devices_list(&cb);
1053 
1054         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1055             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1056                                     machine->kernel_cmdline));
1057         }
1058 
1059         if (spapr->initrd_size) {
1060             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1061                                   spapr->initrd_base));
1062             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1063                                   spapr->initrd_base + spapr->initrd_size));
1064         }
1065 
1066         if (spapr->kernel_size) {
1067             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1068                                   cpu_to_be64(spapr->kernel_size) };
1069 
1070             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1071                          &kprop, sizeof(kprop)));
1072             if (spapr->kernel_le) {
1073                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1074             }
1075         }
1076         if (boot_menu) {
1077             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1078         }
1079         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1080         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1081         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1082 
1083         if (cb && bootlist) {
1084             int i;
1085 
1086             for (i = 0; i < cb; i++) {
1087                 if (bootlist[i] == '\n') {
1088                     bootlist[i] = ' ';
1089                 }
1090             }
1091             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1092         }
1093 
1094         if (boot_device && strlen(boot_device)) {
1095             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1096         }
1097 
1098         if (!spapr->has_graphics && stdout_path) {
1099             /*
1100              * "linux,stdout-path" and "stdout" properties are
1101              * deprecated by linux kernel. New platforms should only
1102              * use the "stdout-path" property. Set the new property
1103              * and continue using older property to remain compatible
1104              * with the existing firmware.
1105              */
1106             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1107             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1108         }
1109 
1110         /*
1111          * We can deal with BAR reallocation just fine, advertise it
1112          * to the guest
1113          */
1114         if (smc->linux_pci_probe) {
1115             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1116         }
1117 
1118         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1119 
1120         g_free(stdout_path);
1121         g_free(bootlist);
1122     }
1123 
1124     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1125 }
1126 
1127 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1128 {
1129     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1130      * KVM to work under pHyp with some guest co-operation */
1131     int hypervisor;
1132     uint8_t hypercall[16];
1133 
1134     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1135     /* indicate KVM hypercall interface */
1136     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1137     if (kvmppc_has_cap_fixup_hcalls()) {
1138         /*
1139          * Older KVM versions with older guest kernels were broken
1140          * with the magic page, don't allow the guest to map it.
1141          */
1142         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1143                                   sizeof(hypercall))) {
1144             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1145                              hypercall, sizeof(hypercall)));
1146         }
1147     }
1148 }
1149 
1150 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1151 {
1152     MachineState *machine = MACHINE(spapr);
1153     MachineClass *mc = MACHINE_GET_CLASS(machine);
1154     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1155     int ret;
1156     void *fdt;
1157     SpaprPhbState *phb;
1158     char *buf;
1159 
1160     fdt = g_malloc0(space);
1161     _FDT((fdt_create_empty_tree(fdt, space)));
1162 
1163     /* Root node */
1164     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1165     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1166     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1167 
1168     /* Guest UUID & Name*/
1169     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1170     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1171     if (qemu_uuid_set) {
1172         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1173     }
1174     g_free(buf);
1175 
1176     if (qemu_get_vm_name()) {
1177         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1178                                 qemu_get_vm_name()));
1179     }
1180 
1181     /* Host Model & Serial Number */
1182     if (spapr->host_model) {
1183         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1184     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1185         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1186         g_free(buf);
1187     }
1188 
1189     if (spapr->host_serial) {
1190         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1191     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1192         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1193         g_free(buf);
1194     }
1195 
1196     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1197     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1198 
1199     /* /interrupt controller */
1200     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1201 
1202     ret = spapr_dt_memory(spapr, fdt);
1203     if (ret < 0) {
1204         error_report("couldn't setup memory nodes in fdt");
1205         exit(1);
1206     }
1207 
1208     /* /vdevice */
1209     spapr_dt_vdevice(spapr->vio_bus, fdt);
1210 
1211     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1212         ret = spapr_dt_rng(fdt);
1213         if (ret < 0) {
1214             error_report("could not set up rng device in the fdt");
1215             exit(1);
1216         }
1217     }
1218 
1219     QLIST_FOREACH(phb, &spapr->phbs, list) {
1220         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1221         if (ret < 0) {
1222             error_report("couldn't setup PCI devices in fdt");
1223             exit(1);
1224         }
1225     }
1226 
1227     spapr_dt_cpus(fdt, spapr);
1228 
1229     if (smc->dr_lmb_enabled) {
1230         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1231     }
1232 
1233     if (mc->has_hotpluggable_cpus) {
1234         int offset = fdt_path_offset(fdt, "/cpus");
1235         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1236         if (ret < 0) {
1237             error_report("Couldn't set up CPU DR device tree properties");
1238             exit(1);
1239         }
1240     }
1241 
1242     /* /event-sources */
1243     spapr_dt_events(spapr, fdt);
1244 
1245     /* /rtas */
1246     spapr_dt_rtas(spapr, fdt);
1247 
1248     /* /chosen */
1249     spapr_dt_chosen(spapr, fdt, reset);
1250 
1251     /* /hypervisor */
1252     if (kvm_enabled()) {
1253         spapr_dt_hypervisor(spapr, fdt);
1254     }
1255 
1256     /* Build memory reserve map */
1257     if (reset) {
1258         if (spapr->kernel_size) {
1259             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1260                                   spapr->kernel_size)));
1261         }
1262         if (spapr->initrd_size) {
1263             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1264                                   spapr->initrd_size)));
1265         }
1266     }
1267 
1268     if (smc->dr_phb_enabled) {
1269         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1270         if (ret < 0) {
1271             error_report("Couldn't set up PHB DR device tree properties");
1272             exit(1);
1273         }
1274     }
1275 
1276     /* NVDIMM devices */
1277     if (mc->nvdimm_supported) {
1278         spapr_dt_persistent_memory(fdt);
1279     }
1280 
1281     return fdt;
1282 }
1283 
1284 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1285 {
1286     SpaprMachineState *spapr = opaque;
1287 
1288     return (addr & 0x0fffffff) + spapr->kernel_addr;
1289 }
1290 
1291 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1292                                     PowerPCCPU *cpu)
1293 {
1294     CPUPPCState *env = &cpu->env;
1295 
1296     /* The TCG path should also be holding the BQL at this point */
1297     g_assert(qemu_mutex_iothread_locked());
1298 
1299     if (msr_pr) {
1300         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1301         env->gpr[3] = H_PRIVILEGE;
1302     } else {
1303         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1304     }
1305 }
1306 
1307 struct LPCRSyncState {
1308     target_ulong value;
1309     target_ulong mask;
1310 };
1311 
1312 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1313 {
1314     struct LPCRSyncState *s = arg.host_ptr;
1315     PowerPCCPU *cpu = POWERPC_CPU(cs);
1316     CPUPPCState *env = &cpu->env;
1317     target_ulong lpcr;
1318 
1319     cpu_synchronize_state(cs);
1320     lpcr = env->spr[SPR_LPCR];
1321     lpcr &= ~s->mask;
1322     lpcr |= s->value;
1323     ppc_store_lpcr(cpu, lpcr);
1324 }
1325 
1326 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1327 {
1328     CPUState *cs;
1329     struct LPCRSyncState s = {
1330         .value = value,
1331         .mask = mask
1332     };
1333     CPU_FOREACH(cs) {
1334         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1335     }
1336 }
1337 
1338 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1339 {
1340     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1341 
1342     /* Copy PATE1:GR into PATE0:HR */
1343     entry->dw0 = spapr->patb_entry & PATE0_HR;
1344     entry->dw1 = spapr->patb_entry;
1345 }
1346 
1347 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1348 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1349 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1350 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1351 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1352 
1353 /*
1354  * Get the fd to access the kernel htab, re-opening it if necessary
1355  */
1356 static int get_htab_fd(SpaprMachineState *spapr)
1357 {
1358     Error *local_err = NULL;
1359 
1360     if (spapr->htab_fd >= 0) {
1361         return spapr->htab_fd;
1362     }
1363 
1364     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1365     if (spapr->htab_fd < 0) {
1366         error_report_err(local_err);
1367     }
1368 
1369     return spapr->htab_fd;
1370 }
1371 
1372 void close_htab_fd(SpaprMachineState *spapr)
1373 {
1374     if (spapr->htab_fd >= 0) {
1375         close(spapr->htab_fd);
1376     }
1377     spapr->htab_fd = -1;
1378 }
1379 
1380 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1381 {
1382     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1383 
1384     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1385 }
1386 
1387 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390 
1391     assert(kvm_enabled());
1392 
1393     if (!spapr->htab) {
1394         return 0;
1395     }
1396 
1397     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1398 }
1399 
1400 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1401                                                 hwaddr ptex, int n)
1402 {
1403     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1404     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1405 
1406     if (!spapr->htab) {
1407         /*
1408          * HTAB is controlled by KVM. Fetch into temporary buffer
1409          */
1410         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1411         kvmppc_read_hptes(hptes, ptex, n);
1412         return hptes;
1413     }
1414 
1415     /*
1416      * HTAB is controlled by QEMU. Just point to the internally
1417      * accessible PTEG.
1418      */
1419     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1420 }
1421 
1422 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1423                               const ppc_hash_pte64_t *hptes,
1424                               hwaddr ptex, int n)
1425 {
1426     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1427 
1428     if (!spapr->htab) {
1429         g_free((void *)hptes);
1430     }
1431 
1432     /* Nothing to do for qemu managed HPT */
1433 }
1434 
1435 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1436                       uint64_t pte0, uint64_t pte1)
1437 {
1438     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1439     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1440 
1441     if (!spapr->htab) {
1442         kvmppc_write_hpte(ptex, pte0, pte1);
1443     } else {
1444         if (pte0 & HPTE64_V_VALID) {
1445             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1446             /*
1447              * When setting valid, we write PTE1 first. This ensures
1448              * proper synchronization with the reading code in
1449              * ppc_hash64_pteg_search()
1450              */
1451             smp_wmb();
1452             stq_p(spapr->htab + offset, pte0);
1453         } else {
1454             stq_p(spapr->htab + offset, pte0);
1455             /*
1456              * When clearing it we set PTE0 first. This ensures proper
1457              * synchronization with the reading code in
1458              * ppc_hash64_pteg_search()
1459              */
1460             smp_wmb();
1461             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1462         }
1463     }
1464 }
1465 
1466 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1467                              uint64_t pte1)
1468 {
1469     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1470     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1471 
1472     if (!spapr->htab) {
1473         /* There should always be a hash table when this is called */
1474         error_report("spapr_hpte_set_c called with no hash table !");
1475         return;
1476     }
1477 
1478     /* The HW performs a non-atomic byte update */
1479     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1480 }
1481 
1482 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1483                              uint64_t pte1)
1484 {
1485     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1486     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1487 
1488     if (!spapr->htab) {
1489         /* There should always be a hash table when this is called */
1490         error_report("spapr_hpte_set_r called with no hash table !");
1491         return;
1492     }
1493 
1494     /* The HW performs a non-atomic byte update */
1495     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1496 }
1497 
1498 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1499 {
1500     int shift;
1501 
1502     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1503      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1504      * that's much more than is needed for Linux guests */
1505     shift = ctz64(pow2ceil(ramsize)) - 7;
1506     shift = MAX(shift, 18); /* Minimum architected size */
1507     shift = MIN(shift, 46); /* Maximum architected size */
1508     return shift;
1509 }
1510 
1511 void spapr_free_hpt(SpaprMachineState *spapr)
1512 {
1513     g_free(spapr->htab);
1514     spapr->htab = NULL;
1515     spapr->htab_shift = 0;
1516     close_htab_fd(spapr);
1517 }
1518 
1519 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1520                           Error **errp)
1521 {
1522     long rc;
1523 
1524     /* Clean up any HPT info from a previous boot */
1525     spapr_free_hpt(spapr);
1526 
1527     rc = kvmppc_reset_htab(shift);
1528     if (rc < 0) {
1529         /* kernel-side HPT needed, but couldn't allocate one */
1530         error_setg_errno(errp, errno,
1531                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1532                          shift);
1533         /* This is almost certainly fatal, but if the caller really
1534          * wants to carry on with shift == 0, it's welcome to try */
1535     } else if (rc > 0) {
1536         /* kernel-side HPT allocated */
1537         if (rc != shift) {
1538             error_setg(errp,
1539                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1540                        shift, rc);
1541         }
1542 
1543         spapr->htab_shift = shift;
1544         spapr->htab = NULL;
1545     } else {
1546         /* kernel-side HPT not needed, allocate in userspace instead */
1547         size_t size = 1ULL << shift;
1548         int i;
1549 
1550         spapr->htab = qemu_memalign(size, size);
1551         if (!spapr->htab) {
1552             error_setg_errno(errp, errno,
1553                              "Could not allocate HPT of order %d", shift);
1554             return;
1555         }
1556 
1557         memset(spapr->htab, 0, size);
1558         spapr->htab_shift = shift;
1559 
1560         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1561             DIRTY_HPTE(HPTE(spapr->htab, i));
1562         }
1563     }
1564     /* We're setting up a hash table, so that means we're not radix */
1565     spapr->patb_entry = 0;
1566     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1567 }
1568 
1569 void spapr_setup_hpt(SpaprMachineState *spapr)
1570 {
1571     int hpt_shift;
1572 
1573     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1574         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1575     } else {
1576         uint64_t current_ram_size;
1577 
1578         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1579         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1580     }
1581     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1582 
1583     if (kvm_enabled()) {
1584         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1585 
1586         /* Check our RMA fits in the possible VRMA */
1587         if (vrma_limit < spapr->rma_size) {
1588             error_report("Unable to create %" HWADDR_PRIu
1589                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1590                          spapr->rma_size / MiB, vrma_limit / MiB);
1591             exit(EXIT_FAILURE);
1592         }
1593     }
1594 }
1595 
1596 static int spapr_reset_drcs(Object *child, void *opaque)
1597 {
1598     SpaprDrc *drc =
1599         (SpaprDrc *) object_dynamic_cast(child,
1600                                                  TYPE_SPAPR_DR_CONNECTOR);
1601 
1602     if (drc) {
1603         spapr_drc_reset(drc);
1604     }
1605 
1606     return 0;
1607 }
1608 
1609 static void spapr_machine_reset(MachineState *machine)
1610 {
1611     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1612     PowerPCCPU *first_ppc_cpu;
1613     hwaddr fdt_addr;
1614     void *fdt;
1615     int rc;
1616 
1617     kvmppc_svm_off(&error_fatal);
1618     spapr_caps_apply(spapr);
1619 
1620     first_ppc_cpu = POWERPC_CPU(first_cpu);
1621     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1622         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1623                               spapr->max_compat_pvr)) {
1624         /*
1625          * If using KVM with radix mode available, VCPUs can be started
1626          * without a HPT because KVM will start them in radix mode.
1627          * Set the GR bit in PATE so that we know there is no HPT.
1628          */
1629         spapr->patb_entry = PATE1_GR;
1630         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1631     } else {
1632         spapr_setup_hpt(spapr);
1633     }
1634 
1635     qemu_devices_reset();
1636 
1637     spapr_ovec_cleanup(spapr->ov5_cas);
1638     spapr->ov5_cas = spapr_ovec_new();
1639 
1640     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1641 
1642     /*
1643      * This is fixing some of the default configuration of the XIVE
1644      * devices. To be called after the reset of the machine devices.
1645      */
1646     spapr_irq_reset(spapr, &error_fatal);
1647 
1648     /*
1649      * There is no CAS under qtest. Simulate one to please the code that
1650      * depends on spapr->ov5_cas. This is especially needed to test device
1651      * unplug, so we do that before resetting the DRCs.
1652      */
1653     if (qtest_enabled()) {
1654         spapr_ovec_cleanup(spapr->ov5_cas);
1655         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1656     }
1657 
1658     /* DRC reset may cause a device to be unplugged. This will cause troubles
1659      * if this device is used by another device (eg, a running vhost backend
1660      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1661      * situations, we reset DRCs after all devices have been reset.
1662      */
1663     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1664 
1665     spapr_clear_pending_events(spapr);
1666 
1667     /*
1668      * We place the device tree and RTAS just below either the top of the RMA,
1669      * or just below 2GB, whichever is lower, so that it can be
1670      * processed with 32-bit real mode code if necessary
1671      */
1672     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1673 
1674     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1675 
1676     rc = fdt_pack(fdt);
1677 
1678     /* Should only fail if we've built a corrupted tree */
1679     assert(rc == 0);
1680 
1681     /* Load the fdt */
1682     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1683     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1684     g_free(spapr->fdt_blob);
1685     spapr->fdt_size = fdt_totalsize(fdt);
1686     spapr->fdt_initial_size = spapr->fdt_size;
1687     spapr->fdt_blob = fdt;
1688 
1689     /* Set up the entry state */
1690     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1691     first_ppc_cpu->env.gpr[5] = 0;
1692 
1693     spapr->fwnmi_system_reset_addr = -1;
1694     spapr->fwnmi_machine_check_addr = -1;
1695     spapr->fwnmi_machine_check_interlock = -1;
1696 
1697     /* Signal all vCPUs waiting on this condition */
1698     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1699 
1700     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1701 }
1702 
1703 static void spapr_create_nvram(SpaprMachineState *spapr)
1704 {
1705     DeviceState *dev = qdev_new("spapr-nvram");
1706     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1707 
1708     if (dinfo) {
1709         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1710                                 &error_fatal);
1711     }
1712 
1713     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1714 
1715     spapr->nvram = (struct SpaprNvram *)dev;
1716 }
1717 
1718 static void spapr_rtc_create(SpaprMachineState *spapr)
1719 {
1720     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1721                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1722                                        &error_fatal, NULL);
1723     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1724     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1725                               "date");
1726 }
1727 
1728 /* Returns whether we want to use VGA or not */
1729 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1730 {
1731     switch (vga_interface_type) {
1732     case VGA_NONE:
1733         return false;
1734     case VGA_DEVICE:
1735         return true;
1736     case VGA_STD:
1737     case VGA_VIRTIO:
1738     case VGA_CIRRUS:
1739         return pci_vga_init(pci_bus) != NULL;
1740     default:
1741         error_setg(errp,
1742                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1743         return false;
1744     }
1745 }
1746 
1747 static int spapr_pre_load(void *opaque)
1748 {
1749     int rc;
1750 
1751     rc = spapr_caps_pre_load(opaque);
1752     if (rc) {
1753         return rc;
1754     }
1755 
1756     return 0;
1757 }
1758 
1759 static int spapr_post_load(void *opaque, int version_id)
1760 {
1761     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1762     int err = 0;
1763 
1764     err = spapr_caps_post_migration(spapr);
1765     if (err) {
1766         return err;
1767     }
1768 
1769     /*
1770      * In earlier versions, there was no separate qdev for the PAPR
1771      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1772      * So when migrating from those versions, poke the incoming offset
1773      * value into the RTC device
1774      */
1775     if (version_id < 3) {
1776         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1777         if (err) {
1778             return err;
1779         }
1780     }
1781 
1782     if (kvm_enabled() && spapr->patb_entry) {
1783         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1784         bool radix = !!(spapr->patb_entry & PATE1_GR);
1785         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1786 
1787         /*
1788          * Update LPCR:HR and UPRT as they may not be set properly in
1789          * the stream
1790          */
1791         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1792                             LPCR_HR | LPCR_UPRT);
1793 
1794         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1795         if (err) {
1796             error_report("Process table config unsupported by the host");
1797             return -EINVAL;
1798         }
1799     }
1800 
1801     err = spapr_irq_post_load(spapr, version_id);
1802     if (err) {
1803         return err;
1804     }
1805 
1806     return err;
1807 }
1808 
1809 static int spapr_pre_save(void *opaque)
1810 {
1811     int rc;
1812 
1813     rc = spapr_caps_pre_save(opaque);
1814     if (rc) {
1815         return rc;
1816     }
1817 
1818     return 0;
1819 }
1820 
1821 static bool version_before_3(void *opaque, int version_id)
1822 {
1823     return version_id < 3;
1824 }
1825 
1826 static bool spapr_pending_events_needed(void *opaque)
1827 {
1828     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1829     return !QTAILQ_EMPTY(&spapr->pending_events);
1830 }
1831 
1832 static const VMStateDescription vmstate_spapr_event_entry = {
1833     .name = "spapr_event_log_entry",
1834     .version_id = 1,
1835     .minimum_version_id = 1,
1836     .fields = (VMStateField[]) {
1837         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1838         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1839         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1840                                      NULL, extended_length),
1841         VMSTATE_END_OF_LIST()
1842     },
1843 };
1844 
1845 static const VMStateDescription vmstate_spapr_pending_events = {
1846     .name = "spapr_pending_events",
1847     .version_id = 1,
1848     .minimum_version_id = 1,
1849     .needed = spapr_pending_events_needed,
1850     .fields = (VMStateField[]) {
1851         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1852                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1853         VMSTATE_END_OF_LIST()
1854     },
1855 };
1856 
1857 static bool spapr_ov5_cas_needed(void *opaque)
1858 {
1859     SpaprMachineState *spapr = opaque;
1860     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1861     bool cas_needed;
1862 
1863     /* Prior to the introduction of SpaprOptionVector, we had two option
1864      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1865      * Both of these options encode machine topology into the device-tree
1866      * in such a way that the now-booted OS should still be able to interact
1867      * appropriately with QEMU regardless of what options were actually
1868      * negotiatied on the source side.
1869      *
1870      * As such, we can avoid migrating the CAS-negotiated options if these
1871      * are the only options available on the current machine/platform.
1872      * Since these are the only options available for pseries-2.7 and
1873      * earlier, this allows us to maintain old->new/new->old migration
1874      * compatibility.
1875      *
1876      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1877      * via default pseries-2.8 machines and explicit command-line parameters.
1878      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1879      * of the actual CAS-negotiated values to continue working properly. For
1880      * example, availability of memory unplug depends on knowing whether
1881      * OV5_HP_EVT was negotiated via CAS.
1882      *
1883      * Thus, for any cases where the set of available CAS-negotiatable
1884      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1885      * include the CAS-negotiated options in the migration stream, unless
1886      * if they affect boot time behaviour only.
1887      */
1888     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1889     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1890     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1891 
1892     /* We need extra information if we have any bits outside the mask
1893      * defined above */
1894     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1895 
1896     spapr_ovec_cleanup(ov5_mask);
1897 
1898     return cas_needed;
1899 }
1900 
1901 static const VMStateDescription vmstate_spapr_ov5_cas = {
1902     .name = "spapr_option_vector_ov5_cas",
1903     .version_id = 1,
1904     .minimum_version_id = 1,
1905     .needed = spapr_ov5_cas_needed,
1906     .fields = (VMStateField[]) {
1907         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1908                                  vmstate_spapr_ovec, SpaprOptionVector),
1909         VMSTATE_END_OF_LIST()
1910     },
1911 };
1912 
1913 static bool spapr_patb_entry_needed(void *opaque)
1914 {
1915     SpaprMachineState *spapr = opaque;
1916 
1917     return !!spapr->patb_entry;
1918 }
1919 
1920 static const VMStateDescription vmstate_spapr_patb_entry = {
1921     .name = "spapr_patb_entry",
1922     .version_id = 1,
1923     .minimum_version_id = 1,
1924     .needed = spapr_patb_entry_needed,
1925     .fields = (VMStateField[]) {
1926         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1927         VMSTATE_END_OF_LIST()
1928     },
1929 };
1930 
1931 static bool spapr_irq_map_needed(void *opaque)
1932 {
1933     SpaprMachineState *spapr = opaque;
1934 
1935     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1936 }
1937 
1938 static const VMStateDescription vmstate_spapr_irq_map = {
1939     .name = "spapr_irq_map",
1940     .version_id = 1,
1941     .minimum_version_id = 1,
1942     .needed = spapr_irq_map_needed,
1943     .fields = (VMStateField[]) {
1944         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1945         VMSTATE_END_OF_LIST()
1946     },
1947 };
1948 
1949 static bool spapr_dtb_needed(void *opaque)
1950 {
1951     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1952 
1953     return smc->update_dt_enabled;
1954 }
1955 
1956 static int spapr_dtb_pre_load(void *opaque)
1957 {
1958     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1959 
1960     g_free(spapr->fdt_blob);
1961     spapr->fdt_blob = NULL;
1962     spapr->fdt_size = 0;
1963 
1964     return 0;
1965 }
1966 
1967 static const VMStateDescription vmstate_spapr_dtb = {
1968     .name = "spapr_dtb",
1969     .version_id = 1,
1970     .minimum_version_id = 1,
1971     .needed = spapr_dtb_needed,
1972     .pre_load = spapr_dtb_pre_load,
1973     .fields = (VMStateField[]) {
1974         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1975         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1976         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1977                                      fdt_size),
1978         VMSTATE_END_OF_LIST()
1979     },
1980 };
1981 
1982 static bool spapr_fwnmi_needed(void *opaque)
1983 {
1984     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1985 
1986     return spapr->fwnmi_machine_check_addr != -1;
1987 }
1988 
1989 static int spapr_fwnmi_pre_save(void *opaque)
1990 {
1991     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1992 
1993     /*
1994      * Check if machine check handling is in progress and print a
1995      * warning message.
1996      */
1997     if (spapr->fwnmi_machine_check_interlock != -1) {
1998         warn_report("A machine check is being handled during migration. The"
1999                 "handler may run and log hardware error on the destination");
2000     }
2001 
2002     return 0;
2003 }
2004 
2005 static const VMStateDescription vmstate_spapr_fwnmi = {
2006     .name = "spapr_fwnmi",
2007     .version_id = 1,
2008     .minimum_version_id = 1,
2009     .needed = spapr_fwnmi_needed,
2010     .pre_save = spapr_fwnmi_pre_save,
2011     .fields = (VMStateField[]) {
2012         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2013         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2014         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2015         VMSTATE_END_OF_LIST()
2016     },
2017 };
2018 
2019 static const VMStateDescription vmstate_spapr = {
2020     .name = "spapr",
2021     .version_id = 3,
2022     .minimum_version_id = 1,
2023     .pre_load = spapr_pre_load,
2024     .post_load = spapr_post_load,
2025     .pre_save = spapr_pre_save,
2026     .fields = (VMStateField[]) {
2027         /* used to be @next_irq */
2028         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2029 
2030         /* RTC offset */
2031         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2032 
2033         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2034         VMSTATE_END_OF_LIST()
2035     },
2036     .subsections = (const VMStateDescription*[]) {
2037         &vmstate_spapr_ov5_cas,
2038         &vmstate_spapr_patb_entry,
2039         &vmstate_spapr_pending_events,
2040         &vmstate_spapr_cap_htm,
2041         &vmstate_spapr_cap_vsx,
2042         &vmstate_spapr_cap_dfp,
2043         &vmstate_spapr_cap_cfpc,
2044         &vmstate_spapr_cap_sbbc,
2045         &vmstate_spapr_cap_ibs,
2046         &vmstate_spapr_cap_hpt_maxpagesize,
2047         &vmstate_spapr_irq_map,
2048         &vmstate_spapr_cap_nested_kvm_hv,
2049         &vmstate_spapr_dtb,
2050         &vmstate_spapr_cap_large_decr,
2051         &vmstate_spapr_cap_ccf_assist,
2052         &vmstate_spapr_cap_fwnmi,
2053         &vmstate_spapr_fwnmi,
2054         NULL
2055     }
2056 };
2057 
2058 static int htab_save_setup(QEMUFile *f, void *opaque)
2059 {
2060     SpaprMachineState *spapr = opaque;
2061 
2062     /* "Iteration" header */
2063     if (!spapr->htab_shift) {
2064         qemu_put_be32(f, -1);
2065     } else {
2066         qemu_put_be32(f, spapr->htab_shift);
2067     }
2068 
2069     if (spapr->htab) {
2070         spapr->htab_save_index = 0;
2071         spapr->htab_first_pass = true;
2072     } else {
2073         if (spapr->htab_shift) {
2074             assert(kvm_enabled());
2075         }
2076     }
2077 
2078 
2079     return 0;
2080 }
2081 
2082 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2083                             int chunkstart, int n_valid, int n_invalid)
2084 {
2085     qemu_put_be32(f, chunkstart);
2086     qemu_put_be16(f, n_valid);
2087     qemu_put_be16(f, n_invalid);
2088     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2089                     HASH_PTE_SIZE_64 * n_valid);
2090 }
2091 
2092 static void htab_save_end_marker(QEMUFile *f)
2093 {
2094     qemu_put_be32(f, 0);
2095     qemu_put_be16(f, 0);
2096     qemu_put_be16(f, 0);
2097 }
2098 
2099 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2100                                  int64_t max_ns)
2101 {
2102     bool has_timeout = max_ns != -1;
2103     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2104     int index = spapr->htab_save_index;
2105     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2106 
2107     assert(spapr->htab_first_pass);
2108 
2109     do {
2110         int chunkstart;
2111 
2112         /* Consume invalid HPTEs */
2113         while ((index < htabslots)
2114                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2115             CLEAN_HPTE(HPTE(spapr->htab, index));
2116             index++;
2117         }
2118 
2119         /* Consume valid HPTEs */
2120         chunkstart = index;
2121         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2122                && HPTE_VALID(HPTE(spapr->htab, index))) {
2123             CLEAN_HPTE(HPTE(spapr->htab, index));
2124             index++;
2125         }
2126 
2127         if (index > chunkstart) {
2128             int n_valid = index - chunkstart;
2129 
2130             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2131 
2132             if (has_timeout &&
2133                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2134                 break;
2135             }
2136         }
2137     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2138 
2139     if (index >= htabslots) {
2140         assert(index == htabslots);
2141         index = 0;
2142         spapr->htab_first_pass = false;
2143     }
2144     spapr->htab_save_index = index;
2145 }
2146 
2147 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2148                                 int64_t max_ns)
2149 {
2150     bool final = max_ns < 0;
2151     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2152     int examined = 0, sent = 0;
2153     int index = spapr->htab_save_index;
2154     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2155 
2156     assert(!spapr->htab_first_pass);
2157 
2158     do {
2159         int chunkstart, invalidstart;
2160 
2161         /* Consume non-dirty HPTEs */
2162         while ((index < htabslots)
2163                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2164             index++;
2165             examined++;
2166         }
2167 
2168         chunkstart = index;
2169         /* Consume valid dirty HPTEs */
2170         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2171                && HPTE_DIRTY(HPTE(spapr->htab, index))
2172                && HPTE_VALID(HPTE(spapr->htab, index))) {
2173             CLEAN_HPTE(HPTE(spapr->htab, index));
2174             index++;
2175             examined++;
2176         }
2177 
2178         invalidstart = index;
2179         /* Consume invalid dirty HPTEs */
2180         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2181                && HPTE_DIRTY(HPTE(spapr->htab, index))
2182                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2183             CLEAN_HPTE(HPTE(spapr->htab, index));
2184             index++;
2185             examined++;
2186         }
2187 
2188         if (index > chunkstart) {
2189             int n_valid = invalidstart - chunkstart;
2190             int n_invalid = index - invalidstart;
2191 
2192             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2193             sent += index - chunkstart;
2194 
2195             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2196                 break;
2197             }
2198         }
2199 
2200         if (examined >= htabslots) {
2201             break;
2202         }
2203 
2204         if (index >= htabslots) {
2205             assert(index == htabslots);
2206             index = 0;
2207         }
2208     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2209 
2210     if (index >= htabslots) {
2211         assert(index == htabslots);
2212         index = 0;
2213     }
2214 
2215     spapr->htab_save_index = index;
2216 
2217     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2218 }
2219 
2220 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2221 #define MAX_KVM_BUF_SIZE    2048
2222 
2223 static int htab_save_iterate(QEMUFile *f, void *opaque)
2224 {
2225     SpaprMachineState *spapr = opaque;
2226     int fd;
2227     int rc = 0;
2228 
2229     /* Iteration header */
2230     if (!spapr->htab_shift) {
2231         qemu_put_be32(f, -1);
2232         return 1;
2233     } else {
2234         qemu_put_be32(f, 0);
2235     }
2236 
2237     if (!spapr->htab) {
2238         assert(kvm_enabled());
2239 
2240         fd = get_htab_fd(spapr);
2241         if (fd < 0) {
2242             return fd;
2243         }
2244 
2245         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2246         if (rc < 0) {
2247             return rc;
2248         }
2249     } else  if (spapr->htab_first_pass) {
2250         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2251     } else {
2252         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2253     }
2254 
2255     htab_save_end_marker(f);
2256 
2257     return rc;
2258 }
2259 
2260 static int htab_save_complete(QEMUFile *f, void *opaque)
2261 {
2262     SpaprMachineState *spapr = opaque;
2263     int fd;
2264 
2265     /* Iteration header */
2266     if (!spapr->htab_shift) {
2267         qemu_put_be32(f, -1);
2268         return 0;
2269     } else {
2270         qemu_put_be32(f, 0);
2271     }
2272 
2273     if (!spapr->htab) {
2274         int rc;
2275 
2276         assert(kvm_enabled());
2277 
2278         fd = get_htab_fd(spapr);
2279         if (fd < 0) {
2280             return fd;
2281         }
2282 
2283         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2284         if (rc < 0) {
2285             return rc;
2286         }
2287     } else {
2288         if (spapr->htab_first_pass) {
2289             htab_save_first_pass(f, spapr, -1);
2290         }
2291         htab_save_later_pass(f, spapr, -1);
2292     }
2293 
2294     /* End marker */
2295     htab_save_end_marker(f);
2296 
2297     return 0;
2298 }
2299 
2300 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2301 {
2302     SpaprMachineState *spapr = opaque;
2303     uint32_t section_hdr;
2304     int fd = -1;
2305     Error *local_err = NULL;
2306 
2307     if (version_id < 1 || version_id > 1) {
2308         error_report("htab_load() bad version");
2309         return -EINVAL;
2310     }
2311 
2312     section_hdr = qemu_get_be32(f);
2313 
2314     if (section_hdr == -1) {
2315         spapr_free_hpt(spapr);
2316         return 0;
2317     }
2318 
2319     if (section_hdr) {
2320         /* First section gives the htab size */
2321         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2322         if (local_err) {
2323             error_report_err(local_err);
2324             return -EINVAL;
2325         }
2326         return 0;
2327     }
2328 
2329     if (!spapr->htab) {
2330         assert(kvm_enabled());
2331 
2332         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2333         if (fd < 0) {
2334             error_report_err(local_err);
2335             return fd;
2336         }
2337     }
2338 
2339     while (true) {
2340         uint32_t index;
2341         uint16_t n_valid, n_invalid;
2342 
2343         index = qemu_get_be32(f);
2344         n_valid = qemu_get_be16(f);
2345         n_invalid = qemu_get_be16(f);
2346 
2347         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2348             /* End of Stream */
2349             break;
2350         }
2351 
2352         if ((index + n_valid + n_invalid) >
2353             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2354             /* Bad index in stream */
2355             error_report(
2356                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2357                 index, n_valid, n_invalid, spapr->htab_shift);
2358             return -EINVAL;
2359         }
2360 
2361         if (spapr->htab) {
2362             if (n_valid) {
2363                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2364                                 HASH_PTE_SIZE_64 * n_valid);
2365             }
2366             if (n_invalid) {
2367                 memset(HPTE(spapr->htab, index + n_valid), 0,
2368                        HASH_PTE_SIZE_64 * n_invalid);
2369             }
2370         } else {
2371             int rc;
2372 
2373             assert(fd >= 0);
2374 
2375             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2376             if (rc < 0) {
2377                 return rc;
2378             }
2379         }
2380     }
2381 
2382     if (!spapr->htab) {
2383         assert(fd >= 0);
2384         close(fd);
2385     }
2386 
2387     return 0;
2388 }
2389 
2390 static void htab_save_cleanup(void *opaque)
2391 {
2392     SpaprMachineState *spapr = opaque;
2393 
2394     close_htab_fd(spapr);
2395 }
2396 
2397 static SaveVMHandlers savevm_htab_handlers = {
2398     .save_setup = htab_save_setup,
2399     .save_live_iterate = htab_save_iterate,
2400     .save_live_complete_precopy = htab_save_complete,
2401     .save_cleanup = htab_save_cleanup,
2402     .load_state = htab_load,
2403 };
2404 
2405 static void spapr_boot_set(void *opaque, const char *boot_device,
2406                            Error **errp)
2407 {
2408     MachineState *machine = MACHINE(opaque);
2409     machine->boot_order = 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 /* pSeries LPAR / sPAPR hardware init */
2672 static void spapr_machine_init(MachineState *machine)
2673 {
2674     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2675     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2676     MachineClass *mc = MACHINE_GET_CLASS(machine);
2677     const char *kernel_filename = machine->kernel_filename;
2678     const char *initrd_filename = machine->initrd_filename;
2679     PCIHostState *phb;
2680     int i;
2681     MemoryRegion *sysmem = get_system_memory();
2682     long load_limit, fw_size;
2683     char *filename;
2684     Error *resize_hpt_err = NULL;
2685 
2686     msi_nonbroken = true;
2687 
2688     QLIST_INIT(&spapr->phbs);
2689     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2690 
2691     /* Determine capabilities to run with */
2692     spapr_caps_init(spapr);
2693 
2694     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2695     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2696         /*
2697          * If the user explicitly requested a mode we should either
2698          * supply it, or fail completely (which we do below).  But if
2699          * it's not set explicitly, we reset our mode to something
2700          * that works
2701          */
2702         if (resize_hpt_err) {
2703             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2704             error_free(resize_hpt_err);
2705             resize_hpt_err = NULL;
2706         } else {
2707             spapr->resize_hpt = smc->resize_hpt_default;
2708         }
2709     }
2710 
2711     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2712 
2713     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2714         /*
2715          * User requested HPT resize, but this host can't supply it.  Bail out
2716          */
2717         error_report_err(resize_hpt_err);
2718         exit(1);
2719     }
2720     error_free(resize_hpt_err);
2721 
2722     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2723 
2724     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2725     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2726 
2727     /*
2728      * VSMT must be set in order to be able to compute VCPU ids, ie to
2729      * call spapr_max_server_number() or spapr_vcpu_id().
2730      */
2731     spapr_set_vsmt_mode(spapr, &error_fatal);
2732 
2733     /* Set up Interrupt Controller before we create the VCPUs */
2734     spapr_irq_init(spapr, &error_fatal);
2735 
2736     /* Set up containers for ibm,client-architecture-support negotiated options
2737      */
2738     spapr->ov5 = spapr_ovec_new();
2739     spapr->ov5_cas = spapr_ovec_new();
2740 
2741     if (smc->dr_lmb_enabled) {
2742         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2743         spapr_validate_node_memory(machine, &error_fatal);
2744     }
2745 
2746     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2747 
2748     /* advertise support for dedicated HP event source to guests */
2749     if (spapr->use_hotplug_event_source) {
2750         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2751     }
2752 
2753     /* advertise support for HPT resizing */
2754     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2755         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2756     }
2757 
2758     /* advertise support for ibm,dyamic-memory-v2 */
2759     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2760 
2761     /* advertise XIVE on POWER9 machines */
2762     if (spapr->irq->xive) {
2763         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2764     }
2765 
2766     /* init CPUs */
2767     spapr_init_cpus(spapr);
2768 
2769     /*
2770      * check we don't have a memory-less/cpu-less NUMA node
2771      * Firmware relies on the existing memory/cpu topology to provide the
2772      * NUMA topology to the kernel.
2773      * And the linux kernel needs to know the NUMA topology at start
2774      * to be able to hotplug CPUs later.
2775      */
2776     if (machine->numa_state->num_nodes) {
2777         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2778             /* check for memory-less node */
2779             if (machine->numa_state->nodes[i].node_mem == 0) {
2780                 CPUState *cs;
2781                 int found = 0;
2782                 /* check for cpu-less node */
2783                 CPU_FOREACH(cs) {
2784                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2785                     if (cpu->node_id == i) {
2786                         found = 1;
2787                         break;
2788                     }
2789                 }
2790                 /* memory-less and cpu-less node */
2791                 if (!found) {
2792                     error_report(
2793                        "Memory-less/cpu-less nodes are not supported (node %d)",
2794                                  i);
2795                     exit(1);
2796                 }
2797             }
2798         }
2799 
2800     }
2801 
2802     /*
2803      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2804      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2805      * called from vPHB reset handler so we initialize the counter here.
2806      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2807      * must be equally distant from any other node.
2808      * The final value of spapr->gpu_numa_id is going to be written to
2809      * max-associativity-domains in spapr_build_fdt().
2810      */
2811     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2812 
2813     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2814         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2815                               spapr->max_compat_pvr)) {
2816         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2817         /* KVM and TCG always allow GTSE with radix... */
2818         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2819     }
2820     /* ... but not with hash (currently). */
2821 
2822     if (kvm_enabled()) {
2823         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2824         kvmppc_enable_logical_ci_hcalls();
2825         kvmppc_enable_set_mode_hcall();
2826 
2827         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2828         kvmppc_enable_clear_ref_mod_hcalls();
2829 
2830         /* Enable H_PAGE_INIT */
2831         kvmppc_enable_h_page_init();
2832     }
2833 
2834     /* map RAM */
2835     memory_region_add_subregion(sysmem, 0, machine->ram);
2836 
2837     /* always allocate the device memory information */
2838     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2839 
2840     /* initialize hotplug memory address space */
2841     if (machine->ram_size < machine->maxram_size) {
2842         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2843         /*
2844          * Limit the number of hotpluggable memory slots to half the number
2845          * slots that KVM supports, leaving the other half for PCI and other
2846          * devices. However ensure that number of slots doesn't drop below 32.
2847          */
2848         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2849                            SPAPR_MAX_RAM_SLOTS;
2850 
2851         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2852             max_memslots = SPAPR_MAX_RAM_SLOTS;
2853         }
2854         if (machine->ram_slots > max_memslots) {
2855             error_report("Specified number of memory slots %"
2856                          PRIu64" exceeds max supported %d",
2857                          machine->ram_slots, max_memslots);
2858             exit(1);
2859         }
2860 
2861         machine->device_memory->base = ROUND_UP(machine->ram_size,
2862                                                 SPAPR_DEVICE_MEM_ALIGN);
2863         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2864                            "device-memory", device_mem_size);
2865         memory_region_add_subregion(sysmem, machine->device_memory->base,
2866                                     &machine->device_memory->mr);
2867     }
2868 
2869     if (smc->dr_lmb_enabled) {
2870         spapr_create_lmb_dr_connectors(spapr);
2871     }
2872 
2873     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2874         /* Create the error string for live migration blocker */
2875         error_setg(&spapr->fwnmi_migration_blocker,
2876             "A machine check is being handled during migration. The handler"
2877             "may run and log hardware error on the destination");
2878     }
2879 
2880     if (mc->nvdimm_supported) {
2881         spapr_create_nvdimm_dr_connectors(spapr);
2882     }
2883 
2884     /* Set up RTAS event infrastructure */
2885     spapr_events_init(spapr);
2886 
2887     /* Set up the RTC RTAS interfaces */
2888     spapr_rtc_create(spapr);
2889 
2890     /* Set up VIO bus */
2891     spapr->vio_bus = spapr_vio_bus_init();
2892 
2893     for (i = 0; i < serial_max_hds(); i++) {
2894         if (serial_hd(i)) {
2895             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2896         }
2897     }
2898 
2899     /* We always have at least the nvram device on VIO */
2900     spapr_create_nvram(spapr);
2901 
2902     /*
2903      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2904      * connectors (described in root DT node's "ibm,drc-types" property)
2905      * are pre-initialized here. additional child connectors (such as
2906      * connectors for a PHBs PCI slots) are added as needed during their
2907      * parent's realization.
2908      */
2909     if (smc->dr_phb_enabled) {
2910         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2911             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2912         }
2913     }
2914 
2915     /* Set up PCI */
2916     spapr_pci_rtas_init();
2917 
2918     phb = spapr_create_default_phb();
2919 
2920     for (i = 0; i < nb_nics; i++) {
2921         NICInfo *nd = &nd_table[i];
2922 
2923         if (!nd->model) {
2924             nd->model = g_strdup("spapr-vlan");
2925         }
2926 
2927         if (g_str_equal(nd->model, "spapr-vlan") ||
2928             g_str_equal(nd->model, "ibmveth")) {
2929             spapr_vlan_create(spapr->vio_bus, nd);
2930         } else {
2931             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2932         }
2933     }
2934 
2935     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2936         spapr_vscsi_create(spapr->vio_bus);
2937     }
2938 
2939     /* Graphics */
2940     if (spapr_vga_init(phb->bus, &error_fatal)) {
2941         spapr->has_graphics = true;
2942         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2943     }
2944 
2945     if (machine->usb) {
2946         if (smc->use_ohci_by_default) {
2947             pci_create_simple(phb->bus, -1, "pci-ohci");
2948         } else {
2949             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2950         }
2951 
2952         if (spapr->has_graphics) {
2953             USBBus *usb_bus = usb_bus_find(-1);
2954 
2955             usb_create_simple(usb_bus, "usb-kbd");
2956             usb_create_simple(usb_bus, "usb-mouse");
2957         }
2958     }
2959 
2960     if (kernel_filename) {
2961         uint64_t lowaddr = 0;
2962 
2963         spapr->kernel_size = load_elf(kernel_filename, NULL,
2964                                       translate_kernel_address, spapr,
2965                                       NULL, &lowaddr, NULL, NULL, 1,
2966                                       PPC_ELF_MACHINE, 0, 0);
2967         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2968             spapr->kernel_size = load_elf(kernel_filename, NULL,
2969                                           translate_kernel_address, spapr, NULL,
2970                                           &lowaddr, NULL, NULL, 0,
2971                                           PPC_ELF_MACHINE,
2972                                           0, 0);
2973             spapr->kernel_le = spapr->kernel_size > 0;
2974         }
2975         if (spapr->kernel_size < 0) {
2976             error_report("error loading %s: %s", kernel_filename,
2977                          load_elf_strerror(spapr->kernel_size));
2978             exit(1);
2979         }
2980 
2981         /* load initrd */
2982         if (initrd_filename) {
2983             /* Try to locate the initrd in the gap between the kernel
2984              * and the firmware. Add a bit of space just in case
2985              */
2986             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2987                                   + 0x1ffff) & ~0xffff;
2988             spapr->initrd_size = load_image_targphys(initrd_filename,
2989                                                      spapr->initrd_base,
2990                                                      load_limit
2991                                                      - spapr->initrd_base);
2992             if (spapr->initrd_size < 0) {
2993                 error_report("could not load initial ram disk '%s'",
2994                              initrd_filename);
2995                 exit(1);
2996             }
2997         }
2998     }
2999 
3000     if (bios_name == NULL) {
3001         bios_name = FW_FILE_NAME;
3002     }
3003     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3004     if (!filename) {
3005         error_report("Could not find LPAR firmware '%s'", bios_name);
3006         exit(1);
3007     }
3008     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3009     if (fw_size <= 0) {
3010         error_report("Could not load LPAR firmware '%s'", filename);
3011         exit(1);
3012     }
3013     g_free(filename);
3014 
3015     /* FIXME: Should register things through the MachineState's qdev
3016      * interface, this is a legacy from the sPAPREnvironment structure
3017      * which predated MachineState but had a similar function */
3018     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3019     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3020                          &savevm_htab_handlers, spapr);
3021 
3022     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3023 
3024     qemu_register_boot_set(spapr_boot_set, spapr);
3025 
3026     /*
3027      * Nothing needs to be done to resume a suspended guest because
3028      * suspending does not change the machine state, so no need for
3029      * a ->wakeup method.
3030      */
3031     qemu_register_wakeup_support();
3032 
3033     if (kvm_enabled()) {
3034         /* to stop and start vmclock */
3035         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3036                                          &spapr->tb);
3037 
3038         kvmppc_spapr_enable_inkernel_multitce();
3039     }
3040 
3041     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3042 }
3043 
3044 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3045 {
3046     if (!vm_type) {
3047         return 0;
3048     }
3049 
3050     if (!strcmp(vm_type, "HV")) {
3051         return 1;
3052     }
3053 
3054     if (!strcmp(vm_type, "PR")) {
3055         return 2;
3056     }
3057 
3058     error_report("Unknown kvm-type specified '%s'", vm_type);
3059     exit(1);
3060 }
3061 
3062 /*
3063  * Implementation of an interface to adjust firmware path
3064  * for the bootindex property handling.
3065  */
3066 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3067                                    DeviceState *dev)
3068 {
3069 #define CAST(type, obj, name) \
3070     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3071     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3072     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3073     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3074 
3075     if (d) {
3076         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3077         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3078         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3079 
3080         if (spapr) {
3081             /*
3082              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3083              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3084              * 0x8000 | (target << 8) | (bus << 5) | lun
3085              * (see the "Logical unit addressing format" table in SAM5)
3086              */
3087             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3088             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3089                                    (uint64_t)id << 48);
3090         } else if (virtio) {
3091             /*
3092              * We use SRP luns of the form 01000000 | (target << 8) | lun
3093              * in the top 32 bits of the 64-bit LUN
3094              * Note: the quote above is from SLOF and it is wrong,
3095              * the actual binding is:
3096              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3097              */
3098             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3099             if (d->lun >= 256) {
3100                 /* Use the LUN "flat space addressing method" */
3101                 id |= 0x4000;
3102             }
3103             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3104                                    (uint64_t)id << 32);
3105         } else if (usb) {
3106             /*
3107              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3108              * in the top 32 bits of the 64-bit LUN
3109              */
3110             unsigned usb_port = atoi(usb->port->path);
3111             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3112             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3113                                    (uint64_t)id << 32);
3114         }
3115     }
3116 
3117     /*
3118      * SLOF probes the USB devices, and if it recognizes that the device is a
3119      * storage device, it changes its name to "storage" instead of "usb-host",
3120      * and additionally adds a child node for the SCSI LUN, so the correct
3121      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3122      */
3123     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3124         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3125         if (usb_host_dev_is_scsi_storage(usbdev)) {
3126             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3127         }
3128     }
3129 
3130     if (phb) {
3131         /* Replace "pci" with "pci@800000020000000" */
3132         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3133     }
3134 
3135     if (vsc) {
3136         /* Same logic as virtio above */
3137         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3138         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3139     }
3140 
3141     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3142         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3143         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3144         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3145     }
3146 
3147     return NULL;
3148 }
3149 
3150 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3151 {
3152     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3153 
3154     return g_strdup(spapr->kvm_type);
3155 }
3156 
3157 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     g_free(spapr->kvm_type);
3162     spapr->kvm_type = g_strdup(value);
3163 }
3164 
3165 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     return spapr->use_hotplug_event_source;
3170 }
3171 
3172 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3173                                             Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     spapr->use_hotplug_event_source = value;
3178 }
3179 
3180 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3181 {
3182     return true;
3183 }
3184 
3185 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     switch (spapr->resize_hpt) {
3190     case SPAPR_RESIZE_HPT_DEFAULT:
3191         return g_strdup("default");
3192     case SPAPR_RESIZE_HPT_DISABLED:
3193         return g_strdup("disabled");
3194     case SPAPR_RESIZE_HPT_ENABLED:
3195         return g_strdup("enabled");
3196     case SPAPR_RESIZE_HPT_REQUIRED:
3197         return g_strdup("required");
3198     }
3199     g_assert_not_reached();
3200 }
3201 
3202 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     if (strcmp(value, "default") == 0) {
3207         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3208     } else if (strcmp(value, "disabled") == 0) {
3209         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3210     } else if (strcmp(value, "enabled") == 0) {
3211         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3212     } else if (strcmp(value, "required") == 0) {
3213         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3214     } else {
3215         error_setg(errp, "Bad value for \"resize-hpt\" property");
3216     }
3217 }
3218 
3219 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3220 {
3221     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3222 
3223     if (spapr->irq == &spapr_irq_xics_legacy) {
3224         return g_strdup("legacy");
3225     } else if (spapr->irq == &spapr_irq_xics) {
3226         return g_strdup("xics");
3227     } else if (spapr->irq == &spapr_irq_xive) {
3228         return g_strdup("xive");
3229     } else if (spapr->irq == &spapr_irq_dual) {
3230         return g_strdup("dual");
3231     }
3232     g_assert_not_reached();
3233 }
3234 
3235 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3236 {
3237     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3238 
3239     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3240         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3241         return;
3242     }
3243 
3244     /* The legacy IRQ backend can not be set */
3245     if (strcmp(value, "xics") == 0) {
3246         spapr->irq = &spapr_irq_xics;
3247     } else if (strcmp(value, "xive") == 0) {
3248         spapr->irq = &spapr_irq_xive;
3249     } else if (strcmp(value, "dual") == 0) {
3250         spapr->irq = &spapr_irq_dual;
3251     } else {
3252         error_setg(errp, "Bad value for \"ic-mode\" property");
3253     }
3254 }
3255 
3256 static char *spapr_get_host_model(Object *obj, Error **errp)
3257 {
3258     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3259 
3260     return g_strdup(spapr->host_model);
3261 }
3262 
3263 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3264 {
3265     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3266 
3267     g_free(spapr->host_model);
3268     spapr->host_model = g_strdup(value);
3269 }
3270 
3271 static char *spapr_get_host_serial(Object *obj, Error **errp)
3272 {
3273     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3274 
3275     return g_strdup(spapr->host_serial);
3276 }
3277 
3278 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3279 {
3280     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3281 
3282     g_free(spapr->host_serial);
3283     spapr->host_serial = g_strdup(value);
3284 }
3285 
3286 static void spapr_instance_init(Object *obj)
3287 {
3288     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3289     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3290 
3291     spapr->htab_fd = -1;
3292     spapr->use_hotplug_event_source = true;
3293     object_property_add_str(obj, "kvm-type",
3294                             spapr_get_kvm_type, spapr_set_kvm_type);
3295     object_property_set_description(obj, "kvm-type",
3296                                     "Specifies the KVM virtualization mode (HV, PR)");
3297     object_property_add_bool(obj, "modern-hotplug-events",
3298                             spapr_get_modern_hotplug_events,
3299                             spapr_set_modern_hotplug_events);
3300     object_property_set_description(obj, "modern-hotplug-events",
3301                                     "Use dedicated hotplug event mechanism in"
3302                                     " place of standard EPOW events when possible"
3303                                     " (required for memory hot-unplug support)");
3304     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3305                             "Maximum permitted CPU compatibility mode");
3306 
3307     object_property_add_str(obj, "resize-hpt",
3308                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3309     object_property_set_description(obj, "resize-hpt",
3310                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3311     object_property_add_uint32_ptr(obj, "vsmt",
3312                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3313     object_property_set_description(obj, "vsmt",
3314                                     "Virtual SMT: KVM behaves as if this were"
3315                                     " the host's SMT mode");
3316 
3317     object_property_add_bool(obj, "vfio-no-msix-emulation",
3318                              spapr_get_msix_emulation, NULL);
3319 
3320     object_property_add_uint64_ptr(obj, "kernel-addr",
3321                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3322     object_property_set_description(obj, "kernel-addr",
3323                                     stringify(KERNEL_LOAD_ADDR)
3324                                     " for -kernel is the default");
3325     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3326     /* The machine class defines the default interrupt controller mode */
3327     spapr->irq = smc->irq;
3328     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3329                             spapr_set_ic_mode);
3330     object_property_set_description(obj, "ic-mode",
3331                  "Specifies the interrupt controller mode (xics, xive, dual)");
3332 
3333     object_property_add_str(obj, "host-model",
3334         spapr_get_host_model, spapr_set_host_model);
3335     object_property_set_description(obj, "host-model",
3336         "Host model to advertise in guest device tree");
3337     object_property_add_str(obj, "host-serial",
3338         spapr_get_host_serial, spapr_set_host_serial);
3339     object_property_set_description(obj, "host-serial",
3340         "Host serial number to advertise in guest device tree");
3341 }
3342 
3343 static void spapr_machine_finalizefn(Object *obj)
3344 {
3345     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3346 
3347     g_free(spapr->kvm_type);
3348 }
3349 
3350 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3351 {
3352     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3353     PowerPCCPU *cpu = POWERPC_CPU(cs);
3354     CPUPPCState *env = &cpu->env;
3355 
3356     cpu_synchronize_state(cs);
3357     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3358     if (spapr->fwnmi_system_reset_addr != -1) {
3359         uint64_t rtas_addr, addr;
3360 
3361         /* get rtas addr from fdt */
3362         rtas_addr = spapr_get_rtas_addr();
3363         if (!rtas_addr) {
3364             qemu_system_guest_panicked(NULL);
3365             return;
3366         }
3367 
3368         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3369         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3370         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3371         env->gpr[3] = addr;
3372     }
3373     ppc_cpu_do_system_reset(cs);
3374     if (spapr->fwnmi_system_reset_addr != -1) {
3375         env->nip = spapr->fwnmi_system_reset_addr;
3376     }
3377 }
3378 
3379 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3380 {
3381     CPUState *cs;
3382 
3383     CPU_FOREACH(cs) {
3384         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3385     }
3386 }
3387 
3388 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3389                           void *fdt, int *fdt_start_offset, Error **errp)
3390 {
3391     uint64_t addr;
3392     uint32_t node;
3393 
3394     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3395     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3396                                     &error_abort);
3397     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3398                                              SPAPR_MEMORY_BLOCK_SIZE);
3399     return 0;
3400 }
3401 
3402 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3403                            bool dedicated_hp_event_source, Error **errp)
3404 {
3405     SpaprDrc *drc;
3406     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3407     int i;
3408     uint64_t addr = addr_start;
3409     bool hotplugged = spapr_drc_hotplugged(dev);
3410     Error *local_err = NULL;
3411 
3412     for (i = 0; i < nr_lmbs; i++) {
3413         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3414                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3415         g_assert(drc);
3416 
3417         spapr_drc_attach(drc, dev, &local_err);
3418         if (local_err) {
3419             while (addr > addr_start) {
3420                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3421                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3422                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3423                 spapr_drc_detach(drc);
3424             }
3425             error_propagate(errp, local_err);
3426             return;
3427         }
3428         if (!hotplugged) {
3429             spapr_drc_reset(drc);
3430         }
3431         addr += SPAPR_MEMORY_BLOCK_SIZE;
3432     }
3433     /* send hotplug notification to the
3434      * guest only in case of hotplugged memory
3435      */
3436     if (hotplugged) {
3437         if (dedicated_hp_event_source) {
3438             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3439                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3440             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3441                                                    nr_lmbs,
3442                                                    spapr_drc_index(drc));
3443         } else {
3444             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3445                                            nr_lmbs);
3446         }
3447     }
3448 }
3449 
3450 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3451                               Error **errp)
3452 {
3453     Error *local_err = NULL;
3454     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3455     PCDIMMDevice *dimm = PC_DIMM(dev);
3456     uint64_t size, addr, slot;
3457     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3458 
3459     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3460 
3461     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3462     if (local_err) {
3463         goto out;
3464     }
3465 
3466     if (!is_nvdimm) {
3467         addr = object_property_get_uint(OBJECT(dimm),
3468                                         PC_DIMM_ADDR_PROP, &local_err);
3469         if (local_err) {
3470             goto out_unplug;
3471         }
3472         spapr_add_lmbs(dev, addr, size,
3473                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3474                        &local_err);
3475     } else {
3476         slot = object_property_get_uint(OBJECT(dimm),
3477                                         PC_DIMM_SLOT_PROP, &local_err);
3478         if (local_err) {
3479             goto out_unplug;
3480         }
3481         spapr_add_nvdimm(dev, slot, &local_err);
3482     }
3483 
3484     if (local_err) {
3485         goto out_unplug;
3486     }
3487 
3488     return;
3489 
3490 out_unplug:
3491     pc_dimm_unplug(dimm, MACHINE(ms));
3492 out:
3493     error_propagate(errp, local_err);
3494 }
3495 
3496 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3497                                   Error **errp)
3498 {
3499     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3500     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3501     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3502     PCDIMMDevice *dimm = PC_DIMM(dev);
3503     Error *local_err = NULL;
3504     uint64_t size;
3505     Object *memdev;
3506     hwaddr pagesize;
3507 
3508     if (!smc->dr_lmb_enabled) {
3509         error_setg(errp, "Memory hotplug not supported for this machine");
3510         return;
3511     }
3512 
3513     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3514     if (local_err) {
3515         error_propagate(errp, local_err);
3516         return;
3517     }
3518 
3519     if (is_nvdimm) {
3520         spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, &local_err);
3521         if (local_err) {
3522             error_propagate(errp, local_err);
3523             return;
3524         }
3525     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3526         error_setg(errp, "Hotplugged memory size must be a multiple of "
3527                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3528         return;
3529     }
3530 
3531     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3532                                       &error_abort);
3533     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3534     spapr_check_pagesize(spapr, pagesize, &local_err);
3535     if (local_err) {
3536         error_propagate(errp, local_err);
3537         return;
3538     }
3539 
3540     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3541 }
3542 
3543 struct SpaprDimmState {
3544     PCDIMMDevice *dimm;
3545     uint32_t nr_lmbs;
3546     QTAILQ_ENTRY(SpaprDimmState) next;
3547 };
3548 
3549 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3550                                                        PCDIMMDevice *dimm)
3551 {
3552     SpaprDimmState *dimm_state = NULL;
3553 
3554     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3555         if (dimm_state->dimm == dimm) {
3556             break;
3557         }
3558     }
3559     return dimm_state;
3560 }
3561 
3562 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3563                                                       uint32_t nr_lmbs,
3564                                                       PCDIMMDevice *dimm)
3565 {
3566     SpaprDimmState *ds = NULL;
3567 
3568     /*
3569      * If this request is for a DIMM whose removal had failed earlier
3570      * (due to guest's refusal to remove the LMBs), we would have this
3571      * dimm already in the pending_dimm_unplugs list. In that
3572      * case don't add again.
3573      */
3574     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3575     if (!ds) {
3576         ds = g_malloc0(sizeof(SpaprDimmState));
3577         ds->nr_lmbs = nr_lmbs;
3578         ds->dimm = dimm;
3579         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3580     }
3581     return ds;
3582 }
3583 
3584 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3585                                               SpaprDimmState *dimm_state)
3586 {
3587     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3588     g_free(dimm_state);
3589 }
3590 
3591 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3592                                                         PCDIMMDevice *dimm)
3593 {
3594     SpaprDrc *drc;
3595     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3596                                                   &error_abort);
3597     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3598     uint32_t avail_lmbs = 0;
3599     uint64_t addr_start, addr;
3600     int i;
3601 
3602     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3603                                          &error_abort);
3604 
3605     addr = addr_start;
3606     for (i = 0; i < nr_lmbs; i++) {
3607         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3608                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3609         g_assert(drc);
3610         if (drc->dev) {
3611             avail_lmbs++;
3612         }
3613         addr += SPAPR_MEMORY_BLOCK_SIZE;
3614     }
3615 
3616     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3617 }
3618 
3619 /* Callback to be called during DRC release. */
3620 void spapr_lmb_release(DeviceState *dev)
3621 {
3622     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3623     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3624     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3625 
3626     /* This information will get lost if a migration occurs
3627      * during the unplug process. In this case recover it. */
3628     if (ds == NULL) {
3629         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3630         g_assert(ds);
3631         /* The DRC being examined by the caller at least must be counted */
3632         g_assert(ds->nr_lmbs);
3633     }
3634 
3635     if (--ds->nr_lmbs) {
3636         return;
3637     }
3638 
3639     /*
3640      * Now that all the LMBs have been removed by the guest, call the
3641      * unplug handler chain. This can never fail.
3642      */
3643     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3644     object_unparent(OBJECT(dev));
3645 }
3646 
3647 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3648 {
3649     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3650     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3651 
3652     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3653     qdev_unrealize(dev);
3654     spapr_pending_dimm_unplugs_remove(spapr, ds);
3655 }
3656 
3657 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3658                                         DeviceState *dev, Error **errp)
3659 {
3660     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3661     Error *local_err = NULL;
3662     PCDIMMDevice *dimm = PC_DIMM(dev);
3663     uint32_t nr_lmbs;
3664     uint64_t size, addr_start, addr;
3665     int i;
3666     SpaprDrc *drc;
3667 
3668     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3669         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3670         return;
3671     }
3672 
3673     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3674     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3675 
3676     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3677                                          &local_err);
3678     if (local_err) {
3679         error_propagate(errp, local_err);
3680         return;
3681     }
3682 
3683     /*
3684      * An existing pending dimm state for this DIMM means that there is an
3685      * unplug operation in progress, waiting for the spapr_lmb_release
3686      * callback to complete the job (BQL can't cover that far). In this case,
3687      * bail out to avoid detaching DRCs that were already released.
3688      */
3689     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3690         error_setg(errp, "Memory unplug already in progress for device %s",
3691                    dev->id);
3692         return;
3693     }
3694 
3695     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3696 
3697     addr = addr_start;
3698     for (i = 0; i < nr_lmbs; i++) {
3699         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3700                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3701         g_assert(drc);
3702 
3703         spapr_drc_detach(drc);
3704         addr += SPAPR_MEMORY_BLOCK_SIZE;
3705     }
3706 
3707     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3708                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3709     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3710                                               nr_lmbs, spapr_drc_index(drc));
3711 }
3712 
3713 /* Callback to be called during DRC release. */
3714 void spapr_core_release(DeviceState *dev)
3715 {
3716     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3717 
3718     /* Call the unplug handler chain. This can never fail. */
3719     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3720     object_unparent(OBJECT(dev));
3721 }
3722 
3723 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3724 {
3725     MachineState *ms = MACHINE(hotplug_dev);
3726     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3727     CPUCore *cc = CPU_CORE(dev);
3728     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3729 
3730     if (smc->pre_2_10_has_unused_icps) {
3731         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3732         int i;
3733 
3734         for (i = 0; i < cc->nr_threads; i++) {
3735             CPUState *cs = CPU(sc->threads[i]);
3736 
3737             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3738         }
3739     }
3740 
3741     assert(core_slot);
3742     core_slot->cpu = NULL;
3743     qdev_unrealize(dev);
3744 }
3745 
3746 static
3747 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3748                                Error **errp)
3749 {
3750     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3751     int index;
3752     SpaprDrc *drc;
3753     CPUCore *cc = CPU_CORE(dev);
3754 
3755     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3756         error_setg(errp, "Unable to find CPU core with core-id: %d",
3757                    cc->core_id);
3758         return;
3759     }
3760     if (index == 0) {
3761         error_setg(errp, "Boot CPU core may not be unplugged");
3762         return;
3763     }
3764 
3765     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3766                           spapr_vcpu_id(spapr, cc->core_id));
3767     g_assert(drc);
3768 
3769     if (!spapr_drc_unplug_requested(drc)) {
3770         spapr_drc_detach(drc);
3771         spapr_hotplug_req_remove_by_index(drc);
3772     }
3773 }
3774 
3775 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3776                            void *fdt, int *fdt_start_offset, Error **errp)
3777 {
3778     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3779     CPUState *cs = CPU(core->threads[0]);
3780     PowerPCCPU *cpu = POWERPC_CPU(cs);
3781     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3782     int id = spapr_get_vcpu_id(cpu);
3783     char *nodename;
3784     int offset;
3785 
3786     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3787     offset = fdt_add_subnode(fdt, 0, nodename);
3788     g_free(nodename);
3789 
3790     spapr_dt_cpu(cs, fdt, offset, spapr);
3791 
3792     *fdt_start_offset = offset;
3793     return 0;
3794 }
3795 
3796 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3797                             Error **errp)
3798 {
3799     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3800     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3801     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3802     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3803     CPUCore *cc = CPU_CORE(dev);
3804     CPUState *cs;
3805     SpaprDrc *drc;
3806     Error *local_err = NULL;
3807     CPUArchId *core_slot;
3808     int index;
3809     bool hotplugged = spapr_drc_hotplugged(dev);
3810     int i;
3811 
3812     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3813     if (!core_slot) {
3814         error_setg(errp, "Unable to find CPU core with core-id: %d",
3815                    cc->core_id);
3816         return;
3817     }
3818     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3819                           spapr_vcpu_id(spapr, cc->core_id));
3820 
3821     g_assert(drc || !mc->has_hotpluggable_cpus);
3822 
3823     if (drc) {
3824         spapr_drc_attach(drc, dev, &local_err);
3825         if (local_err) {
3826             error_propagate(errp, local_err);
3827             return;
3828         }
3829 
3830         if (hotplugged) {
3831             /*
3832              * Send hotplug notification interrupt to the guest only
3833              * in case of hotplugged CPUs.
3834              */
3835             spapr_hotplug_req_add_by_index(drc);
3836         } else {
3837             spapr_drc_reset(drc);
3838         }
3839     }
3840 
3841     core_slot->cpu = OBJECT(dev);
3842 
3843     if (smc->pre_2_10_has_unused_icps) {
3844         for (i = 0; i < cc->nr_threads; i++) {
3845             cs = CPU(core->threads[i]);
3846             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3847         }
3848     }
3849 
3850     /*
3851      * Set compatibility mode to match the boot CPU, which was either set
3852      * by the machine reset code or by CAS.
3853      */
3854     if (hotplugged) {
3855         for (i = 0; i < cc->nr_threads; i++) {
3856             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3857                            &local_err);
3858             if (local_err) {
3859                 error_propagate(errp, local_err);
3860                 return;
3861             }
3862         }
3863     }
3864 }
3865 
3866 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3867                                 Error **errp)
3868 {
3869     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3870     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3871     CPUCore *cc = CPU_CORE(dev);
3872     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3873     const char *type = object_get_typename(OBJECT(dev));
3874     CPUArchId *core_slot;
3875     int index;
3876     unsigned int smp_threads = machine->smp.threads;
3877 
3878     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3879         error_setg(errp, "CPU hotplug not supported for this machine");
3880         return;
3881     }
3882 
3883     if (strcmp(base_core_type, type)) {
3884         error_setg(errp, "CPU core type should be %s", base_core_type);
3885         return;
3886     }
3887 
3888     if (cc->core_id % smp_threads) {
3889         error_setg(errp, "invalid core id %d", cc->core_id);
3890         return;
3891     }
3892 
3893     /*
3894      * In general we should have homogeneous threads-per-core, but old
3895      * (pre hotplug support) machine types allow the last core to have
3896      * reduced threads as a compatibility hack for when we allowed
3897      * total vcpus not a multiple of threads-per-core.
3898      */
3899     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3900         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3901                    smp_threads);
3902         return;
3903     }
3904 
3905     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3906     if (!core_slot) {
3907         error_setg(errp, "core id %d out of range", cc->core_id);
3908         return;
3909     }
3910 
3911     if (core_slot->cpu) {
3912         error_setg(errp, "core %d already populated", cc->core_id);
3913         return;
3914     }
3915 
3916     numa_cpu_pre_plug(core_slot, dev, errp);
3917 }
3918 
3919 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3920                           void *fdt, int *fdt_start_offset, Error **errp)
3921 {
3922     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3923     int intc_phandle;
3924 
3925     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3926     if (intc_phandle <= 0) {
3927         return -1;
3928     }
3929 
3930     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3931         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3932         return -1;
3933     }
3934 
3935     /* generally SLOF creates these, for hotplug it's up to QEMU */
3936     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3937 
3938     return 0;
3939 }
3940 
3941 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3942                                Error **errp)
3943 {
3944     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3945     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3946     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3947     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3948 
3949     if (dev->hotplugged && !smc->dr_phb_enabled) {
3950         error_setg(errp, "PHB hotplug not supported for this machine");
3951         return;
3952     }
3953 
3954     if (sphb->index == (uint32_t)-1) {
3955         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3956         return;
3957     }
3958 
3959     /*
3960      * This will check that sphb->index doesn't exceed the maximum number of
3961      * PHBs for the current machine type.
3962      */
3963     smc->phb_placement(spapr, sphb->index,
3964                        &sphb->buid, &sphb->io_win_addr,
3965                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3966                        windows_supported, sphb->dma_liobn,
3967                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3968                        errp);
3969 }
3970 
3971 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3972                            Error **errp)
3973 {
3974     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3975     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3976     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3977     SpaprDrc *drc;
3978     bool hotplugged = spapr_drc_hotplugged(dev);
3979     Error *local_err = NULL;
3980 
3981     if (!smc->dr_phb_enabled) {
3982         return;
3983     }
3984 
3985     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3986     /* hotplug hooks should check it's enabled before getting this far */
3987     assert(drc);
3988 
3989     spapr_drc_attach(drc, dev, &local_err);
3990     if (local_err) {
3991         error_propagate(errp, local_err);
3992         return;
3993     }
3994 
3995     if (hotplugged) {
3996         spapr_hotplug_req_add_by_index(drc);
3997     } else {
3998         spapr_drc_reset(drc);
3999     }
4000 }
4001 
4002 void spapr_phb_release(DeviceState *dev)
4003 {
4004     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4005 
4006     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4007     object_unparent(OBJECT(dev));
4008 }
4009 
4010 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4011 {
4012     qdev_unrealize(dev);
4013 }
4014 
4015 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4016                                      DeviceState *dev, Error **errp)
4017 {
4018     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4019     SpaprDrc *drc;
4020 
4021     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4022     assert(drc);
4023 
4024     if (!spapr_drc_unplug_requested(drc)) {
4025         spapr_drc_detach(drc);
4026         spapr_hotplug_req_remove_by_index(drc);
4027     }
4028 }
4029 
4030 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4031                                  Error **errp)
4032 {
4033     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4034     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4035 
4036     if (spapr->tpm_proxy != NULL) {
4037         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4038         return;
4039     }
4040 
4041     spapr->tpm_proxy = tpm_proxy;
4042 }
4043 
4044 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4045 {
4046     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4047 
4048     qdev_unrealize(dev);
4049     object_unparent(OBJECT(dev));
4050     spapr->tpm_proxy = NULL;
4051 }
4052 
4053 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4054                                       DeviceState *dev, Error **errp)
4055 {
4056     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4057         spapr_memory_plug(hotplug_dev, dev, errp);
4058     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4059         spapr_core_plug(hotplug_dev, dev, errp);
4060     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4061         spapr_phb_plug(hotplug_dev, dev, errp);
4062     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4063         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4064     }
4065 }
4066 
4067 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4068                                         DeviceState *dev, Error **errp)
4069 {
4070     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4071         spapr_memory_unplug(hotplug_dev, dev);
4072     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4073         spapr_core_unplug(hotplug_dev, dev);
4074     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4075         spapr_phb_unplug(hotplug_dev, dev);
4076     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4077         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4078     }
4079 }
4080 
4081 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4082                                                 DeviceState *dev, Error **errp)
4083 {
4084     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4085     MachineClass *mc = MACHINE_GET_CLASS(sms);
4086     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4087 
4088     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4089         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4090             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4091         } else {
4092             /* NOTE: this means there is a window after guest reset, prior to
4093              * CAS negotiation, where unplug requests will fail due to the
4094              * capability not being detected yet. This is a bit different than
4095              * the case with PCI unplug, where the events will be queued and
4096              * eventually handled by the guest after boot
4097              */
4098             error_setg(errp, "Memory hot unplug not supported for this guest");
4099         }
4100     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4101         if (!mc->has_hotpluggable_cpus) {
4102             error_setg(errp, "CPU hot unplug not supported on this machine");
4103             return;
4104         }
4105         spapr_core_unplug_request(hotplug_dev, dev, errp);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4107         if (!smc->dr_phb_enabled) {
4108             error_setg(errp, "PHB hot unplug not supported on this machine");
4109             return;
4110         }
4111         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4112     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4113         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4114     }
4115 }
4116 
4117 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4118                                           DeviceState *dev, Error **errp)
4119 {
4120     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4121         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4122     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4123         spapr_core_pre_plug(hotplug_dev, dev, errp);
4124     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4125         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4126     }
4127 }
4128 
4129 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4130                                                  DeviceState *dev)
4131 {
4132     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4133         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4134         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4135         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4136         return HOTPLUG_HANDLER(machine);
4137     }
4138     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4139         PCIDevice *pcidev = PCI_DEVICE(dev);
4140         PCIBus *root = pci_device_root_bus(pcidev);
4141         SpaprPhbState *phb =
4142             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4143                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4144 
4145         if (phb) {
4146             return HOTPLUG_HANDLER(phb);
4147         }
4148     }
4149     return NULL;
4150 }
4151 
4152 static CpuInstanceProperties
4153 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4154 {
4155     CPUArchId *core_slot;
4156     MachineClass *mc = MACHINE_GET_CLASS(machine);
4157 
4158     /* make sure possible_cpu are intialized */
4159     mc->possible_cpu_arch_ids(machine);
4160     /* get CPU core slot containing thread that matches cpu_index */
4161     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4162     assert(core_slot);
4163     return core_slot->props;
4164 }
4165 
4166 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4167 {
4168     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4169 }
4170 
4171 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4172 {
4173     int i;
4174     unsigned int smp_threads = machine->smp.threads;
4175     unsigned int smp_cpus = machine->smp.cpus;
4176     const char *core_type;
4177     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4178     MachineClass *mc = MACHINE_GET_CLASS(machine);
4179 
4180     if (!mc->has_hotpluggable_cpus) {
4181         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4182     }
4183     if (machine->possible_cpus) {
4184         assert(machine->possible_cpus->len == spapr_max_cores);
4185         return machine->possible_cpus;
4186     }
4187 
4188     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4189     if (!core_type) {
4190         error_report("Unable to find sPAPR CPU Core definition");
4191         exit(1);
4192     }
4193 
4194     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4195                              sizeof(CPUArchId) * spapr_max_cores);
4196     machine->possible_cpus->len = spapr_max_cores;
4197     for (i = 0; i < machine->possible_cpus->len; i++) {
4198         int core_id = i * smp_threads;
4199 
4200         machine->possible_cpus->cpus[i].type = core_type;
4201         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4202         machine->possible_cpus->cpus[i].arch_id = core_id;
4203         machine->possible_cpus->cpus[i].props.has_core_id = true;
4204         machine->possible_cpus->cpus[i].props.core_id = core_id;
4205     }
4206     return machine->possible_cpus;
4207 }
4208 
4209 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4210                                 uint64_t *buid, hwaddr *pio,
4211                                 hwaddr *mmio32, hwaddr *mmio64,
4212                                 unsigned n_dma, uint32_t *liobns,
4213                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4214 {
4215     /*
4216      * New-style PHB window placement.
4217      *
4218      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4219      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4220      * windows.
4221      *
4222      * Some guest kernels can't work with MMIO windows above 1<<46
4223      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4224      *
4225      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4226      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4227      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4228      * 1TiB 64-bit MMIO windows for each PHB.
4229      */
4230     const uint64_t base_buid = 0x800000020000000ULL;
4231     int i;
4232 
4233     /* Sanity check natural alignments */
4234     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4235     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4236     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4237     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4238     /* Sanity check bounds */
4239     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4240                       SPAPR_PCI_MEM32_WIN_SIZE);
4241     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4242                       SPAPR_PCI_MEM64_WIN_SIZE);
4243 
4244     if (index >= SPAPR_MAX_PHBS) {
4245         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4246                    SPAPR_MAX_PHBS - 1);
4247         return;
4248     }
4249 
4250     *buid = base_buid + index;
4251     for (i = 0; i < n_dma; ++i) {
4252         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4253     }
4254 
4255     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4256     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4257     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4258 
4259     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4260     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4261 }
4262 
4263 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4264 {
4265     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4266 
4267     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4268 }
4269 
4270 static void spapr_ics_resend(XICSFabric *dev)
4271 {
4272     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4273 
4274     ics_resend(spapr->ics);
4275 }
4276 
4277 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4278 {
4279     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4280 
4281     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4282 }
4283 
4284 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4285                                  Monitor *mon)
4286 {
4287     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4288 
4289     spapr_irq_print_info(spapr, mon);
4290     monitor_printf(mon, "irqchip: %s\n",
4291                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4292 }
4293 
4294 /*
4295  * This is a XIVE only operation
4296  */
4297 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4298                            uint8_t nvt_blk, uint32_t nvt_idx,
4299                            bool cam_ignore, uint8_t priority,
4300                            uint32_t logic_serv, XiveTCTXMatch *match)
4301 {
4302     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4303     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4304     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4305     int count;
4306 
4307     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4308                            priority, logic_serv, match);
4309     if (count < 0) {
4310         return count;
4311     }
4312 
4313     /*
4314      * When we implement the save and restore of the thread interrupt
4315      * contexts in the enter/exit CPU handlers of the machine and the
4316      * escalations in QEMU, we should be able to handle non dispatched
4317      * vCPUs.
4318      *
4319      * Until this is done, the sPAPR machine should find at least one
4320      * matching context always.
4321      */
4322     if (count == 0) {
4323         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4324                       nvt_blk, nvt_idx);
4325     }
4326 
4327     return count;
4328 }
4329 
4330 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4331 {
4332     return cpu->vcpu_id;
4333 }
4334 
4335 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4336 {
4337     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4338     MachineState *ms = MACHINE(spapr);
4339     int vcpu_id;
4340 
4341     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4342 
4343     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4344         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4345         error_append_hint(errp, "Adjust the number of cpus to %d "
4346                           "or try to raise the number of threads per core\n",
4347                           vcpu_id * ms->smp.threads / spapr->vsmt);
4348         return;
4349     }
4350 
4351     cpu->vcpu_id = vcpu_id;
4352 }
4353 
4354 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4355 {
4356     CPUState *cs;
4357 
4358     CPU_FOREACH(cs) {
4359         PowerPCCPU *cpu = POWERPC_CPU(cs);
4360 
4361         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4362             return cpu;
4363         }
4364     }
4365 
4366     return NULL;
4367 }
4368 
4369 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4370 {
4371     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4372 
4373     /* These are only called by TCG, KVM maintains dispatch state */
4374 
4375     spapr_cpu->prod = false;
4376     if (spapr_cpu->vpa_addr) {
4377         CPUState *cs = CPU(cpu);
4378         uint32_t dispatch;
4379 
4380         dispatch = ldl_be_phys(cs->as,
4381                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4382         dispatch++;
4383         if ((dispatch & 1) != 0) {
4384             qemu_log_mask(LOG_GUEST_ERROR,
4385                           "VPA: incorrect dispatch counter value for "
4386                           "dispatched partition %u, correcting.\n", dispatch);
4387             dispatch++;
4388         }
4389         stl_be_phys(cs->as,
4390                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4391     }
4392 }
4393 
4394 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4395 {
4396     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4397 
4398     if (spapr_cpu->vpa_addr) {
4399         CPUState *cs = CPU(cpu);
4400         uint32_t dispatch;
4401 
4402         dispatch = ldl_be_phys(cs->as,
4403                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4404         dispatch++;
4405         if ((dispatch & 1) != 1) {
4406             qemu_log_mask(LOG_GUEST_ERROR,
4407                           "VPA: incorrect dispatch counter value for "
4408                           "preempted partition %u, correcting.\n", dispatch);
4409             dispatch++;
4410         }
4411         stl_be_phys(cs->as,
4412                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4413     }
4414 }
4415 
4416 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4417 {
4418     MachineClass *mc = MACHINE_CLASS(oc);
4419     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4420     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4421     NMIClass *nc = NMI_CLASS(oc);
4422     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4423     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4424     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4425     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4426     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4427 
4428     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4429     mc->ignore_boot_device_suffixes = true;
4430 
4431     /*
4432      * We set up the default / latest behaviour here.  The class_init
4433      * functions for the specific versioned machine types can override
4434      * these details for backwards compatibility
4435      */
4436     mc->init = spapr_machine_init;
4437     mc->reset = spapr_machine_reset;
4438     mc->block_default_type = IF_SCSI;
4439     mc->max_cpus = 1024;
4440     mc->no_parallel = 1;
4441     mc->default_boot_order = "";
4442     mc->default_ram_size = 512 * MiB;
4443     mc->default_ram_id = "ppc_spapr.ram";
4444     mc->default_display = "std";
4445     mc->kvm_type = spapr_kvm_type;
4446     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4447     mc->pci_allow_0_address = true;
4448     assert(!mc->get_hotplug_handler);
4449     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4450     hc->pre_plug = spapr_machine_device_pre_plug;
4451     hc->plug = spapr_machine_device_plug;
4452     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4453     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4454     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4455     hc->unplug_request = spapr_machine_device_unplug_request;
4456     hc->unplug = spapr_machine_device_unplug;
4457 
4458     smc->dr_lmb_enabled = true;
4459     smc->update_dt_enabled = true;
4460     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4461     mc->has_hotpluggable_cpus = true;
4462     mc->nvdimm_supported = true;
4463     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4464     fwc->get_dev_path = spapr_get_fw_dev_path;
4465     nc->nmi_monitor_handler = spapr_nmi;
4466     smc->phb_placement = spapr_phb_placement;
4467     vhc->hypercall = emulate_spapr_hypercall;
4468     vhc->hpt_mask = spapr_hpt_mask;
4469     vhc->map_hptes = spapr_map_hptes;
4470     vhc->unmap_hptes = spapr_unmap_hptes;
4471     vhc->hpte_set_c = spapr_hpte_set_c;
4472     vhc->hpte_set_r = spapr_hpte_set_r;
4473     vhc->get_pate = spapr_get_pate;
4474     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4475     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4476     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4477     xic->ics_get = spapr_ics_get;
4478     xic->ics_resend = spapr_ics_resend;
4479     xic->icp_get = spapr_icp_get;
4480     ispc->print_info = spapr_pic_print_info;
4481     /* Force NUMA node memory size to be a multiple of
4482      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4483      * in which LMBs are represented and hot-added
4484      */
4485     mc->numa_mem_align_shift = 28;
4486     mc->auto_enable_numa = true;
4487 
4488     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4489     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4490     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4491     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4492     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4493     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4494     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4495     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4496     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4497     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4498     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4499     spapr_caps_add_properties(smc);
4500     smc->irq = &spapr_irq_dual;
4501     smc->dr_phb_enabled = true;
4502     smc->linux_pci_probe = true;
4503     smc->smp_threads_vsmt = true;
4504     smc->nr_xirqs = SPAPR_NR_XIRQS;
4505     xfc->match_nvt = spapr_match_nvt;
4506 }
4507 
4508 static const TypeInfo spapr_machine_info = {
4509     .name          = TYPE_SPAPR_MACHINE,
4510     .parent        = TYPE_MACHINE,
4511     .abstract      = true,
4512     .instance_size = sizeof(SpaprMachineState),
4513     .instance_init = spapr_instance_init,
4514     .instance_finalize = spapr_machine_finalizefn,
4515     .class_size    = sizeof(SpaprMachineClass),
4516     .class_init    = spapr_machine_class_init,
4517     .interfaces = (InterfaceInfo[]) {
4518         { TYPE_FW_PATH_PROVIDER },
4519         { TYPE_NMI },
4520         { TYPE_HOTPLUG_HANDLER },
4521         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4522         { TYPE_XICS_FABRIC },
4523         { TYPE_INTERRUPT_STATS_PROVIDER },
4524         { TYPE_XIVE_FABRIC },
4525         { }
4526     },
4527 };
4528 
4529 static void spapr_machine_latest_class_options(MachineClass *mc)
4530 {
4531     mc->alias = "pseries";
4532     mc->is_default = true;
4533 }
4534 
4535 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4536     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4537                                                     void *data)      \
4538     {                                                                \
4539         MachineClass *mc = MACHINE_CLASS(oc);                        \
4540         spapr_machine_##suffix##_class_options(mc);                  \
4541         if (latest) {                                                \
4542             spapr_machine_latest_class_options(mc);                  \
4543         }                                                            \
4544     }                                                                \
4545     static const TypeInfo spapr_machine_##suffix##_info = {          \
4546         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4547         .parent = TYPE_SPAPR_MACHINE,                                \
4548         .class_init = spapr_machine_##suffix##_class_init,           \
4549     };                                                               \
4550     static void spapr_machine_register_##suffix(void)                \
4551     {                                                                \
4552         type_register(&spapr_machine_##suffix##_info);               \
4553     }                                                                \
4554     type_init(spapr_machine_register_##suffix)
4555 
4556 /*
4557  * pseries-5.2
4558  */
4559 static void spapr_machine_5_2_class_options(MachineClass *mc)
4560 {
4561     /* Defaults for the latest behaviour inherited from the base class */
4562 }
4563 
4564 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4565 
4566 /*
4567  * pseries-5.1
4568  */
4569 static void spapr_machine_5_1_class_options(MachineClass *mc)
4570 {
4571     spapr_machine_5_2_class_options(mc);
4572     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4573 }
4574 
4575 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4576 
4577 /*
4578  * pseries-5.0
4579  */
4580 static void spapr_machine_5_0_class_options(MachineClass *mc)
4581 {
4582     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4583     static GlobalProperty compat[] = {
4584         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4585     };
4586 
4587     spapr_machine_5_1_class_options(mc);
4588     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4589     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4590     mc->numa_mem_supported = true;
4591     smc->pre_5_1_assoc_refpoints = true;
4592 }
4593 
4594 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4595 
4596 /*
4597  * pseries-4.2
4598  */
4599 static void spapr_machine_4_2_class_options(MachineClass *mc)
4600 {
4601     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4602 
4603     spapr_machine_5_0_class_options(mc);
4604     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4605     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4606     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4607     smc->rma_limit = 16 * GiB;
4608     mc->nvdimm_supported = false;
4609 }
4610 
4611 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4612 
4613 /*
4614  * pseries-4.1
4615  */
4616 static void spapr_machine_4_1_class_options(MachineClass *mc)
4617 {
4618     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4619     static GlobalProperty compat[] = {
4620         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4621         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4622     };
4623 
4624     spapr_machine_4_2_class_options(mc);
4625     smc->linux_pci_probe = false;
4626     smc->smp_threads_vsmt = false;
4627     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4628     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4629 }
4630 
4631 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4632 
4633 /*
4634  * pseries-4.0
4635  */
4636 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4637                               uint64_t *buid, hwaddr *pio,
4638                               hwaddr *mmio32, hwaddr *mmio64,
4639                               unsigned n_dma, uint32_t *liobns,
4640                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4641 {
4642     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4643                         nv2gpa, nv2atsd, errp);
4644     *nv2gpa = 0;
4645     *nv2atsd = 0;
4646 }
4647 
4648 static void spapr_machine_4_0_class_options(MachineClass *mc)
4649 {
4650     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4651 
4652     spapr_machine_4_1_class_options(mc);
4653     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4654     smc->phb_placement = phb_placement_4_0;
4655     smc->irq = &spapr_irq_xics;
4656     smc->pre_4_1_migration = true;
4657 }
4658 
4659 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4660 
4661 /*
4662  * pseries-3.1
4663  */
4664 static void spapr_machine_3_1_class_options(MachineClass *mc)
4665 {
4666     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4667 
4668     spapr_machine_4_0_class_options(mc);
4669     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4670 
4671     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4672     smc->update_dt_enabled = false;
4673     smc->dr_phb_enabled = false;
4674     smc->broken_host_serial_model = true;
4675     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4676     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4677     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4678     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4679 }
4680 
4681 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4682 
4683 /*
4684  * pseries-3.0
4685  */
4686 
4687 static void spapr_machine_3_0_class_options(MachineClass *mc)
4688 {
4689     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4690 
4691     spapr_machine_3_1_class_options(mc);
4692     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4693 
4694     smc->legacy_irq_allocation = true;
4695     smc->nr_xirqs = 0x400;
4696     smc->irq = &spapr_irq_xics_legacy;
4697 }
4698 
4699 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4700 
4701 /*
4702  * pseries-2.12
4703  */
4704 static void spapr_machine_2_12_class_options(MachineClass *mc)
4705 {
4706     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4707     static GlobalProperty compat[] = {
4708         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4709         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4710     };
4711 
4712     spapr_machine_3_0_class_options(mc);
4713     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4714     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4715 
4716     /* We depend on kvm_enabled() to choose a default value for the
4717      * hpt-max-page-size capability. Of course we can't do it here
4718      * because this is too early and the HW accelerator isn't initialzed
4719      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4720      */
4721     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4722 }
4723 
4724 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4725 
4726 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4727 {
4728     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4729 
4730     spapr_machine_2_12_class_options(mc);
4731     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4732     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4733     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4734 }
4735 
4736 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4737 
4738 /*
4739  * pseries-2.11
4740  */
4741 
4742 static void spapr_machine_2_11_class_options(MachineClass *mc)
4743 {
4744     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4745 
4746     spapr_machine_2_12_class_options(mc);
4747     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4748     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4749 }
4750 
4751 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4752 
4753 /*
4754  * pseries-2.10
4755  */
4756 
4757 static void spapr_machine_2_10_class_options(MachineClass *mc)
4758 {
4759     spapr_machine_2_11_class_options(mc);
4760     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4761 }
4762 
4763 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4764 
4765 /*
4766  * pseries-2.9
4767  */
4768 
4769 static void spapr_machine_2_9_class_options(MachineClass *mc)
4770 {
4771     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4772     static GlobalProperty compat[] = {
4773         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4774     };
4775 
4776     spapr_machine_2_10_class_options(mc);
4777     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4778     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4779     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4780     smc->pre_2_10_has_unused_icps = true;
4781     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4785 
4786 /*
4787  * pseries-2.8
4788  */
4789 
4790 static void spapr_machine_2_8_class_options(MachineClass *mc)
4791 {
4792     static GlobalProperty compat[] = {
4793         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4794     };
4795 
4796     spapr_machine_2_9_class_options(mc);
4797     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4798     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4799     mc->numa_mem_align_shift = 23;
4800 }
4801 
4802 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4803 
4804 /*
4805  * pseries-2.7
4806  */
4807 
4808 static void phb_placement_2_7(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     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4815     const uint64_t base_buid = 0x800000020000000ULL;
4816     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4817     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4818     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4819     const uint32_t max_index = 255;
4820     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4821 
4822     uint64_t ram_top = MACHINE(spapr)->ram_size;
4823     hwaddr phb0_base, phb_base;
4824     int i;
4825 
4826     /* Do we have device memory? */
4827     if (MACHINE(spapr)->maxram_size > ram_top) {
4828         /* Can't just use maxram_size, because there may be an
4829          * alignment gap between normal and device memory regions
4830          */
4831         ram_top = MACHINE(spapr)->device_memory->base +
4832             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4833     }
4834 
4835     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4836 
4837     if (index > max_index) {
4838         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4839                    max_index);
4840         return;
4841     }
4842 
4843     *buid = base_buid + index;
4844     for (i = 0; i < n_dma; ++i) {
4845         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4846     }
4847 
4848     phb_base = phb0_base + index * phb_spacing;
4849     *pio = phb_base + pio_offset;
4850     *mmio32 = phb_base + mmio_offset;
4851     /*
4852      * We don't set the 64-bit MMIO window, relying on the PHB's
4853      * fallback behaviour of automatically splitting a large "32-bit"
4854      * window into contiguous 32-bit and 64-bit windows
4855      */
4856 
4857     *nv2gpa = 0;
4858     *nv2atsd = 0;
4859 }
4860 
4861 static void spapr_machine_2_7_class_options(MachineClass *mc)
4862 {
4863     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4864     static GlobalProperty compat[] = {
4865         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4867         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4868         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4869     };
4870 
4871     spapr_machine_2_8_class_options(mc);
4872     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4873     mc->default_machine_opts = "modern-hotplug-events=off";
4874     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4875     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4876     smc->phb_placement = phb_placement_2_7;
4877 }
4878 
4879 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4880 
4881 /*
4882  * pseries-2.6
4883  */
4884 
4885 static void spapr_machine_2_6_class_options(MachineClass *mc)
4886 {
4887     static GlobalProperty compat[] = {
4888         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4889     };
4890 
4891     spapr_machine_2_7_class_options(mc);
4892     mc->has_hotpluggable_cpus = false;
4893     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4894     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4895 }
4896 
4897 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4898 
4899 /*
4900  * pseries-2.5
4901  */
4902 
4903 static void spapr_machine_2_5_class_options(MachineClass *mc)
4904 {
4905     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4906     static GlobalProperty compat[] = {
4907         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4908     };
4909 
4910     spapr_machine_2_6_class_options(mc);
4911     smc->use_ohci_by_default = true;
4912     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4913     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4914 }
4915 
4916 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4917 
4918 /*
4919  * pseries-2.4
4920  */
4921 
4922 static void spapr_machine_2_4_class_options(MachineClass *mc)
4923 {
4924     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4925 
4926     spapr_machine_2_5_class_options(mc);
4927     smc->dr_lmb_enabled = false;
4928     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4929 }
4930 
4931 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4932 
4933 /*
4934  * pseries-2.3
4935  */
4936 
4937 static void spapr_machine_2_3_class_options(MachineClass *mc)
4938 {
4939     static GlobalProperty compat[] = {
4940         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4941     };
4942     spapr_machine_2_4_class_options(mc);
4943     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4944     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4945 }
4946 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4947 
4948 /*
4949  * pseries-2.2
4950  */
4951 
4952 static void spapr_machine_2_2_class_options(MachineClass *mc)
4953 {
4954     static GlobalProperty compat[] = {
4955         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4956     };
4957 
4958     spapr_machine_2_3_class_options(mc);
4959     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4960     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4961     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4962 }
4963 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4964 
4965 /*
4966  * pseries-2.1
4967  */
4968 
4969 static void spapr_machine_2_1_class_options(MachineClass *mc)
4970 {
4971     spapr_machine_2_2_class_options(mc);
4972     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4973 }
4974 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4975 
4976 static void spapr_machine_register_types(void)
4977 {
4978     type_register_static(&spapr_machine_info);
4979 }
4980 
4981 type_init(spapr_machine_register_types)
4982