xref: /openbmc/qemu/hw/ppc/spapr.c (revision 8f86a408241264db605da9a1215a409475a60bd0)
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 void spapr_dt_pa_features(SpaprMachineState *spapr,
206                                  PowerPCCPU *cpu,
207                                  void *fdt, int offset)
208 {
209     uint8_t pa_features_206[] = { 6, 0,
210         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
211     uint8_t pa_features_207[] = { 24, 0,
212         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
213         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
214         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
215         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
216     uint8_t pa_features_300[] = { 66, 0,
217         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
218         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
219         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
220         /* 6: DS207 */
221         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
222         /* 16: Vector */
223         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
224         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
225         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
226         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
227         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
228         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
229         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
230         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
232         /* 42: PM, 44: PC RA, 46: SC vec'd */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
234         /* 48: SIMD, 50: QP BFP, 52: String */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
236         /* 54: DecFP, 56: DecI, 58: SHA */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
238         /* 60: NM atomic, 62: RNG */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
240     };
241     uint8_t *pa_features = NULL;
242     size_t pa_size;
243 
244     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
245         pa_features = pa_features_206;
246         pa_size = sizeof(pa_features_206);
247     }
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_207;
250         pa_size = sizeof(pa_features_207);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_300;
254         pa_size = sizeof(pa_features_300);
255     }
256     if (!pa_features) {
257         return;
258     }
259 
260     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
261         /*
262          * Note: we keep CI large pages off by default because a 64K capable
263          * guest provisioned with large pages might otherwise try to map a qemu
264          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
265          * even if that qemu runs on a 4k host.
266          * We dd this bit back here if we are confident this is not an issue
267          */
268         pa_features[3] |= 0x20;
269     }
270     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
271         pa_features[24] |= 0x80;    /* Transactional memory support */
272     }
273     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
274         /* Workaround for broken kernels that attempt (guest) radix
275          * mode when they can't handle it, if they see the radix bit set
276          * in pa-features. So hide it from them. */
277         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
278     }
279 
280     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
281 }
282 
283 static hwaddr spapr_node0_size(MachineState *machine)
284 {
285     if (machine->numa_state->num_nodes) {
286         int i;
287         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
288             if (machine->numa_state->nodes[i].node_mem) {
289                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
290                            machine->ram_size);
291             }
292         }
293     }
294     return machine->ram_size;
295 }
296 
297 static void add_str(GString *s, const gchar *s1)
298 {
299     g_string_append_len(s, s1, strlen(s1) + 1);
300 }
301 
302 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
303                                 hwaddr start, hwaddr size)
304 {
305     char mem_name[32];
306     uint64_t mem_reg_property[2];
307     int off;
308 
309     mem_reg_property[0] = cpu_to_be64(start);
310     mem_reg_property[1] = cpu_to_be64(size);
311 
312     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
313     off = fdt_add_subnode(fdt, 0, mem_name);
314     _FDT(off);
315     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
316     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
317                       sizeof(mem_reg_property))));
318     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
319     return off;
320 }
321 
322 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
323 {
324     MemoryDeviceInfoList *info;
325 
326     for (info = list; info; info = info->next) {
327         MemoryDeviceInfo *value = info->value;
328 
329         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
330             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
331 
332             if (addr >= pcdimm_info->addr &&
333                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
334                 return pcdimm_info->node;
335             }
336         }
337     }
338 
339     return -1;
340 }
341 
342 struct sPAPRDrconfCellV2 {
343      uint32_t seq_lmbs;
344      uint64_t base_addr;
345      uint32_t drc_index;
346      uint32_t aa_index;
347      uint32_t flags;
348 } QEMU_PACKED;
349 
350 typedef struct DrconfCellQueue {
351     struct sPAPRDrconfCellV2 cell;
352     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
353 } DrconfCellQueue;
354 
355 static DrconfCellQueue *
356 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
357                       uint32_t drc_index, uint32_t aa_index,
358                       uint32_t flags)
359 {
360     DrconfCellQueue *elem;
361 
362     elem = g_malloc0(sizeof(*elem));
363     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
364     elem->cell.base_addr = cpu_to_be64(base_addr);
365     elem->cell.drc_index = cpu_to_be32(drc_index);
366     elem->cell.aa_index = cpu_to_be32(aa_index);
367     elem->cell.flags = cpu_to_be32(flags);
368 
369     return elem;
370 }
371 
372 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
373                                       int offset, MemoryDeviceInfoList *dimms)
374 {
375     MachineState *machine = MACHINE(spapr);
376     uint8_t *int_buf, *cur_index;
377     int ret;
378     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
379     uint64_t addr, cur_addr, size;
380     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
381     uint64_t mem_end = machine->device_memory->base +
382                        memory_region_size(&machine->device_memory->mr);
383     uint32_t node, buf_len, nr_entries = 0;
384     SpaprDrc *drc;
385     DrconfCellQueue *elem, *next;
386     MemoryDeviceInfoList *info;
387     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
388         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
389 
390     /* Entry to cover RAM and the gap area */
391     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
392                                  SPAPR_LMB_FLAGS_RESERVED |
393                                  SPAPR_LMB_FLAGS_DRC_INVALID);
394     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
395     nr_entries++;
396 
397     cur_addr = machine->device_memory->base;
398     for (info = dimms; info; info = info->next) {
399         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
400 
401         addr = di->addr;
402         size = di->size;
403         node = di->node;
404 
405         /*
406          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
407          * area is marked hotpluggable in the next iteration for the bigger
408          * chunk including the NVDIMM occupied area.
409          */
410         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
411             continue;
412 
413         /* Entry for hot-pluggable area */
414         if (cur_addr < addr) {
415             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
416             g_assert(drc);
417             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
418                                          cur_addr, spapr_drc_index(drc), -1, 0);
419             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
420             nr_entries++;
421         }
422 
423         /* Entry for DIMM */
424         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
425         g_assert(drc);
426         elem = spapr_get_drconf_cell(size / lmb_size, addr,
427                                      spapr_drc_index(drc), node,
428                                      (SPAPR_LMB_FLAGS_ASSIGNED |
429                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
430         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
431         nr_entries++;
432         cur_addr = addr + size;
433     }
434 
435     /* Entry for remaining hotpluggable area */
436     if (cur_addr < mem_end) {
437         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
438         g_assert(drc);
439         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
440                                      cur_addr, spapr_drc_index(drc), -1, 0);
441         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
442         nr_entries++;
443     }
444 
445     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
446     int_buf = cur_index = g_malloc0(buf_len);
447     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
448     cur_index += sizeof(nr_entries);
449 
450     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
451         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
452         cur_index += sizeof(elem->cell);
453         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
454         g_free(elem);
455     }
456 
457     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
458     g_free(int_buf);
459     if (ret < 0) {
460         return -1;
461     }
462     return 0;
463 }
464 
465 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
466                                    int offset, MemoryDeviceInfoList *dimms)
467 {
468     MachineState *machine = MACHINE(spapr);
469     int i, ret;
470     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
471     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
472     uint32_t nr_lmbs = (machine->device_memory->base +
473                        memory_region_size(&machine->device_memory->mr)) /
474                        lmb_size;
475     uint32_t *int_buf, *cur_index, buf_len;
476 
477     /*
478      * Allocate enough buffer size to fit in ibm,dynamic-memory
479      */
480     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
481     cur_index = int_buf = g_malloc0(buf_len);
482     int_buf[0] = cpu_to_be32(nr_lmbs);
483     cur_index++;
484     for (i = 0; i < nr_lmbs; i++) {
485         uint64_t addr = i * lmb_size;
486         uint32_t *dynamic_memory = cur_index;
487 
488         if (i >= device_lmb_start) {
489             SpaprDrc *drc;
490 
491             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
492             g_assert(drc);
493 
494             dynamic_memory[0] = cpu_to_be32(addr >> 32);
495             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
496             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
497             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
498             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
499             if (memory_region_present(get_system_memory(), addr)) {
500                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
501             } else {
502                 dynamic_memory[5] = cpu_to_be32(0);
503             }
504         } else {
505             /*
506              * LMB information for RMA, boot time RAM and gap b/n RAM and
507              * device memory region -- all these are marked as reserved
508              * and as having no valid DRC.
509              */
510             dynamic_memory[0] = cpu_to_be32(addr >> 32);
511             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
512             dynamic_memory[2] = cpu_to_be32(0);
513             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
514             dynamic_memory[4] = cpu_to_be32(-1);
515             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
516                                             SPAPR_LMB_FLAGS_DRC_INVALID);
517         }
518 
519         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
520     }
521     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
522     g_free(int_buf);
523     if (ret < 0) {
524         return -1;
525     }
526     return 0;
527 }
528 
529 /*
530  * Adds ibm,dynamic-reconfiguration-memory node.
531  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
532  * of this device tree node.
533  */
534 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
535                                                    void *fdt)
536 {
537     MachineState *machine = MACHINE(spapr);
538     int nb_numa_nodes = machine->numa_state->num_nodes;
539     int ret, i, offset;
540     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
541     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
542                                 cpu_to_be32(lmb_size & 0xffffffff)};
543     uint32_t *int_buf, *cur_index, buf_len;
544     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
545     MemoryDeviceInfoList *dimms = NULL;
546 
547     /*
548      * Don't create the node if there is no device memory
549      */
550     if (machine->ram_size == machine->maxram_size) {
551         return 0;
552     }
553 
554     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
555 
556     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557                     sizeof(prop_lmb_size));
558     if (ret < 0) {
559         return ret;
560     }
561 
562     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
563     if (ret < 0) {
564         return ret;
565     }
566 
567     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
568     if (ret < 0) {
569         return ret;
570     }
571 
572     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573     dimms = qmp_memory_device_list();
574     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
576     } else {
577         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
578     }
579     qapi_free_MemoryDeviceInfoList(dimms);
580 
581     if (ret < 0) {
582         return ret;
583     }
584 
585     /* ibm,associativity-lookup-arrays */
586     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
587     cur_index = int_buf = g_malloc0(buf_len);
588     int_buf[0] = cpu_to_be32(nr_nodes);
589     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
590     cur_index += 2;
591     for (i = 0; i < nr_nodes; i++) {
592         uint32_t associativity[] = {
593             cpu_to_be32(0x0),
594             cpu_to_be32(0x0),
595             cpu_to_be32(0x0),
596             cpu_to_be32(i)
597         };
598         memcpy(cur_index, associativity, sizeof(associativity));
599         cur_index += 4;
600     }
601     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
602             (cur_index - int_buf) * sizeof(uint32_t));
603     g_free(int_buf);
604 
605     return ret;
606 }
607 
608 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
609 {
610     MachineState *machine = MACHINE(spapr);
611     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
612     hwaddr mem_start, node_size;
613     int i, nb_nodes = machine->numa_state->num_nodes;
614     NodeInfo *nodes = machine->numa_state->nodes;
615 
616     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
617         if (!nodes[i].node_mem) {
618             continue;
619         }
620         if (mem_start >= machine->ram_size) {
621             node_size = 0;
622         } else {
623             node_size = nodes[i].node_mem;
624             if (node_size > machine->ram_size - mem_start) {
625                 node_size = machine->ram_size - mem_start;
626             }
627         }
628         if (!mem_start) {
629             /* spapr_machine_init() checks for rma_size <= node0_size
630              * already */
631             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
632             mem_start += spapr->rma_size;
633             node_size -= spapr->rma_size;
634         }
635         for ( ; node_size; ) {
636             hwaddr sizetmp = pow2floor(node_size);
637 
638             /* mem_start != 0 here */
639             if (ctzl(mem_start) < ctzl(sizetmp)) {
640                 sizetmp = 1ULL << ctzl(mem_start);
641             }
642 
643             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
644             node_size -= sizetmp;
645             mem_start += sizetmp;
646         }
647     }
648 
649     /* Generate ibm,dynamic-reconfiguration-memory node if required */
650     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
651         int ret;
652 
653         g_assert(smc->dr_lmb_enabled);
654         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
655         if (ret) {
656             return ret;
657         }
658     }
659 
660     return 0;
661 }
662 
663 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
664                          SpaprMachineState *spapr)
665 {
666     MachineState *ms = MACHINE(spapr);
667     PowerPCCPU *cpu = POWERPC_CPU(cs);
668     CPUPPCState *env = &cpu->env;
669     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
670     int index = spapr_get_vcpu_id(cpu);
671     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
672                        0xffffffff, 0xffffffff};
673     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
674         : SPAPR_TIMEBASE_FREQ;
675     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
676     uint32_t page_sizes_prop[64];
677     size_t page_sizes_prop_size;
678     unsigned int smp_threads = ms->smp.threads;
679     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
680     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
681     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
682     SpaprDrc *drc;
683     int drc_index;
684     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
685     int i;
686 
687     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
688     if (drc) {
689         drc_index = spapr_drc_index(drc);
690         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
691     }
692 
693     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
694     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
695 
696     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
697     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
698                            env->dcache_line_size)));
699     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
700                            env->dcache_line_size)));
701     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
702                            env->icache_line_size)));
703     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
704                            env->icache_line_size)));
705 
706     if (pcc->l1_dcache_size) {
707         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
708                                pcc->l1_dcache_size)));
709     } else {
710         warn_report("Unknown L1 dcache size for cpu");
711     }
712     if (pcc->l1_icache_size) {
713         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
714                                pcc->l1_icache_size)));
715     } else {
716         warn_report("Unknown L1 icache size for cpu");
717     }
718 
719     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
720     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
721     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
723     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
724     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
725 
726     if (env->spr_cb[SPR_PURR].oea_read) {
727         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
728     }
729     if (env->spr_cb[SPR_SPURR].oea_read) {
730         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
731     }
732 
733     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
734         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
735                           segs, sizeof(segs))));
736     }
737 
738     /* Advertise VSX (vector extensions) if available
739      *   1               == VMX / Altivec available
740      *   2               == VSX available
741      *
742      * Only CPUs for which we create core types in spapr_cpu_core.c
743      * are possible, and all of those have VMX */
744     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
745         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
746     } else {
747         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
748     }
749 
750     /* Advertise DFP (Decimal Floating Point) if available
751      *   0 / no property == no DFP
752      *   1               == DFP available */
753     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
754         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
755     }
756 
757     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
758                                                       sizeof(page_sizes_prop));
759     if (page_sizes_prop_size) {
760         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
761                           page_sizes_prop, page_sizes_prop_size)));
762     }
763 
764     spapr_dt_pa_features(spapr, cpu, fdt, offset);
765 
766     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
767                            cs->cpu_index / vcpus_per_socket)));
768 
769     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
770                       pft_size_prop, sizeof(pft_size_prop))));
771 
772     if (ms->numa_state->num_nodes > 1) {
773         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
774     }
775 
776     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
777 
778     if (pcc->radix_page_info) {
779         for (i = 0; i < pcc->radix_page_info->count; i++) {
780             radix_AP_encodings[i] =
781                 cpu_to_be32(pcc->radix_page_info->entries[i]);
782         }
783         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
784                           radix_AP_encodings,
785                           pcc->radix_page_info->count *
786                           sizeof(radix_AP_encodings[0]))));
787     }
788 
789     /*
790      * We set this property to let the guest know that it can use the large
791      * decrementer and its width in bits.
792      */
793     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
794         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
795                               pcc->lrg_decr_bits)));
796 }
797 
798 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
799 {
800     CPUState **rev;
801     CPUState *cs;
802     int n_cpus;
803     int cpus_offset;
804     char *nodename;
805     int i;
806 
807     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
808     _FDT(cpus_offset);
809     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
810     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
811 
812     /*
813      * We walk the CPUs in reverse order to ensure that CPU DT nodes
814      * created by fdt_add_subnode() end up in the right order in FDT
815      * for the guest kernel the enumerate the CPUs correctly.
816      *
817      * The CPU list cannot be traversed in reverse order, so we need
818      * to do extra work.
819      */
820     n_cpus = 0;
821     rev = NULL;
822     CPU_FOREACH(cs) {
823         rev = g_renew(CPUState *, rev, n_cpus + 1);
824         rev[n_cpus++] = cs;
825     }
826 
827     for (i = n_cpus - 1; i >= 0; i--) {
828         CPUState *cs = rev[i];
829         PowerPCCPU *cpu = POWERPC_CPU(cs);
830         int index = spapr_get_vcpu_id(cpu);
831         DeviceClass *dc = DEVICE_GET_CLASS(cs);
832         int offset;
833 
834         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
835             continue;
836         }
837 
838         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
839         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
840         g_free(nodename);
841         _FDT(offset);
842         spapr_dt_cpu(cs, fdt, offset, spapr);
843     }
844 
845     g_free(rev);
846 }
847 
848 static int spapr_dt_rng(void *fdt)
849 {
850     int node;
851     int ret;
852 
853     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
854     if (node <= 0) {
855         return -1;
856     }
857     ret = fdt_setprop_string(fdt, node, "device_type",
858                              "ibm,platform-facilities");
859     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
860     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
861 
862     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
863     if (node <= 0) {
864         return -1;
865     }
866     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
867 
868     return ret ? -1 : 0;
869 }
870 
871 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
872 {
873     MachineState *ms = MACHINE(spapr);
874     int rtas;
875     GString *hypertas = g_string_sized_new(256);
876     GString *qemu_hypertas = g_string_sized_new(256);
877     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
878         memory_region_size(&MACHINE(spapr)->device_memory->mr);
879     uint32_t lrdr_capacity[] = {
880         cpu_to_be32(max_device_addr >> 32),
881         cpu_to_be32(max_device_addr & 0xffffffff),
882         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
883         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
884         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
885     };
886 
887     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
888 
889     /* hypertas */
890     add_str(hypertas, "hcall-pft");
891     add_str(hypertas, "hcall-term");
892     add_str(hypertas, "hcall-dabr");
893     add_str(hypertas, "hcall-interrupt");
894     add_str(hypertas, "hcall-tce");
895     add_str(hypertas, "hcall-vio");
896     add_str(hypertas, "hcall-splpar");
897     add_str(hypertas, "hcall-join");
898     add_str(hypertas, "hcall-bulk");
899     add_str(hypertas, "hcall-set-mode");
900     add_str(hypertas, "hcall-sprg0");
901     add_str(hypertas, "hcall-copy");
902     add_str(hypertas, "hcall-debug");
903     add_str(hypertas, "hcall-vphn");
904     add_str(qemu_hypertas, "hcall-memop1");
905 
906     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
907         add_str(hypertas, "hcall-multi-tce");
908     }
909 
910     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
911         add_str(hypertas, "hcall-hpt-resize");
912     }
913 
914     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
915                      hypertas->str, hypertas->len));
916     g_string_free(hypertas, TRUE);
917     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
918                      qemu_hypertas->str, qemu_hypertas->len));
919     g_string_free(qemu_hypertas, TRUE);
920 
921     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
922 
923     /*
924      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
925      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
926      *
927      * The system reset requirements are driven by existing Linux and PowerVM
928      * implementation which (contrary to PAPR) saves r3 in the error log
929      * structure like machine check, so Linux expects to find the saved r3
930      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
931      * does not look at the error value).
932      *
933      * System reset interrupts are not subject to interlock like machine
934      * check, so this memory area could be corrupted if the sreset is
935      * interrupted by a machine check (or vice versa) if it was shared. To
936      * prevent this, system reset uses per-CPU areas for the sreset save
937      * area. A system reset that interrupts a system reset handler could
938      * still overwrite this area, but Linux doesn't try to recover in that
939      * case anyway.
940      *
941      * The extra 8 bytes is required because Linux's FWNMI error log check
942      * is off-by-one.
943      */
944     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
945 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
946     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
947                           RTAS_ERROR_LOG_MAX));
948     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
949                           RTAS_EVENT_SCAN_RATE));
950 
951     g_assert(msi_nonbroken);
952     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
953 
954     /*
955      * According to PAPR, rtas ibm,os-term does not guarantee a return
956      * back to the guest cpu.
957      *
958      * While an additional ibm,extended-os-term property indicates
959      * that rtas call return will always occur. Set this property.
960      */
961     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
962 
963     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
964                      lrdr_capacity, sizeof(lrdr_capacity)));
965 
966     spapr_dt_rtas_tokens(fdt, rtas);
967 }
968 
969 /*
970  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
971  * and the XIVE features that the guest may request and thus the valid
972  * values for bytes 23..26 of option vector 5:
973  */
974 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
975                                           int chosen)
976 {
977     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
978 
979     char val[2 * 4] = {
980         23, 0x00, /* XICS / XIVE mode */
981         24, 0x00, /* Hash/Radix, filled in below. */
982         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
983         26, 0x40, /* Radix options: GTSE == yes. */
984     };
985 
986     if (spapr->irq->xics && spapr->irq->xive) {
987         val[1] = SPAPR_OV5_XIVE_BOTH;
988     } else if (spapr->irq->xive) {
989         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
990     } else {
991         assert(spapr->irq->xics);
992         val[1] = SPAPR_OV5_XIVE_LEGACY;
993     }
994 
995     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
996                           first_ppc_cpu->compat_pvr)) {
997         /*
998          * If we're in a pre POWER9 compat mode then the guest should
999          * do hash and use the legacy interrupt mode
1000          */
1001         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1002         val[3] = 0x00; /* Hash */
1003     } else if (kvm_enabled()) {
1004         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1005             val[3] = 0x80; /* OV5_MMU_BOTH */
1006         } else if (kvmppc_has_cap_mmu_radix()) {
1007             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1008         } else {
1009             val[3] = 0x00; /* Hash */
1010         }
1011     } else {
1012         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1013         val[3] = 0xC0;
1014     }
1015     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1016                      val, sizeof(val)));
1017 }
1018 
1019 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1020 {
1021     MachineState *machine = MACHINE(spapr);
1022     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1023     int chosen;
1024 
1025     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1026 
1027     if (reset) {
1028         const char *boot_device = machine->boot_order;
1029         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1030         size_t cb = 0;
1031         char *bootlist = get_boot_devices_list(&cb);
1032 
1033         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1034             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1035                                     machine->kernel_cmdline));
1036         }
1037 
1038         if (spapr->initrd_size) {
1039             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1040                                   spapr->initrd_base));
1041             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1042                                   spapr->initrd_base + spapr->initrd_size));
1043         }
1044 
1045         if (spapr->kernel_size) {
1046             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1047                                   cpu_to_be64(spapr->kernel_size) };
1048 
1049             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1050                          &kprop, sizeof(kprop)));
1051             if (spapr->kernel_le) {
1052                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1053             }
1054         }
1055         if (boot_menu) {
1056             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1057         }
1058         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1059         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1060         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1061 
1062         if (cb && bootlist) {
1063             int i;
1064 
1065             for (i = 0; i < cb; i++) {
1066                 if (bootlist[i] == '\n') {
1067                     bootlist[i] = ' ';
1068                 }
1069             }
1070             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1071         }
1072 
1073         if (boot_device && strlen(boot_device)) {
1074             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1075         }
1076 
1077         if (!spapr->has_graphics && stdout_path) {
1078             /*
1079              * "linux,stdout-path" and "stdout" properties are
1080              * deprecated by linux kernel. New platforms should only
1081              * use the "stdout-path" property. Set the new property
1082              * and continue using older property to remain compatible
1083              * with the existing firmware.
1084              */
1085             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1086             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1087         }
1088 
1089         /*
1090          * We can deal with BAR reallocation just fine, advertise it
1091          * to the guest
1092          */
1093         if (smc->linux_pci_probe) {
1094             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1095         }
1096 
1097         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1098 
1099         g_free(stdout_path);
1100         g_free(bootlist);
1101     }
1102 
1103     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1104 }
1105 
1106 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1107 {
1108     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1109      * KVM to work under pHyp with some guest co-operation */
1110     int hypervisor;
1111     uint8_t hypercall[16];
1112 
1113     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1114     /* indicate KVM hypercall interface */
1115     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1116     if (kvmppc_has_cap_fixup_hcalls()) {
1117         /*
1118          * Older KVM versions with older guest kernels were broken
1119          * with the magic page, don't allow the guest to map it.
1120          */
1121         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1122                                   sizeof(hypercall))) {
1123             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1124                              hypercall, sizeof(hypercall)));
1125         }
1126     }
1127 }
1128 
1129 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1130 {
1131     MachineState *machine = MACHINE(spapr);
1132     MachineClass *mc = MACHINE_GET_CLASS(machine);
1133     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1134     int ret;
1135     void *fdt;
1136     SpaprPhbState *phb;
1137     char *buf;
1138 
1139     fdt = g_malloc0(space);
1140     _FDT((fdt_create_empty_tree(fdt, space)));
1141 
1142     /* Root node */
1143     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1144     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1145     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1146 
1147     /* Guest UUID & Name*/
1148     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1149     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1150     if (qemu_uuid_set) {
1151         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1152     }
1153     g_free(buf);
1154 
1155     if (qemu_get_vm_name()) {
1156         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1157                                 qemu_get_vm_name()));
1158     }
1159 
1160     /* Host Model & Serial Number */
1161     if (spapr->host_model) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1163     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1164         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1165         g_free(buf);
1166     }
1167 
1168     if (spapr->host_serial) {
1169         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1170     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1171         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1172         g_free(buf);
1173     }
1174 
1175     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1176     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1177 
1178     /* /interrupt controller */
1179     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1180 
1181     ret = spapr_dt_memory(spapr, fdt);
1182     if (ret < 0) {
1183         error_report("couldn't setup memory nodes in fdt");
1184         exit(1);
1185     }
1186 
1187     /* /vdevice */
1188     spapr_dt_vdevice(spapr->vio_bus, fdt);
1189 
1190     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1191         ret = spapr_dt_rng(fdt);
1192         if (ret < 0) {
1193             error_report("could not set up rng device in the fdt");
1194             exit(1);
1195         }
1196     }
1197 
1198     QLIST_FOREACH(phb, &spapr->phbs, list) {
1199         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1200         if (ret < 0) {
1201             error_report("couldn't setup PCI devices in fdt");
1202             exit(1);
1203         }
1204     }
1205 
1206     spapr_dt_cpus(fdt, spapr);
1207 
1208     if (smc->dr_lmb_enabled) {
1209         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1210     }
1211 
1212     if (mc->has_hotpluggable_cpus) {
1213         int offset = fdt_path_offset(fdt, "/cpus");
1214         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1215         if (ret < 0) {
1216             error_report("Couldn't set up CPU DR device tree properties");
1217             exit(1);
1218         }
1219     }
1220 
1221     /* /event-sources */
1222     spapr_dt_events(spapr, fdt);
1223 
1224     /* /rtas */
1225     spapr_dt_rtas(spapr, fdt);
1226 
1227     /* /chosen */
1228     spapr_dt_chosen(spapr, fdt, reset);
1229 
1230     /* /hypervisor */
1231     if (kvm_enabled()) {
1232         spapr_dt_hypervisor(spapr, fdt);
1233     }
1234 
1235     /* Build memory reserve map */
1236     if (reset) {
1237         if (spapr->kernel_size) {
1238             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1239                                   spapr->kernel_size)));
1240         }
1241         if (spapr->initrd_size) {
1242             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1243                                   spapr->initrd_size)));
1244         }
1245     }
1246 
1247     if (smc->dr_phb_enabled) {
1248         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1249         if (ret < 0) {
1250             error_report("Couldn't set up PHB DR device tree properties");
1251             exit(1);
1252         }
1253     }
1254 
1255     /* NVDIMM devices */
1256     if (mc->nvdimm_supported) {
1257         spapr_dt_persistent_memory(spapr, fdt);
1258     }
1259 
1260     return fdt;
1261 }
1262 
1263 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1264 {
1265     SpaprMachineState *spapr = opaque;
1266 
1267     return (addr & 0x0fffffff) + spapr->kernel_addr;
1268 }
1269 
1270 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1271                                     PowerPCCPU *cpu)
1272 {
1273     CPUPPCState *env = &cpu->env;
1274 
1275     /* The TCG path should also be holding the BQL at this point */
1276     g_assert(qemu_mutex_iothread_locked());
1277 
1278     if (msr_pr) {
1279         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1280         env->gpr[3] = H_PRIVILEGE;
1281     } else {
1282         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1283     }
1284 }
1285 
1286 struct LPCRSyncState {
1287     target_ulong value;
1288     target_ulong mask;
1289 };
1290 
1291 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1292 {
1293     struct LPCRSyncState *s = arg.host_ptr;
1294     PowerPCCPU *cpu = POWERPC_CPU(cs);
1295     CPUPPCState *env = &cpu->env;
1296     target_ulong lpcr;
1297 
1298     cpu_synchronize_state(cs);
1299     lpcr = env->spr[SPR_LPCR];
1300     lpcr &= ~s->mask;
1301     lpcr |= s->value;
1302     ppc_store_lpcr(cpu, lpcr);
1303 }
1304 
1305 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1306 {
1307     CPUState *cs;
1308     struct LPCRSyncState s = {
1309         .value = value,
1310         .mask = mask
1311     };
1312     CPU_FOREACH(cs) {
1313         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1314     }
1315 }
1316 
1317 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1318 {
1319     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1320 
1321     /* Copy PATE1:GR into PATE0:HR */
1322     entry->dw0 = spapr->patb_entry & PATE0_HR;
1323     entry->dw1 = spapr->patb_entry;
1324 }
1325 
1326 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1327 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1328 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1329 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1330 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1331 
1332 /*
1333  * Get the fd to access the kernel htab, re-opening it if necessary
1334  */
1335 static int get_htab_fd(SpaprMachineState *spapr)
1336 {
1337     Error *local_err = NULL;
1338 
1339     if (spapr->htab_fd >= 0) {
1340         return spapr->htab_fd;
1341     }
1342 
1343     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1344     if (spapr->htab_fd < 0) {
1345         error_report_err(local_err);
1346     }
1347 
1348     return spapr->htab_fd;
1349 }
1350 
1351 void close_htab_fd(SpaprMachineState *spapr)
1352 {
1353     if (spapr->htab_fd >= 0) {
1354         close(spapr->htab_fd);
1355     }
1356     spapr->htab_fd = -1;
1357 }
1358 
1359 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1360 {
1361     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1362 
1363     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1364 }
1365 
1366 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1367 {
1368     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1369 
1370     assert(kvm_enabled());
1371 
1372     if (!spapr->htab) {
1373         return 0;
1374     }
1375 
1376     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1377 }
1378 
1379 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1380                                                 hwaddr ptex, int n)
1381 {
1382     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1383     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1384 
1385     if (!spapr->htab) {
1386         /*
1387          * HTAB is controlled by KVM. Fetch into temporary buffer
1388          */
1389         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1390         kvmppc_read_hptes(hptes, ptex, n);
1391         return hptes;
1392     }
1393 
1394     /*
1395      * HTAB is controlled by QEMU. Just point to the internally
1396      * accessible PTEG.
1397      */
1398     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1399 }
1400 
1401 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1402                               const ppc_hash_pte64_t *hptes,
1403                               hwaddr ptex, int n)
1404 {
1405     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1406 
1407     if (!spapr->htab) {
1408         g_free((void *)hptes);
1409     }
1410 
1411     /* Nothing to do for qemu managed HPT */
1412 }
1413 
1414 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1415                       uint64_t pte0, uint64_t pte1)
1416 {
1417     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1418     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1419 
1420     if (!spapr->htab) {
1421         kvmppc_write_hpte(ptex, pte0, pte1);
1422     } else {
1423         if (pte0 & HPTE64_V_VALID) {
1424             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1425             /*
1426              * When setting valid, we write PTE1 first. This ensures
1427              * proper synchronization with the reading code in
1428              * ppc_hash64_pteg_search()
1429              */
1430             smp_wmb();
1431             stq_p(spapr->htab + offset, pte0);
1432         } else {
1433             stq_p(spapr->htab + offset, pte0);
1434             /*
1435              * When clearing it we set PTE0 first. This ensures proper
1436              * synchronization with the reading code in
1437              * ppc_hash64_pteg_search()
1438              */
1439             smp_wmb();
1440             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1441         }
1442     }
1443 }
1444 
1445 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1446                              uint64_t pte1)
1447 {
1448     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1449     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1450 
1451     if (!spapr->htab) {
1452         /* There should always be a hash table when this is called */
1453         error_report("spapr_hpte_set_c called with no hash table !");
1454         return;
1455     }
1456 
1457     /* The HW performs a non-atomic byte update */
1458     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1459 }
1460 
1461 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1462                              uint64_t pte1)
1463 {
1464     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1465     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1466 
1467     if (!spapr->htab) {
1468         /* There should always be a hash table when this is called */
1469         error_report("spapr_hpte_set_r called with no hash table !");
1470         return;
1471     }
1472 
1473     /* The HW performs a non-atomic byte update */
1474     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1475 }
1476 
1477 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1478 {
1479     int shift;
1480 
1481     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1482      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1483      * that's much more than is needed for Linux guests */
1484     shift = ctz64(pow2ceil(ramsize)) - 7;
1485     shift = MAX(shift, 18); /* Minimum architected size */
1486     shift = MIN(shift, 46); /* Maximum architected size */
1487     return shift;
1488 }
1489 
1490 void spapr_free_hpt(SpaprMachineState *spapr)
1491 {
1492     g_free(spapr->htab);
1493     spapr->htab = NULL;
1494     spapr->htab_shift = 0;
1495     close_htab_fd(spapr);
1496 }
1497 
1498 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1499                           Error **errp)
1500 {
1501     long rc;
1502 
1503     /* Clean up any HPT info from a previous boot */
1504     spapr_free_hpt(spapr);
1505 
1506     rc = kvmppc_reset_htab(shift);
1507     if (rc < 0) {
1508         /* kernel-side HPT needed, but couldn't allocate one */
1509         error_setg_errno(errp, errno,
1510                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1511                          shift);
1512         /* This is almost certainly fatal, but if the caller really
1513          * wants to carry on with shift == 0, it's welcome to try */
1514     } else if (rc > 0) {
1515         /* kernel-side HPT allocated */
1516         if (rc != shift) {
1517             error_setg(errp,
1518                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1519                        shift, rc);
1520         }
1521 
1522         spapr->htab_shift = shift;
1523         spapr->htab = NULL;
1524     } else {
1525         /* kernel-side HPT not needed, allocate in userspace instead */
1526         size_t size = 1ULL << shift;
1527         int i;
1528 
1529         spapr->htab = qemu_memalign(size, size);
1530         if (!spapr->htab) {
1531             error_setg_errno(errp, errno,
1532                              "Could not allocate HPT of order %d", shift);
1533             return;
1534         }
1535 
1536         memset(spapr->htab, 0, size);
1537         spapr->htab_shift = shift;
1538 
1539         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1540             DIRTY_HPTE(HPTE(spapr->htab, i));
1541         }
1542     }
1543     /* We're setting up a hash table, so that means we're not radix */
1544     spapr->patb_entry = 0;
1545     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1546 }
1547 
1548 void spapr_setup_hpt(SpaprMachineState *spapr)
1549 {
1550     int hpt_shift;
1551 
1552     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1553         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1554     } else {
1555         uint64_t current_ram_size;
1556 
1557         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1558         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1559     }
1560     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1561 
1562     if (kvm_enabled()) {
1563         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1564 
1565         /* Check our RMA fits in the possible VRMA */
1566         if (vrma_limit < spapr->rma_size) {
1567             error_report("Unable to create %" HWADDR_PRIu
1568                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1569                          spapr->rma_size / MiB, vrma_limit / MiB);
1570             exit(EXIT_FAILURE);
1571         }
1572     }
1573 }
1574 
1575 static int spapr_reset_drcs(Object *child, void *opaque)
1576 {
1577     SpaprDrc *drc =
1578         (SpaprDrc *) object_dynamic_cast(child,
1579                                                  TYPE_SPAPR_DR_CONNECTOR);
1580 
1581     if (drc) {
1582         spapr_drc_reset(drc);
1583     }
1584 
1585     return 0;
1586 }
1587 
1588 static void spapr_machine_reset(MachineState *machine)
1589 {
1590     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1591     PowerPCCPU *first_ppc_cpu;
1592     hwaddr fdt_addr;
1593     void *fdt;
1594     int rc;
1595 
1596     kvmppc_svm_off(&error_fatal);
1597     spapr_caps_apply(spapr);
1598 
1599     first_ppc_cpu = POWERPC_CPU(first_cpu);
1600     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1601         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1602                               spapr->max_compat_pvr)) {
1603         /*
1604          * If using KVM with radix mode available, VCPUs can be started
1605          * without a HPT because KVM will start them in radix mode.
1606          * Set the GR bit in PATE so that we know there is no HPT.
1607          */
1608         spapr->patb_entry = PATE1_GR;
1609         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1610     } else {
1611         spapr_setup_hpt(spapr);
1612     }
1613 
1614     qemu_devices_reset();
1615 
1616     spapr_ovec_cleanup(spapr->ov5_cas);
1617     spapr->ov5_cas = spapr_ovec_new();
1618 
1619     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1620 
1621     /*
1622      * This is fixing some of the default configuration of the XIVE
1623      * devices. To be called after the reset of the machine devices.
1624      */
1625     spapr_irq_reset(spapr, &error_fatal);
1626 
1627     /*
1628      * There is no CAS under qtest. Simulate one to please the code that
1629      * depends on spapr->ov5_cas. This is especially needed to test device
1630      * unplug, so we do that before resetting the DRCs.
1631      */
1632     if (qtest_enabled()) {
1633         spapr_ovec_cleanup(spapr->ov5_cas);
1634         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1635     }
1636 
1637     /* DRC reset may cause a device to be unplugged. This will cause troubles
1638      * if this device is used by another device (eg, a running vhost backend
1639      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1640      * situations, we reset DRCs after all devices have been reset.
1641      */
1642     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1643 
1644     spapr_clear_pending_events(spapr);
1645 
1646     /*
1647      * We place the device tree and RTAS just below either the top of the RMA,
1648      * or just below 2GB, whichever is lower, so that it can be
1649      * processed with 32-bit real mode code if necessary
1650      */
1651     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1652 
1653     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1654 
1655     rc = fdt_pack(fdt);
1656 
1657     /* Should only fail if we've built a corrupted tree */
1658     assert(rc == 0);
1659 
1660     /* Load the fdt */
1661     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1662     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1663     g_free(spapr->fdt_blob);
1664     spapr->fdt_size = fdt_totalsize(fdt);
1665     spapr->fdt_initial_size = spapr->fdt_size;
1666     spapr->fdt_blob = fdt;
1667 
1668     /* Set up the entry state */
1669     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1670     first_ppc_cpu->env.gpr[5] = 0;
1671 
1672     spapr->fwnmi_system_reset_addr = -1;
1673     spapr->fwnmi_machine_check_addr = -1;
1674     spapr->fwnmi_machine_check_interlock = -1;
1675 
1676     /* Signal all vCPUs waiting on this condition */
1677     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1678 
1679     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1680 }
1681 
1682 static void spapr_create_nvram(SpaprMachineState *spapr)
1683 {
1684     DeviceState *dev = qdev_new("spapr-nvram");
1685     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1686 
1687     if (dinfo) {
1688         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1689                                 &error_fatal);
1690     }
1691 
1692     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1693 
1694     spapr->nvram = (struct SpaprNvram *)dev;
1695 }
1696 
1697 static void spapr_rtc_create(SpaprMachineState *spapr)
1698 {
1699     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1700                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1701                                        &error_fatal, NULL);
1702     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1703     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1704                               "date");
1705 }
1706 
1707 /* Returns whether we want to use VGA or not */
1708 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1709 {
1710     switch (vga_interface_type) {
1711     case VGA_NONE:
1712         return false;
1713     case VGA_DEVICE:
1714         return true;
1715     case VGA_STD:
1716     case VGA_VIRTIO:
1717     case VGA_CIRRUS:
1718         return pci_vga_init(pci_bus) != NULL;
1719     default:
1720         error_setg(errp,
1721                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1722         return false;
1723     }
1724 }
1725 
1726 static int spapr_pre_load(void *opaque)
1727 {
1728     int rc;
1729 
1730     rc = spapr_caps_pre_load(opaque);
1731     if (rc) {
1732         return rc;
1733     }
1734 
1735     return 0;
1736 }
1737 
1738 static int spapr_post_load(void *opaque, int version_id)
1739 {
1740     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1741     int err = 0;
1742 
1743     err = spapr_caps_post_migration(spapr);
1744     if (err) {
1745         return err;
1746     }
1747 
1748     /*
1749      * In earlier versions, there was no separate qdev for the PAPR
1750      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1751      * So when migrating from those versions, poke the incoming offset
1752      * value into the RTC device
1753      */
1754     if (version_id < 3) {
1755         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1756         if (err) {
1757             return err;
1758         }
1759     }
1760 
1761     if (kvm_enabled() && spapr->patb_entry) {
1762         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1763         bool radix = !!(spapr->patb_entry & PATE1_GR);
1764         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1765 
1766         /*
1767          * Update LPCR:HR and UPRT as they may not be set properly in
1768          * the stream
1769          */
1770         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1771                             LPCR_HR | LPCR_UPRT);
1772 
1773         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1774         if (err) {
1775             error_report("Process table config unsupported by the host");
1776             return -EINVAL;
1777         }
1778     }
1779 
1780     err = spapr_irq_post_load(spapr, version_id);
1781     if (err) {
1782         return err;
1783     }
1784 
1785     return err;
1786 }
1787 
1788 static int spapr_pre_save(void *opaque)
1789 {
1790     int rc;
1791 
1792     rc = spapr_caps_pre_save(opaque);
1793     if (rc) {
1794         return rc;
1795     }
1796 
1797     return 0;
1798 }
1799 
1800 static bool version_before_3(void *opaque, int version_id)
1801 {
1802     return version_id < 3;
1803 }
1804 
1805 static bool spapr_pending_events_needed(void *opaque)
1806 {
1807     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1808     return !QTAILQ_EMPTY(&spapr->pending_events);
1809 }
1810 
1811 static const VMStateDescription vmstate_spapr_event_entry = {
1812     .name = "spapr_event_log_entry",
1813     .version_id = 1,
1814     .minimum_version_id = 1,
1815     .fields = (VMStateField[]) {
1816         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1817         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1818         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1819                                      NULL, extended_length),
1820         VMSTATE_END_OF_LIST()
1821     },
1822 };
1823 
1824 static const VMStateDescription vmstate_spapr_pending_events = {
1825     .name = "spapr_pending_events",
1826     .version_id = 1,
1827     .minimum_version_id = 1,
1828     .needed = spapr_pending_events_needed,
1829     .fields = (VMStateField[]) {
1830         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1831                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1832         VMSTATE_END_OF_LIST()
1833     },
1834 };
1835 
1836 static bool spapr_ov5_cas_needed(void *opaque)
1837 {
1838     SpaprMachineState *spapr = opaque;
1839     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1840     bool cas_needed;
1841 
1842     /* Prior to the introduction of SpaprOptionVector, we had two option
1843      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1844      * Both of these options encode machine topology into the device-tree
1845      * in such a way that the now-booted OS should still be able to interact
1846      * appropriately with QEMU regardless of what options were actually
1847      * negotiatied on the source side.
1848      *
1849      * As such, we can avoid migrating the CAS-negotiated options if these
1850      * are the only options available on the current machine/platform.
1851      * Since these are the only options available for pseries-2.7 and
1852      * earlier, this allows us to maintain old->new/new->old migration
1853      * compatibility.
1854      *
1855      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1856      * via default pseries-2.8 machines and explicit command-line parameters.
1857      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1858      * of the actual CAS-negotiated values to continue working properly. For
1859      * example, availability of memory unplug depends on knowing whether
1860      * OV5_HP_EVT was negotiated via CAS.
1861      *
1862      * Thus, for any cases where the set of available CAS-negotiatable
1863      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1864      * include the CAS-negotiated options in the migration stream, unless
1865      * if they affect boot time behaviour only.
1866      */
1867     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1868     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1869     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1870 
1871     /* We need extra information if we have any bits outside the mask
1872      * defined above */
1873     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1874 
1875     spapr_ovec_cleanup(ov5_mask);
1876 
1877     return cas_needed;
1878 }
1879 
1880 static const VMStateDescription vmstate_spapr_ov5_cas = {
1881     .name = "spapr_option_vector_ov5_cas",
1882     .version_id = 1,
1883     .minimum_version_id = 1,
1884     .needed = spapr_ov5_cas_needed,
1885     .fields = (VMStateField[]) {
1886         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1887                                  vmstate_spapr_ovec, SpaprOptionVector),
1888         VMSTATE_END_OF_LIST()
1889     },
1890 };
1891 
1892 static bool spapr_patb_entry_needed(void *opaque)
1893 {
1894     SpaprMachineState *spapr = opaque;
1895 
1896     return !!spapr->patb_entry;
1897 }
1898 
1899 static const VMStateDescription vmstate_spapr_patb_entry = {
1900     .name = "spapr_patb_entry",
1901     .version_id = 1,
1902     .minimum_version_id = 1,
1903     .needed = spapr_patb_entry_needed,
1904     .fields = (VMStateField[]) {
1905         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1906         VMSTATE_END_OF_LIST()
1907     },
1908 };
1909 
1910 static bool spapr_irq_map_needed(void *opaque)
1911 {
1912     SpaprMachineState *spapr = opaque;
1913 
1914     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1915 }
1916 
1917 static const VMStateDescription vmstate_spapr_irq_map = {
1918     .name = "spapr_irq_map",
1919     .version_id = 1,
1920     .minimum_version_id = 1,
1921     .needed = spapr_irq_map_needed,
1922     .fields = (VMStateField[]) {
1923         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1924         VMSTATE_END_OF_LIST()
1925     },
1926 };
1927 
1928 static bool spapr_dtb_needed(void *opaque)
1929 {
1930     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1931 
1932     return smc->update_dt_enabled;
1933 }
1934 
1935 static int spapr_dtb_pre_load(void *opaque)
1936 {
1937     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1938 
1939     g_free(spapr->fdt_blob);
1940     spapr->fdt_blob = NULL;
1941     spapr->fdt_size = 0;
1942 
1943     return 0;
1944 }
1945 
1946 static const VMStateDescription vmstate_spapr_dtb = {
1947     .name = "spapr_dtb",
1948     .version_id = 1,
1949     .minimum_version_id = 1,
1950     .needed = spapr_dtb_needed,
1951     .pre_load = spapr_dtb_pre_load,
1952     .fields = (VMStateField[]) {
1953         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1954         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1955         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1956                                      fdt_size),
1957         VMSTATE_END_OF_LIST()
1958     },
1959 };
1960 
1961 static bool spapr_fwnmi_needed(void *opaque)
1962 {
1963     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1964 
1965     return spapr->fwnmi_machine_check_addr != -1;
1966 }
1967 
1968 static int spapr_fwnmi_pre_save(void *opaque)
1969 {
1970     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1971 
1972     /*
1973      * Check if machine check handling is in progress and print a
1974      * warning message.
1975      */
1976     if (spapr->fwnmi_machine_check_interlock != -1) {
1977         warn_report("A machine check is being handled during migration. The"
1978                 "handler may run and log hardware error on the destination");
1979     }
1980 
1981     return 0;
1982 }
1983 
1984 static const VMStateDescription vmstate_spapr_fwnmi = {
1985     .name = "spapr_fwnmi",
1986     .version_id = 1,
1987     .minimum_version_id = 1,
1988     .needed = spapr_fwnmi_needed,
1989     .pre_save = spapr_fwnmi_pre_save,
1990     .fields = (VMStateField[]) {
1991         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1992         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1993         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1994         VMSTATE_END_OF_LIST()
1995     },
1996 };
1997 
1998 static const VMStateDescription vmstate_spapr = {
1999     .name = "spapr",
2000     .version_id = 3,
2001     .minimum_version_id = 1,
2002     .pre_load = spapr_pre_load,
2003     .post_load = spapr_post_load,
2004     .pre_save = spapr_pre_save,
2005     .fields = (VMStateField[]) {
2006         /* used to be @next_irq */
2007         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2008 
2009         /* RTC offset */
2010         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2011 
2012         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2013         VMSTATE_END_OF_LIST()
2014     },
2015     .subsections = (const VMStateDescription*[]) {
2016         &vmstate_spapr_ov5_cas,
2017         &vmstate_spapr_patb_entry,
2018         &vmstate_spapr_pending_events,
2019         &vmstate_spapr_cap_htm,
2020         &vmstate_spapr_cap_vsx,
2021         &vmstate_spapr_cap_dfp,
2022         &vmstate_spapr_cap_cfpc,
2023         &vmstate_spapr_cap_sbbc,
2024         &vmstate_spapr_cap_ibs,
2025         &vmstate_spapr_cap_hpt_maxpagesize,
2026         &vmstate_spapr_irq_map,
2027         &vmstate_spapr_cap_nested_kvm_hv,
2028         &vmstate_spapr_dtb,
2029         &vmstate_spapr_cap_large_decr,
2030         &vmstate_spapr_cap_ccf_assist,
2031         &vmstate_spapr_cap_fwnmi,
2032         &vmstate_spapr_fwnmi,
2033         NULL
2034     }
2035 };
2036 
2037 static int htab_save_setup(QEMUFile *f, void *opaque)
2038 {
2039     SpaprMachineState *spapr = opaque;
2040 
2041     /* "Iteration" header */
2042     if (!spapr->htab_shift) {
2043         qemu_put_be32(f, -1);
2044     } else {
2045         qemu_put_be32(f, spapr->htab_shift);
2046     }
2047 
2048     if (spapr->htab) {
2049         spapr->htab_save_index = 0;
2050         spapr->htab_first_pass = true;
2051     } else {
2052         if (spapr->htab_shift) {
2053             assert(kvm_enabled());
2054         }
2055     }
2056 
2057 
2058     return 0;
2059 }
2060 
2061 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2062                             int chunkstart, int n_valid, int n_invalid)
2063 {
2064     qemu_put_be32(f, chunkstart);
2065     qemu_put_be16(f, n_valid);
2066     qemu_put_be16(f, n_invalid);
2067     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2068                     HASH_PTE_SIZE_64 * n_valid);
2069 }
2070 
2071 static void htab_save_end_marker(QEMUFile *f)
2072 {
2073     qemu_put_be32(f, 0);
2074     qemu_put_be16(f, 0);
2075     qemu_put_be16(f, 0);
2076 }
2077 
2078 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2079                                  int64_t max_ns)
2080 {
2081     bool has_timeout = max_ns != -1;
2082     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2083     int index = spapr->htab_save_index;
2084     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2085 
2086     assert(spapr->htab_first_pass);
2087 
2088     do {
2089         int chunkstart;
2090 
2091         /* Consume invalid HPTEs */
2092         while ((index < htabslots)
2093                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2094             CLEAN_HPTE(HPTE(spapr->htab, index));
2095             index++;
2096         }
2097 
2098         /* Consume valid HPTEs */
2099         chunkstart = index;
2100         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2101                && HPTE_VALID(HPTE(spapr->htab, index))) {
2102             CLEAN_HPTE(HPTE(spapr->htab, index));
2103             index++;
2104         }
2105 
2106         if (index > chunkstart) {
2107             int n_valid = index - chunkstart;
2108 
2109             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2110 
2111             if (has_timeout &&
2112                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2113                 break;
2114             }
2115         }
2116     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2117 
2118     if (index >= htabslots) {
2119         assert(index == htabslots);
2120         index = 0;
2121         spapr->htab_first_pass = false;
2122     }
2123     spapr->htab_save_index = index;
2124 }
2125 
2126 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2127                                 int64_t max_ns)
2128 {
2129     bool final = max_ns < 0;
2130     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2131     int examined = 0, sent = 0;
2132     int index = spapr->htab_save_index;
2133     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2134 
2135     assert(!spapr->htab_first_pass);
2136 
2137     do {
2138         int chunkstart, invalidstart;
2139 
2140         /* Consume non-dirty HPTEs */
2141         while ((index < htabslots)
2142                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2143             index++;
2144             examined++;
2145         }
2146 
2147         chunkstart = index;
2148         /* Consume valid dirty HPTEs */
2149         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2150                && HPTE_DIRTY(HPTE(spapr->htab, index))
2151                && HPTE_VALID(HPTE(spapr->htab, index))) {
2152             CLEAN_HPTE(HPTE(spapr->htab, index));
2153             index++;
2154             examined++;
2155         }
2156 
2157         invalidstart = index;
2158         /* Consume invalid dirty HPTEs */
2159         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2160                && HPTE_DIRTY(HPTE(spapr->htab, index))
2161                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2162             CLEAN_HPTE(HPTE(spapr->htab, index));
2163             index++;
2164             examined++;
2165         }
2166 
2167         if (index > chunkstart) {
2168             int n_valid = invalidstart - chunkstart;
2169             int n_invalid = index - invalidstart;
2170 
2171             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2172             sent += index - chunkstart;
2173 
2174             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2175                 break;
2176             }
2177         }
2178 
2179         if (examined >= htabslots) {
2180             break;
2181         }
2182 
2183         if (index >= htabslots) {
2184             assert(index == htabslots);
2185             index = 0;
2186         }
2187     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2188 
2189     if (index >= htabslots) {
2190         assert(index == htabslots);
2191         index = 0;
2192     }
2193 
2194     spapr->htab_save_index = index;
2195 
2196     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2197 }
2198 
2199 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2200 #define MAX_KVM_BUF_SIZE    2048
2201 
2202 static int htab_save_iterate(QEMUFile *f, void *opaque)
2203 {
2204     SpaprMachineState *spapr = opaque;
2205     int fd;
2206     int rc = 0;
2207 
2208     /* Iteration header */
2209     if (!spapr->htab_shift) {
2210         qemu_put_be32(f, -1);
2211         return 1;
2212     } else {
2213         qemu_put_be32(f, 0);
2214     }
2215 
2216     if (!spapr->htab) {
2217         assert(kvm_enabled());
2218 
2219         fd = get_htab_fd(spapr);
2220         if (fd < 0) {
2221             return fd;
2222         }
2223 
2224         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2225         if (rc < 0) {
2226             return rc;
2227         }
2228     } else  if (spapr->htab_first_pass) {
2229         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2230     } else {
2231         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2232     }
2233 
2234     htab_save_end_marker(f);
2235 
2236     return rc;
2237 }
2238 
2239 static int htab_save_complete(QEMUFile *f, void *opaque)
2240 {
2241     SpaprMachineState *spapr = opaque;
2242     int fd;
2243 
2244     /* Iteration header */
2245     if (!spapr->htab_shift) {
2246         qemu_put_be32(f, -1);
2247         return 0;
2248     } else {
2249         qemu_put_be32(f, 0);
2250     }
2251 
2252     if (!spapr->htab) {
2253         int rc;
2254 
2255         assert(kvm_enabled());
2256 
2257         fd = get_htab_fd(spapr);
2258         if (fd < 0) {
2259             return fd;
2260         }
2261 
2262         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2263         if (rc < 0) {
2264             return rc;
2265         }
2266     } else {
2267         if (spapr->htab_first_pass) {
2268             htab_save_first_pass(f, spapr, -1);
2269         }
2270         htab_save_later_pass(f, spapr, -1);
2271     }
2272 
2273     /* End marker */
2274     htab_save_end_marker(f);
2275 
2276     return 0;
2277 }
2278 
2279 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2280 {
2281     SpaprMachineState *spapr = opaque;
2282     uint32_t section_hdr;
2283     int fd = -1;
2284     Error *local_err = NULL;
2285 
2286     if (version_id < 1 || version_id > 1) {
2287         error_report("htab_load() bad version");
2288         return -EINVAL;
2289     }
2290 
2291     section_hdr = qemu_get_be32(f);
2292 
2293     if (section_hdr == -1) {
2294         spapr_free_hpt(spapr);
2295         return 0;
2296     }
2297 
2298     if (section_hdr) {
2299         /* First section gives the htab size */
2300         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2301         if (local_err) {
2302             error_report_err(local_err);
2303             return -EINVAL;
2304         }
2305         return 0;
2306     }
2307 
2308     if (!spapr->htab) {
2309         assert(kvm_enabled());
2310 
2311         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2312         if (fd < 0) {
2313             error_report_err(local_err);
2314             return fd;
2315         }
2316     }
2317 
2318     while (true) {
2319         uint32_t index;
2320         uint16_t n_valid, n_invalid;
2321 
2322         index = qemu_get_be32(f);
2323         n_valid = qemu_get_be16(f);
2324         n_invalid = qemu_get_be16(f);
2325 
2326         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2327             /* End of Stream */
2328             break;
2329         }
2330 
2331         if ((index + n_valid + n_invalid) >
2332             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2333             /* Bad index in stream */
2334             error_report(
2335                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2336                 index, n_valid, n_invalid, spapr->htab_shift);
2337             return -EINVAL;
2338         }
2339 
2340         if (spapr->htab) {
2341             if (n_valid) {
2342                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2343                                 HASH_PTE_SIZE_64 * n_valid);
2344             }
2345             if (n_invalid) {
2346                 memset(HPTE(spapr->htab, index + n_valid), 0,
2347                        HASH_PTE_SIZE_64 * n_invalid);
2348             }
2349         } else {
2350             int rc;
2351 
2352             assert(fd >= 0);
2353 
2354             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2355             if (rc < 0) {
2356                 return rc;
2357             }
2358         }
2359     }
2360 
2361     if (!spapr->htab) {
2362         assert(fd >= 0);
2363         close(fd);
2364     }
2365 
2366     return 0;
2367 }
2368 
2369 static void htab_save_cleanup(void *opaque)
2370 {
2371     SpaprMachineState *spapr = opaque;
2372 
2373     close_htab_fd(spapr);
2374 }
2375 
2376 static SaveVMHandlers savevm_htab_handlers = {
2377     .save_setup = htab_save_setup,
2378     .save_live_iterate = htab_save_iterate,
2379     .save_live_complete_precopy = htab_save_complete,
2380     .save_cleanup = htab_save_cleanup,
2381     .load_state = htab_load,
2382 };
2383 
2384 static void spapr_boot_set(void *opaque, const char *boot_device,
2385                            Error **errp)
2386 {
2387     MachineState *machine = MACHINE(opaque);
2388     machine->boot_order = g_strdup(boot_device);
2389 }
2390 
2391 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2392 {
2393     MachineState *machine = MACHINE(spapr);
2394     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2395     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2396     int i;
2397 
2398     for (i = 0; i < nr_lmbs; i++) {
2399         uint64_t addr;
2400 
2401         addr = i * lmb_size + machine->device_memory->base;
2402         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2403                                addr / lmb_size);
2404     }
2405 }
2406 
2407 /*
2408  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2409  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2410  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2411  */
2412 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2413 {
2414     int i;
2415 
2416     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2417         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2418                    " is not aligned to %" PRIu64 " MiB",
2419                    machine->ram_size,
2420                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2421         return;
2422     }
2423 
2424     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2425         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2426                    " is not aligned to %" PRIu64 " MiB",
2427                    machine->ram_size,
2428                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2429         return;
2430     }
2431 
2432     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2433         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2434             error_setg(errp,
2435                        "Node %d memory size 0x%" PRIx64
2436                        " is not aligned to %" PRIu64 " MiB",
2437                        i, machine->numa_state->nodes[i].node_mem,
2438                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2439             return;
2440         }
2441     }
2442 }
2443 
2444 /* find cpu slot in machine->possible_cpus by core_id */
2445 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2446 {
2447     int index = id / ms->smp.threads;
2448 
2449     if (index >= ms->possible_cpus->len) {
2450         return NULL;
2451     }
2452     if (idx) {
2453         *idx = index;
2454     }
2455     return &ms->possible_cpus->cpus[index];
2456 }
2457 
2458 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2459 {
2460     MachineState *ms = MACHINE(spapr);
2461     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2462     Error *local_err = NULL;
2463     bool vsmt_user = !!spapr->vsmt;
2464     int kvm_smt = kvmppc_smt_threads();
2465     int ret;
2466     unsigned int smp_threads = ms->smp.threads;
2467 
2468     if (!kvm_enabled() && (smp_threads > 1)) {
2469         error_setg(errp, "TCG cannot support more than 1 thread/core "
2470                    "on a pseries machine");
2471         return;
2472     }
2473     if (!is_power_of_2(smp_threads)) {
2474         error_setg(errp, "Cannot support %d threads/core on a pseries "
2475                    "machine because it must be a power of 2", smp_threads);
2476         return;
2477     }
2478 
2479     /* Detemine the VSMT mode to use: */
2480     if (vsmt_user) {
2481         if (spapr->vsmt < smp_threads) {
2482             error_setg(errp, "Cannot support VSMT mode %d"
2483                        " because it must be >= threads/core (%d)",
2484                        spapr->vsmt, smp_threads);
2485             return;
2486         }
2487         /* In this case, spapr->vsmt has been set by the command line */
2488     } else if (!smc->smp_threads_vsmt) {
2489         /*
2490          * Default VSMT value is tricky, because we need it to be as
2491          * consistent as possible (for migration), but this requires
2492          * changing it for at least some existing cases.  We pick 8 as
2493          * the value that we'd get with KVM on POWER8, the
2494          * overwhelmingly common case in production systems.
2495          */
2496         spapr->vsmt = MAX(8, smp_threads);
2497     } else {
2498         spapr->vsmt = smp_threads;
2499     }
2500 
2501     /* KVM: If necessary, set the SMT mode: */
2502     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2503         ret = kvmppc_set_smt_threads(spapr->vsmt);
2504         if (ret) {
2505             /* Looks like KVM isn't able to change VSMT mode */
2506             error_setg(&local_err,
2507                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2508                        spapr->vsmt, ret);
2509             /* We can live with that if the default one is big enough
2510              * for the number of threads, and a submultiple of the one
2511              * we want.  In this case we'll waste some vcpu ids, but
2512              * behaviour will be correct */
2513             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2514                 warn_report_err(local_err);
2515             } else {
2516                 if (!vsmt_user) {
2517                     error_append_hint(&local_err,
2518                                       "On PPC, a VM with %d threads/core"
2519                                       " on a host with %d threads/core"
2520                                       " requires the use of VSMT mode %d.\n",
2521                                       smp_threads, kvm_smt, spapr->vsmt);
2522                 }
2523                 kvmppc_error_append_smt_possible_hint(&local_err);
2524                 error_propagate(errp, local_err);
2525             }
2526         }
2527     }
2528     /* else TCG: nothing to do currently */
2529 }
2530 
2531 static void spapr_init_cpus(SpaprMachineState *spapr)
2532 {
2533     MachineState *machine = MACHINE(spapr);
2534     MachineClass *mc = MACHINE_GET_CLASS(machine);
2535     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2536     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2537     const CPUArchIdList *possible_cpus;
2538     unsigned int smp_cpus = machine->smp.cpus;
2539     unsigned int smp_threads = machine->smp.threads;
2540     unsigned int max_cpus = machine->smp.max_cpus;
2541     int boot_cores_nr = smp_cpus / smp_threads;
2542     int i;
2543 
2544     possible_cpus = mc->possible_cpu_arch_ids(machine);
2545     if (mc->has_hotpluggable_cpus) {
2546         if (smp_cpus % smp_threads) {
2547             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2548                          smp_cpus, smp_threads);
2549             exit(1);
2550         }
2551         if (max_cpus % smp_threads) {
2552             error_report("max_cpus (%u) must be multiple of threads (%u)",
2553                          max_cpus, smp_threads);
2554             exit(1);
2555         }
2556     } else {
2557         if (max_cpus != smp_cpus) {
2558             error_report("This machine version does not support CPU hotplug");
2559             exit(1);
2560         }
2561         boot_cores_nr = possible_cpus->len;
2562     }
2563 
2564     if (smc->pre_2_10_has_unused_icps) {
2565         int i;
2566 
2567         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2568             /* Dummy entries get deregistered when real ICPState objects
2569              * are registered during CPU core hotplug.
2570              */
2571             pre_2_10_vmstate_register_dummy_icp(i);
2572         }
2573     }
2574 
2575     for (i = 0; i < possible_cpus->len; i++) {
2576         int core_id = i * smp_threads;
2577 
2578         if (mc->has_hotpluggable_cpus) {
2579             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2580                                    spapr_vcpu_id(spapr, core_id));
2581         }
2582 
2583         if (i < boot_cores_nr) {
2584             Object *core  = object_new(type);
2585             int nr_threads = smp_threads;
2586 
2587             /* Handle the partially filled core for older machine types */
2588             if ((i + 1) * smp_threads >= smp_cpus) {
2589                 nr_threads = smp_cpus - i * smp_threads;
2590             }
2591 
2592             object_property_set_int(core, "nr-threads", nr_threads,
2593                                     &error_fatal);
2594             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2595                                     &error_fatal);
2596             qdev_realize(DEVICE(core), NULL, &error_fatal);
2597 
2598             object_unref(core);
2599         }
2600     }
2601 }
2602 
2603 static PCIHostState *spapr_create_default_phb(void)
2604 {
2605     DeviceState *dev;
2606 
2607     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2608     qdev_prop_set_uint32(dev, "index", 0);
2609     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2610 
2611     return PCI_HOST_BRIDGE(dev);
2612 }
2613 
2614 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2615 {
2616     MachineState *machine = MACHINE(spapr);
2617     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2618     hwaddr rma_size = machine->ram_size;
2619     hwaddr node0_size = spapr_node0_size(machine);
2620 
2621     /* RMA has to fit in the first NUMA node */
2622     rma_size = MIN(rma_size, node0_size);
2623 
2624     /*
2625      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2626      * never exceed that
2627      */
2628     rma_size = MIN(rma_size, 1 * TiB);
2629 
2630     /*
2631      * Clamp the RMA size based on machine type.  This is for
2632      * migration compatibility with older qemu versions, which limited
2633      * the RMA size for complicated and mostly bad reasons.
2634      */
2635     if (smc->rma_limit) {
2636         rma_size = MIN(rma_size, smc->rma_limit);
2637     }
2638 
2639     if (rma_size < MIN_RMA_SLOF) {
2640         error_setg(errp,
2641                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2642                    "ldMiB guest RMA (Real Mode Area memory)",
2643                    MIN_RMA_SLOF / MiB);
2644         return 0;
2645     }
2646 
2647     return rma_size;
2648 }
2649 
2650 /* pSeries LPAR / sPAPR hardware init */
2651 static void spapr_machine_init(MachineState *machine)
2652 {
2653     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2654     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2655     MachineClass *mc = MACHINE_GET_CLASS(machine);
2656     const char *kernel_filename = machine->kernel_filename;
2657     const char *initrd_filename = machine->initrd_filename;
2658     PCIHostState *phb;
2659     int i;
2660     MemoryRegion *sysmem = get_system_memory();
2661     long load_limit, fw_size;
2662     char *filename;
2663     Error *resize_hpt_err = NULL;
2664 
2665     msi_nonbroken = true;
2666 
2667     QLIST_INIT(&spapr->phbs);
2668     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2669 
2670     /* Determine capabilities to run with */
2671     spapr_caps_init(spapr);
2672 
2673     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2674     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2675         /*
2676          * If the user explicitly requested a mode we should either
2677          * supply it, or fail completely (which we do below).  But if
2678          * it's not set explicitly, we reset our mode to something
2679          * that works
2680          */
2681         if (resize_hpt_err) {
2682             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2683             error_free(resize_hpt_err);
2684             resize_hpt_err = NULL;
2685         } else {
2686             spapr->resize_hpt = smc->resize_hpt_default;
2687         }
2688     }
2689 
2690     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2691 
2692     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2693         /*
2694          * User requested HPT resize, but this host can't supply it.  Bail out
2695          */
2696         error_report_err(resize_hpt_err);
2697         exit(1);
2698     }
2699     error_free(resize_hpt_err);
2700 
2701     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2702 
2703     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2704     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2705 
2706     /*
2707      * VSMT must be set in order to be able to compute VCPU ids, ie to
2708      * call spapr_max_server_number() or spapr_vcpu_id().
2709      */
2710     spapr_set_vsmt_mode(spapr, &error_fatal);
2711 
2712     /* Set up Interrupt Controller before we create the VCPUs */
2713     spapr_irq_init(spapr, &error_fatal);
2714 
2715     /* Set up containers for ibm,client-architecture-support negotiated options
2716      */
2717     spapr->ov5 = spapr_ovec_new();
2718     spapr->ov5_cas = spapr_ovec_new();
2719 
2720     if (smc->dr_lmb_enabled) {
2721         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2722         spapr_validate_node_memory(machine, &error_fatal);
2723     }
2724 
2725     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2726 
2727     /* advertise support for dedicated HP event source to guests */
2728     if (spapr->use_hotplug_event_source) {
2729         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2730     }
2731 
2732     /* advertise support for HPT resizing */
2733     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2734         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2735     }
2736 
2737     /* advertise support for ibm,dyamic-memory-v2 */
2738     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2739 
2740     /* advertise XIVE on POWER9 machines */
2741     if (spapr->irq->xive) {
2742         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2743     }
2744 
2745     /* init CPUs */
2746     spapr_init_cpus(spapr);
2747 
2748     /*
2749      * check we don't have a memory-less/cpu-less NUMA node
2750      * Firmware relies on the existing memory/cpu topology to provide the
2751      * NUMA topology to the kernel.
2752      * And the linux kernel needs to know the NUMA topology at start
2753      * to be able to hotplug CPUs later.
2754      */
2755     if (machine->numa_state->num_nodes) {
2756         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2757             /* check for memory-less node */
2758             if (machine->numa_state->nodes[i].node_mem == 0) {
2759                 CPUState *cs;
2760                 int found = 0;
2761                 /* check for cpu-less node */
2762                 CPU_FOREACH(cs) {
2763                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2764                     if (cpu->node_id == i) {
2765                         found = 1;
2766                         break;
2767                     }
2768                 }
2769                 /* memory-less and cpu-less node */
2770                 if (!found) {
2771                     error_report(
2772                        "Memory-less/cpu-less nodes are not supported (node %d)",
2773                                  i);
2774                     exit(1);
2775                 }
2776             }
2777         }
2778 
2779     }
2780 
2781     /*
2782      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2783      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2784      * called from vPHB reset handler so we initialize the counter here.
2785      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2786      * must be equally distant from any other node.
2787      * The final value of spapr->gpu_numa_id is going to be written to
2788      * max-associativity-domains in spapr_build_fdt().
2789      */
2790     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2791 
2792     /* Init numa_assoc_array */
2793     spapr_numa_associativity_init(spapr, machine);
2794 
2795     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2796         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2797                               spapr->max_compat_pvr)) {
2798         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2799         /* KVM and TCG always allow GTSE with radix... */
2800         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2801     }
2802     /* ... but not with hash (currently). */
2803 
2804     if (kvm_enabled()) {
2805         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2806         kvmppc_enable_logical_ci_hcalls();
2807         kvmppc_enable_set_mode_hcall();
2808 
2809         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2810         kvmppc_enable_clear_ref_mod_hcalls();
2811 
2812         /* Enable H_PAGE_INIT */
2813         kvmppc_enable_h_page_init();
2814     }
2815 
2816     /* map RAM */
2817     memory_region_add_subregion(sysmem, 0, machine->ram);
2818 
2819     /* always allocate the device memory information */
2820     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2821 
2822     /* initialize hotplug memory address space */
2823     if (machine->ram_size < machine->maxram_size) {
2824         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2825         /*
2826          * Limit the number of hotpluggable memory slots to half the number
2827          * slots that KVM supports, leaving the other half for PCI and other
2828          * devices. However ensure that number of slots doesn't drop below 32.
2829          */
2830         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2831                            SPAPR_MAX_RAM_SLOTS;
2832 
2833         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2834             max_memslots = SPAPR_MAX_RAM_SLOTS;
2835         }
2836         if (machine->ram_slots > max_memslots) {
2837             error_report("Specified number of memory slots %"
2838                          PRIu64" exceeds max supported %d",
2839                          machine->ram_slots, max_memslots);
2840             exit(1);
2841         }
2842 
2843         machine->device_memory->base = ROUND_UP(machine->ram_size,
2844                                                 SPAPR_DEVICE_MEM_ALIGN);
2845         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2846                            "device-memory", device_mem_size);
2847         memory_region_add_subregion(sysmem, machine->device_memory->base,
2848                                     &machine->device_memory->mr);
2849     }
2850 
2851     if (smc->dr_lmb_enabled) {
2852         spapr_create_lmb_dr_connectors(spapr);
2853     }
2854 
2855     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2856         /* Create the error string for live migration blocker */
2857         error_setg(&spapr->fwnmi_migration_blocker,
2858             "A machine check is being handled during migration. The handler"
2859             "may run and log hardware error on the destination");
2860     }
2861 
2862     if (mc->nvdimm_supported) {
2863         spapr_create_nvdimm_dr_connectors(spapr);
2864     }
2865 
2866     /* Set up RTAS event infrastructure */
2867     spapr_events_init(spapr);
2868 
2869     /* Set up the RTC RTAS interfaces */
2870     spapr_rtc_create(spapr);
2871 
2872     /* Set up VIO bus */
2873     spapr->vio_bus = spapr_vio_bus_init();
2874 
2875     for (i = 0; i < serial_max_hds(); i++) {
2876         if (serial_hd(i)) {
2877             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2878         }
2879     }
2880 
2881     /* We always have at least the nvram device on VIO */
2882     spapr_create_nvram(spapr);
2883 
2884     /*
2885      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2886      * connectors (described in root DT node's "ibm,drc-types" property)
2887      * are pre-initialized here. additional child connectors (such as
2888      * connectors for a PHBs PCI slots) are added as needed during their
2889      * parent's realization.
2890      */
2891     if (smc->dr_phb_enabled) {
2892         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2893             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2894         }
2895     }
2896 
2897     /* Set up PCI */
2898     spapr_pci_rtas_init();
2899 
2900     phb = spapr_create_default_phb();
2901 
2902     for (i = 0; i < nb_nics; i++) {
2903         NICInfo *nd = &nd_table[i];
2904 
2905         if (!nd->model) {
2906             nd->model = g_strdup("spapr-vlan");
2907         }
2908 
2909         if (g_str_equal(nd->model, "spapr-vlan") ||
2910             g_str_equal(nd->model, "ibmveth")) {
2911             spapr_vlan_create(spapr->vio_bus, nd);
2912         } else {
2913             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2914         }
2915     }
2916 
2917     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2918         spapr_vscsi_create(spapr->vio_bus);
2919     }
2920 
2921     /* Graphics */
2922     if (spapr_vga_init(phb->bus, &error_fatal)) {
2923         spapr->has_graphics = true;
2924         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2925     }
2926 
2927     if (machine->usb) {
2928         if (smc->use_ohci_by_default) {
2929             pci_create_simple(phb->bus, -1, "pci-ohci");
2930         } else {
2931             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2932         }
2933 
2934         if (spapr->has_graphics) {
2935             USBBus *usb_bus = usb_bus_find(-1);
2936 
2937             usb_create_simple(usb_bus, "usb-kbd");
2938             usb_create_simple(usb_bus, "usb-mouse");
2939         }
2940     }
2941 
2942     if (kernel_filename) {
2943         uint64_t lowaddr = 0;
2944 
2945         spapr->kernel_size = load_elf(kernel_filename, NULL,
2946                                       translate_kernel_address, spapr,
2947                                       NULL, &lowaddr, NULL, NULL, 1,
2948                                       PPC_ELF_MACHINE, 0, 0);
2949         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2950             spapr->kernel_size = load_elf(kernel_filename, NULL,
2951                                           translate_kernel_address, spapr, NULL,
2952                                           &lowaddr, NULL, NULL, 0,
2953                                           PPC_ELF_MACHINE,
2954                                           0, 0);
2955             spapr->kernel_le = spapr->kernel_size > 0;
2956         }
2957         if (spapr->kernel_size < 0) {
2958             error_report("error loading %s: %s", kernel_filename,
2959                          load_elf_strerror(spapr->kernel_size));
2960             exit(1);
2961         }
2962 
2963         /* load initrd */
2964         if (initrd_filename) {
2965             /* Try to locate the initrd in the gap between the kernel
2966              * and the firmware. Add a bit of space just in case
2967              */
2968             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2969                                   + 0x1ffff) & ~0xffff;
2970             spapr->initrd_size = load_image_targphys(initrd_filename,
2971                                                      spapr->initrd_base,
2972                                                      load_limit
2973                                                      - spapr->initrd_base);
2974             if (spapr->initrd_size < 0) {
2975                 error_report("could not load initial ram disk '%s'",
2976                              initrd_filename);
2977                 exit(1);
2978             }
2979         }
2980     }
2981 
2982     if (bios_name == NULL) {
2983         bios_name = FW_FILE_NAME;
2984     }
2985     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2986     if (!filename) {
2987         error_report("Could not find LPAR firmware '%s'", bios_name);
2988         exit(1);
2989     }
2990     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2991     if (fw_size <= 0) {
2992         error_report("Could not load LPAR firmware '%s'", filename);
2993         exit(1);
2994     }
2995     g_free(filename);
2996 
2997     /* FIXME: Should register things through the MachineState's qdev
2998      * interface, this is a legacy from the sPAPREnvironment structure
2999      * which predated MachineState but had a similar function */
3000     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3001     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3002                          &savevm_htab_handlers, spapr);
3003 
3004     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3005 
3006     qemu_register_boot_set(spapr_boot_set, spapr);
3007 
3008     /*
3009      * Nothing needs to be done to resume a suspended guest because
3010      * suspending does not change the machine state, so no need for
3011      * a ->wakeup method.
3012      */
3013     qemu_register_wakeup_support();
3014 
3015     if (kvm_enabled()) {
3016         /* to stop and start vmclock */
3017         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3018                                          &spapr->tb);
3019 
3020         kvmppc_spapr_enable_inkernel_multitce();
3021     }
3022 
3023     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3024 }
3025 
3026 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3027 {
3028     if (!vm_type) {
3029         return 0;
3030     }
3031 
3032     if (!strcmp(vm_type, "HV")) {
3033         return 1;
3034     }
3035 
3036     if (!strcmp(vm_type, "PR")) {
3037         return 2;
3038     }
3039 
3040     error_report("Unknown kvm-type specified '%s'", vm_type);
3041     exit(1);
3042 }
3043 
3044 /*
3045  * Implementation of an interface to adjust firmware path
3046  * for the bootindex property handling.
3047  */
3048 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3049                                    DeviceState *dev)
3050 {
3051 #define CAST(type, obj, name) \
3052     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3053     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3054     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3055     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3056 
3057     if (d) {
3058         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3059         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3060         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3061 
3062         if (spapr) {
3063             /*
3064              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3065              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3066              * 0x8000 | (target << 8) | (bus << 5) | lun
3067              * (see the "Logical unit addressing format" table in SAM5)
3068              */
3069             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3070             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3071                                    (uint64_t)id << 48);
3072         } else if (virtio) {
3073             /*
3074              * We use SRP luns of the form 01000000 | (target << 8) | lun
3075              * in the top 32 bits of the 64-bit LUN
3076              * Note: the quote above is from SLOF and it is wrong,
3077              * the actual binding is:
3078              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3079              */
3080             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3081             if (d->lun >= 256) {
3082                 /* Use the LUN "flat space addressing method" */
3083                 id |= 0x4000;
3084             }
3085             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3086                                    (uint64_t)id << 32);
3087         } else if (usb) {
3088             /*
3089              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3090              * in the top 32 bits of the 64-bit LUN
3091              */
3092             unsigned usb_port = atoi(usb->port->path);
3093             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3094             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3095                                    (uint64_t)id << 32);
3096         }
3097     }
3098 
3099     /*
3100      * SLOF probes the USB devices, and if it recognizes that the device is a
3101      * storage device, it changes its name to "storage" instead of "usb-host",
3102      * and additionally adds a child node for the SCSI LUN, so the correct
3103      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3104      */
3105     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3106         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3107         if (usb_host_dev_is_scsi_storage(usbdev)) {
3108             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3109         }
3110     }
3111 
3112     if (phb) {
3113         /* Replace "pci" with "pci@800000020000000" */
3114         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3115     }
3116 
3117     if (vsc) {
3118         /* Same logic as virtio above */
3119         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3120         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3121     }
3122 
3123     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3124         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3125         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3126         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3127     }
3128 
3129     return NULL;
3130 }
3131 
3132 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3133 {
3134     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3135 
3136     return g_strdup(spapr->kvm_type);
3137 }
3138 
3139 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3140 {
3141     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3142 
3143     g_free(spapr->kvm_type);
3144     spapr->kvm_type = g_strdup(value);
3145 }
3146 
3147 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3148 {
3149     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3150 
3151     return spapr->use_hotplug_event_source;
3152 }
3153 
3154 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3155                                             Error **errp)
3156 {
3157     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3158 
3159     spapr->use_hotplug_event_source = value;
3160 }
3161 
3162 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3163 {
3164     return true;
3165 }
3166 
3167 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3168 {
3169     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3170 
3171     switch (spapr->resize_hpt) {
3172     case SPAPR_RESIZE_HPT_DEFAULT:
3173         return g_strdup("default");
3174     case SPAPR_RESIZE_HPT_DISABLED:
3175         return g_strdup("disabled");
3176     case SPAPR_RESIZE_HPT_ENABLED:
3177         return g_strdup("enabled");
3178     case SPAPR_RESIZE_HPT_REQUIRED:
3179         return g_strdup("required");
3180     }
3181     g_assert_not_reached();
3182 }
3183 
3184 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3185 {
3186     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3187 
3188     if (strcmp(value, "default") == 0) {
3189         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3190     } else if (strcmp(value, "disabled") == 0) {
3191         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3192     } else if (strcmp(value, "enabled") == 0) {
3193         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3194     } else if (strcmp(value, "required") == 0) {
3195         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3196     } else {
3197         error_setg(errp, "Bad value for \"resize-hpt\" property");
3198     }
3199 }
3200 
3201 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3202 {
3203     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3204 
3205     if (spapr->irq == &spapr_irq_xics_legacy) {
3206         return g_strdup("legacy");
3207     } else if (spapr->irq == &spapr_irq_xics) {
3208         return g_strdup("xics");
3209     } else if (spapr->irq == &spapr_irq_xive) {
3210         return g_strdup("xive");
3211     } else if (spapr->irq == &spapr_irq_dual) {
3212         return g_strdup("dual");
3213     }
3214     g_assert_not_reached();
3215 }
3216 
3217 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3218 {
3219     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3220 
3221     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3222         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3223         return;
3224     }
3225 
3226     /* The legacy IRQ backend can not be set */
3227     if (strcmp(value, "xics") == 0) {
3228         spapr->irq = &spapr_irq_xics;
3229     } else if (strcmp(value, "xive") == 0) {
3230         spapr->irq = &spapr_irq_xive;
3231     } else if (strcmp(value, "dual") == 0) {
3232         spapr->irq = &spapr_irq_dual;
3233     } else {
3234         error_setg(errp, "Bad value for \"ic-mode\" property");
3235     }
3236 }
3237 
3238 static char *spapr_get_host_model(Object *obj, Error **errp)
3239 {
3240     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3241 
3242     return g_strdup(spapr->host_model);
3243 }
3244 
3245 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3246 {
3247     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3248 
3249     g_free(spapr->host_model);
3250     spapr->host_model = g_strdup(value);
3251 }
3252 
3253 static char *spapr_get_host_serial(Object *obj, Error **errp)
3254 {
3255     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3256 
3257     return g_strdup(spapr->host_serial);
3258 }
3259 
3260 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3261 {
3262     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3263 
3264     g_free(spapr->host_serial);
3265     spapr->host_serial = g_strdup(value);
3266 }
3267 
3268 static void spapr_instance_init(Object *obj)
3269 {
3270     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3271     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3272 
3273     spapr->htab_fd = -1;
3274     spapr->use_hotplug_event_source = true;
3275     object_property_add_str(obj, "kvm-type",
3276                             spapr_get_kvm_type, spapr_set_kvm_type);
3277     object_property_set_description(obj, "kvm-type",
3278                                     "Specifies the KVM virtualization mode (HV, PR)");
3279     object_property_add_bool(obj, "modern-hotplug-events",
3280                             spapr_get_modern_hotplug_events,
3281                             spapr_set_modern_hotplug_events);
3282     object_property_set_description(obj, "modern-hotplug-events",
3283                                     "Use dedicated hotplug event mechanism in"
3284                                     " place of standard EPOW events when possible"
3285                                     " (required for memory hot-unplug support)");
3286     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3287                             "Maximum permitted CPU compatibility mode");
3288 
3289     object_property_add_str(obj, "resize-hpt",
3290                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3291     object_property_set_description(obj, "resize-hpt",
3292                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3293     object_property_add_uint32_ptr(obj, "vsmt",
3294                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3295     object_property_set_description(obj, "vsmt",
3296                                     "Virtual SMT: KVM behaves as if this were"
3297                                     " the host's SMT mode");
3298 
3299     object_property_add_bool(obj, "vfio-no-msix-emulation",
3300                              spapr_get_msix_emulation, NULL);
3301 
3302     object_property_add_uint64_ptr(obj, "kernel-addr",
3303                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3304     object_property_set_description(obj, "kernel-addr",
3305                                     stringify(KERNEL_LOAD_ADDR)
3306                                     " for -kernel is the default");
3307     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3308     /* The machine class defines the default interrupt controller mode */
3309     spapr->irq = smc->irq;
3310     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3311                             spapr_set_ic_mode);
3312     object_property_set_description(obj, "ic-mode",
3313                  "Specifies the interrupt controller mode (xics, xive, dual)");
3314 
3315     object_property_add_str(obj, "host-model",
3316         spapr_get_host_model, spapr_set_host_model);
3317     object_property_set_description(obj, "host-model",
3318         "Host model to advertise in guest device tree");
3319     object_property_add_str(obj, "host-serial",
3320         spapr_get_host_serial, spapr_set_host_serial);
3321     object_property_set_description(obj, "host-serial",
3322         "Host serial number to advertise in guest device tree");
3323 }
3324 
3325 static void spapr_machine_finalizefn(Object *obj)
3326 {
3327     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3328 
3329     g_free(spapr->kvm_type);
3330 }
3331 
3332 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3333 {
3334     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3335     PowerPCCPU *cpu = POWERPC_CPU(cs);
3336     CPUPPCState *env = &cpu->env;
3337 
3338     cpu_synchronize_state(cs);
3339     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3340     if (spapr->fwnmi_system_reset_addr != -1) {
3341         uint64_t rtas_addr, addr;
3342 
3343         /* get rtas addr from fdt */
3344         rtas_addr = spapr_get_rtas_addr();
3345         if (!rtas_addr) {
3346             qemu_system_guest_panicked(NULL);
3347             return;
3348         }
3349 
3350         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3351         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3352         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3353         env->gpr[3] = addr;
3354     }
3355     ppc_cpu_do_system_reset(cs);
3356     if (spapr->fwnmi_system_reset_addr != -1) {
3357         env->nip = spapr->fwnmi_system_reset_addr;
3358     }
3359 }
3360 
3361 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3362 {
3363     CPUState *cs;
3364 
3365     CPU_FOREACH(cs) {
3366         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3367     }
3368 }
3369 
3370 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3371                           void *fdt, int *fdt_start_offset, Error **errp)
3372 {
3373     uint64_t addr;
3374     uint32_t node;
3375 
3376     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3377     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3378                                     &error_abort);
3379     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3380                                              SPAPR_MEMORY_BLOCK_SIZE);
3381     return 0;
3382 }
3383 
3384 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3385                            bool dedicated_hp_event_source, Error **errp)
3386 {
3387     SpaprDrc *drc;
3388     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3389     int i;
3390     uint64_t addr = addr_start;
3391     bool hotplugged = spapr_drc_hotplugged(dev);
3392     Error *local_err = NULL;
3393 
3394     for (i = 0; i < nr_lmbs; i++) {
3395         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3396                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3397         g_assert(drc);
3398 
3399         spapr_drc_attach(drc, dev, &local_err);
3400         if (local_err) {
3401             while (addr > addr_start) {
3402                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3403                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3404                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3405                 spapr_drc_detach(drc);
3406             }
3407             error_propagate(errp, local_err);
3408             return;
3409         }
3410         if (!hotplugged) {
3411             spapr_drc_reset(drc);
3412         }
3413         addr += SPAPR_MEMORY_BLOCK_SIZE;
3414     }
3415     /* send hotplug notification to the
3416      * guest only in case of hotplugged memory
3417      */
3418     if (hotplugged) {
3419         if (dedicated_hp_event_source) {
3420             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3421                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3422             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3423                                                    nr_lmbs,
3424                                                    spapr_drc_index(drc));
3425         } else {
3426             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3427                                            nr_lmbs);
3428         }
3429     }
3430 }
3431 
3432 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3433                               Error **errp)
3434 {
3435     Error *local_err = NULL;
3436     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3437     PCDIMMDevice *dimm = PC_DIMM(dev);
3438     uint64_t size, addr, slot;
3439     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3440 
3441     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3442 
3443     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3444     if (local_err) {
3445         goto out;
3446     }
3447 
3448     if (!is_nvdimm) {
3449         addr = object_property_get_uint(OBJECT(dimm),
3450                                         PC_DIMM_ADDR_PROP, &local_err);
3451         if (local_err) {
3452             goto out_unplug;
3453         }
3454         spapr_add_lmbs(dev, addr, size,
3455                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3456                        &local_err);
3457     } else {
3458         slot = object_property_get_uint(OBJECT(dimm),
3459                                         PC_DIMM_SLOT_PROP, &local_err);
3460         if (local_err) {
3461             goto out_unplug;
3462         }
3463         spapr_add_nvdimm(dev, slot, &local_err);
3464     }
3465 
3466     if (local_err) {
3467         goto out_unplug;
3468     }
3469 
3470     return;
3471 
3472 out_unplug:
3473     pc_dimm_unplug(dimm, MACHINE(ms));
3474 out:
3475     error_propagate(errp, local_err);
3476 }
3477 
3478 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3479                                   Error **errp)
3480 {
3481     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3482     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3483     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3484     PCDIMMDevice *dimm = PC_DIMM(dev);
3485     Error *local_err = NULL;
3486     uint64_t size;
3487     Object *memdev;
3488     hwaddr pagesize;
3489 
3490     if (!smc->dr_lmb_enabled) {
3491         error_setg(errp, "Memory hotplug not supported for this machine");
3492         return;
3493     }
3494 
3495     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3496     if (local_err) {
3497         error_propagate(errp, local_err);
3498         return;
3499     }
3500 
3501     if (is_nvdimm) {
3502         spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, &local_err);
3503         if (local_err) {
3504             error_propagate(errp, local_err);
3505             return;
3506         }
3507     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3508         error_setg(errp, "Hotplugged memory size must be a multiple of "
3509                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3510         return;
3511     }
3512 
3513     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3514                                       &error_abort);
3515     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3516     spapr_check_pagesize(spapr, pagesize, &local_err);
3517     if (local_err) {
3518         error_propagate(errp, local_err);
3519         return;
3520     }
3521 
3522     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3523 }
3524 
3525 struct SpaprDimmState {
3526     PCDIMMDevice *dimm;
3527     uint32_t nr_lmbs;
3528     QTAILQ_ENTRY(SpaprDimmState) next;
3529 };
3530 
3531 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3532                                                        PCDIMMDevice *dimm)
3533 {
3534     SpaprDimmState *dimm_state = NULL;
3535 
3536     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3537         if (dimm_state->dimm == dimm) {
3538             break;
3539         }
3540     }
3541     return dimm_state;
3542 }
3543 
3544 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3545                                                       uint32_t nr_lmbs,
3546                                                       PCDIMMDevice *dimm)
3547 {
3548     SpaprDimmState *ds = NULL;
3549 
3550     /*
3551      * If this request is for a DIMM whose removal had failed earlier
3552      * (due to guest's refusal to remove the LMBs), we would have this
3553      * dimm already in the pending_dimm_unplugs list. In that
3554      * case don't add again.
3555      */
3556     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3557     if (!ds) {
3558         ds = g_malloc0(sizeof(SpaprDimmState));
3559         ds->nr_lmbs = nr_lmbs;
3560         ds->dimm = dimm;
3561         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3562     }
3563     return ds;
3564 }
3565 
3566 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3567                                               SpaprDimmState *dimm_state)
3568 {
3569     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3570     g_free(dimm_state);
3571 }
3572 
3573 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3574                                                         PCDIMMDevice *dimm)
3575 {
3576     SpaprDrc *drc;
3577     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3578                                                   &error_abort);
3579     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3580     uint32_t avail_lmbs = 0;
3581     uint64_t addr_start, addr;
3582     int i;
3583 
3584     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3585                                          &error_abort);
3586 
3587     addr = addr_start;
3588     for (i = 0; i < nr_lmbs; i++) {
3589         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3590                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3591         g_assert(drc);
3592         if (drc->dev) {
3593             avail_lmbs++;
3594         }
3595         addr += SPAPR_MEMORY_BLOCK_SIZE;
3596     }
3597 
3598     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3599 }
3600 
3601 /* Callback to be called during DRC release. */
3602 void spapr_lmb_release(DeviceState *dev)
3603 {
3604     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3605     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3606     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3607 
3608     /* This information will get lost if a migration occurs
3609      * during the unplug process. In this case recover it. */
3610     if (ds == NULL) {
3611         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3612         g_assert(ds);
3613         /* The DRC being examined by the caller at least must be counted */
3614         g_assert(ds->nr_lmbs);
3615     }
3616 
3617     if (--ds->nr_lmbs) {
3618         return;
3619     }
3620 
3621     /*
3622      * Now that all the LMBs have been removed by the guest, call the
3623      * unplug handler chain. This can never fail.
3624      */
3625     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3626     object_unparent(OBJECT(dev));
3627 }
3628 
3629 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3630 {
3631     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3632     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3633 
3634     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3635     qdev_unrealize(dev);
3636     spapr_pending_dimm_unplugs_remove(spapr, ds);
3637 }
3638 
3639 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3640                                         DeviceState *dev, Error **errp)
3641 {
3642     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3643     Error *local_err = NULL;
3644     PCDIMMDevice *dimm = PC_DIMM(dev);
3645     uint32_t nr_lmbs;
3646     uint64_t size, addr_start, addr;
3647     int i;
3648     SpaprDrc *drc;
3649 
3650     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3651         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3652         return;
3653     }
3654 
3655     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3656     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3657 
3658     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3659                                          &local_err);
3660     if (local_err) {
3661         error_propagate(errp, local_err);
3662         return;
3663     }
3664 
3665     /*
3666      * An existing pending dimm state for this DIMM means that there is an
3667      * unplug operation in progress, waiting for the spapr_lmb_release
3668      * callback to complete the job (BQL can't cover that far). In this case,
3669      * bail out to avoid detaching DRCs that were already released.
3670      */
3671     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3672         error_setg(errp, "Memory unplug already in progress for device %s",
3673                    dev->id);
3674         return;
3675     }
3676 
3677     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3678 
3679     addr = addr_start;
3680     for (i = 0; i < nr_lmbs; i++) {
3681         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3682                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3683         g_assert(drc);
3684 
3685         spapr_drc_detach(drc);
3686         addr += SPAPR_MEMORY_BLOCK_SIZE;
3687     }
3688 
3689     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3690                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3691     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3692                                               nr_lmbs, spapr_drc_index(drc));
3693 }
3694 
3695 /* Callback to be called during DRC release. */
3696 void spapr_core_release(DeviceState *dev)
3697 {
3698     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3699 
3700     /* Call the unplug handler chain. This can never fail. */
3701     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3702     object_unparent(OBJECT(dev));
3703 }
3704 
3705 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3706 {
3707     MachineState *ms = MACHINE(hotplug_dev);
3708     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3709     CPUCore *cc = CPU_CORE(dev);
3710     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3711 
3712     if (smc->pre_2_10_has_unused_icps) {
3713         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3714         int i;
3715 
3716         for (i = 0; i < cc->nr_threads; i++) {
3717             CPUState *cs = CPU(sc->threads[i]);
3718 
3719             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3720         }
3721     }
3722 
3723     assert(core_slot);
3724     core_slot->cpu = NULL;
3725     qdev_unrealize(dev);
3726 }
3727 
3728 static
3729 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3730                                Error **errp)
3731 {
3732     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3733     int index;
3734     SpaprDrc *drc;
3735     CPUCore *cc = CPU_CORE(dev);
3736 
3737     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3738         error_setg(errp, "Unable to find CPU core with core-id: %d",
3739                    cc->core_id);
3740         return;
3741     }
3742     if (index == 0) {
3743         error_setg(errp, "Boot CPU core may not be unplugged");
3744         return;
3745     }
3746 
3747     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3748                           spapr_vcpu_id(spapr, cc->core_id));
3749     g_assert(drc);
3750 
3751     if (!spapr_drc_unplug_requested(drc)) {
3752         spapr_drc_detach(drc);
3753         spapr_hotplug_req_remove_by_index(drc);
3754     }
3755 }
3756 
3757 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3758                            void *fdt, int *fdt_start_offset, Error **errp)
3759 {
3760     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3761     CPUState *cs = CPU(core->threads[0]);
3762     PowerPCCPU *cpu = POWERPC_CPU(cs);
3763     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3764     int id = spapr_get_vcpu_id(cpu);
3765     char *nodename;
3766     int offset;
3767 
3768     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3769     offset = fdt_add_subnode(fdt, 0, nodename);
3770     g_free(nodename);
3771 
3772     spapr_dt_cpu(cs, fdt, offset, spapr);
3773 
3774     *fdt_start_offset = offset;
3775     return 0;
3776 }
3777 
3778 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3779                             Error **errp)
3780 {
3781     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3782     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3783     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3784     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3785     CPUCore *cc = CPU_CORE(dev);
3786     CPUState *cs;
3787     SpaprDrc *drc;
3788     Error *local_err = NULL;
3789     CPUArchId *core_slot;
3790     int index;
3791     bool hotplugged = spapr_drc_hotplugged(dev);
3792     int i;
3793 
3794     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3795     if (!core_slot) {
3796         error_setg(errp, "Unable to find CPU core with core-id: %d",
3797                    cc->core_id);
3798         return;
3799     }
3800     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3801                           spapr_vcpu_id(spapr, cc->core_id));
3802 
3803     g_assert(drc || !mc->has_hotpluggable_cpus);
3804 
3805     if (drc) {
3806         spapr_drc_attach(drc, dev, &local_err);
3807         if (local_err) {
3808             error_propagate(errp, local_err);
3809             return;
3810         }
3811 
3812         if (hotplugged) {
3813             /*
3814              * Send hotplug notification interrupt to the guest only
3815              * in case of hotplugged CPUs.
3816              */
3817             spapr_hotplug_req_add_by_index(drc);
3818         } else {
3819             spapr_drc_reset(drc);
3820         }
3821     }
3822 
3823     core_slot->cpu = OBJECT(dev);
3824 
3825     if (smc->pre_2_10_has_unused_icps) {
3826         for (i = 0; i < cc->nr_threads; i++) {
3827             cs = CPU(core->threads[i]);
3828             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3829         }
3830     }
3831 
3832     /*
3833      * Set compatibility mode to match the boot CPU, which was either set
3834      * by the machine reset code or by CAS.
3835      */
3836     if (hotplugged) {
3837         for (i = 0; i < cc->nr_threads; i++) {
3838             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3839                            &local_err);
3840             if (local_err) {
3841                 error_propagate(errp, local_err);
3842                 return;
3843             }
3844         }
3845     }
3846 }
3847 
3848 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3849                                 Error **errp)
3850 {
3851     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3852     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3853     CPUCore *cc = CPU_CORE(dev);
3854     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3855     const char *type = object_get_typename(OBJECT(dev));
3856     CPUArchId *core_slot;
3857     int index;
3858     unsigned int smp_threads = machine->smp.threads;
3859 
3860     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3861         error_setg(errp, "CPU hotplug not supported for this machine");
3862         return;
3863     }
3864 
3865     if (strcmp(base_core_type, type)) {
3866         error_setg(errp, "CPU core type should be %s", base_core_type);
3867         return;
3868     }
3869 
3870     if (cc->core_id % smp_threads) {
3871         error_setg(errp, "invalid core id %d", cc->core_id);
3872         return;
3873     }
3874 
3875     /*
3876      * In general we should have homogeneous threads-per-core, but old
3877      * (pre hotplug support) machine types allow the last core to have
3878      * reduced threads as a compatibility hack for when we allowed
3879      * total vcpus not a multiple of threads-per-core.
3880      */
3881     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3882         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3883                    smp_threads);
3884         return;
3885     }
3886 
3887     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3888     if (!core_slot) {
3889         error_setg(errp, "core id %d out of range", cc->core_id);
3890         return;
3891     }
3892 
3893     if (core_slot->cpu) {
3894         error_setg(errp, "core %d already populated", cc->core_id);
3895         return;
3896     }
3897 
3898     numa_cpu_pre_plug(core_slot, dev, errp);
3899 }
3900 
3901 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3902                           void *fdt, int *fdt_start_offset, Error **errp)
3903 {
3904     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3905     int intc_phandle;
3906 
3907     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3908     if (intc_phandle <= 0) {
3909         return -1;
3910     }
3911 
3912     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3913         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3914         return -1;
3915     }
3916 
3917     /* generally SLOF creates these, for hotplug it's up to QEMU */
3918     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3919 
3920     return 0;
3921 }
3922 
3923 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3924                                Error **errp)
3925 {
3926     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3927     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3928     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3929     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3930 
3931     if (dev->hotplugged && !smc->dr_phb_enabled) {
3932         error_setg(errp, "PHB hotplug not supported for this machine");
3933         return;
3934     }
3935 
3936     if (sphb->index == (uint32_t)-1) {
3937         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3938         return;
3939     }
3940 
3941     /*
3942      * This will check that sphb->index doesn't exceed the maximum number of
3943      * PHBs for the current machine type.
3944      */
3945     smc->phb_placement(spapr, sphb->index,
3946                        &sphb->buid, &sphb->io_win_addr,
3947                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3948                        windows_supported, sphb->dma_liobn,
3949                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3950                        errp);
3951 }
3952 
3953 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3954                            Error **errp)
3955 {
3956     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3957     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3958     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3959     SpaprDrc *drc;
3960     bool hotplugged = spapr_drc_hotplugged(dev);
3961     Error *local_err = NULL;
3962 
3963     if (!smc->dr_phb_enabled) {
3964         return;
3965     }
3966 
3967     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3968     /* hotplug hooks should check it's enabled before getting this far */
3969     assert(drc);
3970 
3971     spapr_drc_attach(drc, dev, &local_err);
3972     if (local_err) {
3973         error_propagate(errp, local_err);
3974         return;
3975     }
3976 
3977     if (hotplugged) {
3978         spapr_hotplug_req_add_by_index(drc);
3979     } else {
3980         spapr_drc_reset(drc);
3981     }
3982 }
3983 
3984 void spapr_phb_release(DeviceState *dev)
3985 {
3986     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3987 
3988     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3989     object_unparent(OBJECT(dev));
3990 }
3991 
3992 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3993 {
3994     qdev_unrealize(dev);
3995 }
3996 
3997 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3998                                      DeviceState *dev, Error **errp)
3999 {
4000     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4001     SpaprDrc *drc;
4002 
4003     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4004     assert(drc);
4005 
4006     if (!spapr_drc_unplug_requested(drc)) {
4007         spapr_drc_detach(drc);
4008         spapr_hotplug_req_remove_by_index(drc);
4009     }
4010 }
4011 
4012 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4013                                  Error **errp)
4014 {
4015     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4016     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4017 
4018     if (spapr->tpm_proxy != NULL) {
4019         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4020         return;
4021     }
4022 
4023     spapr->tpm_proxy = tpm_proxy;
4024 }
4025 
4026 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4027 {
4028     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4029 
4030     qdev_unrealize(dev);
4031     object_unparent(OBJECT(dev));
4032     spapr->tpm_proxy = NULL;
4033 }
4034 
4035 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4036                                       DeviceState *dev, Error **errp)
4037 {
4038     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4039         spapr_memory_plug(hotplug_dev, dev, errp);
4040     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4041         spapr_core_plug(hotplug_dev, dev, errp);
4042     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4043         spapr_phb_plug(hotplug_dev, dev, errp);
4044     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4045         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4046     }
4047 }
4048 
4049 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4050                                         DeviceState *dev, Error **errp)
4051 {
4052     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4053         spapr_memory_unplug(hotplug_dev, dev);
4054     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4055         spapr_core_unplug(hotplug_dev, dev);
4056     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4057         spapr_phb_unplug(hotplug_dev, dev);
4058     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4059         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4060     }
4061 }
4062 
4063 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4064                                                 DeviceState *dev, Error **errp)
4065 {
4066     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4067     MachineClass *mc = MACHINE_GET_CLASS(sms);
4068     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4069 
4070     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4071         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4072             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4073         } else {
4074             /* NOTE: this means there is a window after guest reset, prior to
4075              * CAS negotiation, where unplug requests will fail due to the
4076              * capability not being detected yet. This is a bit different than
4077              * the case with PCI unplug, where the events will be queued and
4078              * eventually handled by the guest after boot
4079              */
4080             error_setg(errp, "Memory hot unplug not supported for this guest");
4081         }
4082     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4083         if (!mc->has_hotpluggable_cpus) {
4084             error_setg(errp, "CPU hot unplug not supported on this machine");
4085             return;
4086         }
4087         spapr_core_unplug_request(hotplug_dev, dev, errp);
4088     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4089         if (!smc->dr_phb_enabled) {
4090             error_setg(errp, "PHB hot unplug not supported on this machine");
4091             return;
4092         }
4093         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4094     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4095         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4096     }
4097 }
4098 
4099 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4100                                           DeviceState *dev, Error **errp)
4101 {
4102     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4103         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4105         spapr_core_pre_plug(hotplug_dev, dev, errp);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4107         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4108     }
4109 }
4110 
4111 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4112                                                  DeviceState *dev)
4113 {
4114     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4115         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4116         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4117         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4118         return HOTPLUG_HANDLER(machine);
4119     }
4120     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4121         PCIDevice *pcidev = PCI_DEVICE(dev);
4122         PCIBus *root = pci_device_root_bus(pcidev);
4123         SpaprPhbState *phb =
4124             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4125                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4126 
4127         if (phb) {
4128             return HOTPLUG_HANDLER(phb);
4129         }
4130     }
4131     return NULL;
4132 }
4133 
4134 static CpuInstanceProperties
4135 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4136 {
4137     CPUArchId *core_slot;
4138     MachineClass *mc = MACHINE_GET_CLASS(machine);
4139 
4140     /* make sure possible_cpu are intialized */
4141     mc->possible_cpu_arch_ids(machine);
4142     /* get CPU core slot containing thread that matches cpu_index */
4143     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4144     assert(core_slot);
4145     return core_slot->props;
4146 }
4147 
4148 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4149 {
4150     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4151 }
4152 
4153 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4154 {
4155     int i;
4156     unsigned int smp_threads = machine->smp.threads;
4157     unsigned int smp_cpus = machine->smp.cpus;
4158     const char *core_type;
4159     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4160     MachineClass *mc = MACHINE_GET_CLASS(machine);
4161 
4162     if (!mc->has_hotpluggable_cpus) {
4163         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4164     }
4165     if (machine->possible_cpus) {
4166         assert(machine->possible_cpus->len == spapr_max_cores);
4167         return machine->possible_cpus;
4168     }
4169 
4170     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4171     if (!core_type) {
4172         error_report("Unable to find sPAPR CPU Core definition");
4173         exit(1);
4174     }
4175 
4176     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4177                              sizeof(CPUArchId) * spapr_max_cores);
4178     machine->possible_cpus->len = spapr_max_cores;
4179     for (i = 0; i < machine->possible_cpus->len; i++) {
4180         int core_id = i * smp_threads;
4181 
4182         machine->possible_cpus->cpus[i].type = core_type;
4183         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4184         machine->possible_cpus->cpus[i].arch_id = core_id;
4185         machine->possible_cpus->cpus[i].props.has_core_id = true;
4186         machine->possible_cpus->cpus[i].props.core_id = core_id;
4187     }
4188     return machine->possible_cpus;
4189 }
4190 
4191 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4192                                 uint64_t *buid, hwaddr *pio,
4193                                 hwaddr *mmio32, hwaddr *mmio64,
4194                                 unsigned n_dma, uint32_t *liobns,
4195                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4196 {
4197     /*
4198      * New-style PHB window placement.
4199      *
4200      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4201      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4202      * windows.
4203      *
4204      * Some guest kernels can't work with MMIO windows above 1<<46
4205      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4206      *
4207      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4208      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4209      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4210      * 1TiB 64-bit MMIO windows for each PHB.
4211      */
4212     const uint64_t base_buid = 0x800000020000000ULL;
4213     int i;
4214 
4215     /* Sanity check natural alignments */
4216     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4217     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4218     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4219     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4220     /* Sanity check bounds */
4221     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4222                       SPAPR_PCI_MEM32_WIN_SIZE);
4223     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4224                       SPAPR_PCI_MEM64_WIN_SIZE);
4225 
4226     if (index >= SPAPR_MAX_PHBS) {
4227         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4228                    SPAPR_MAX_PHBS - 1);
4229         return;
4230     }
4231 
4232     *buid = base_buid + index;
4233     for (i = 0; i < n_dma; ++i) {
4234         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4235     }
4236 
4237     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4238     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4239     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4240 
4241     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4242     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4243 }
4244 
4245 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4246 {
4247     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4248 
4249     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4250 }
4251 
4252 static void spapr_ics_resend(XICSFabric *dev)
4253 {
4254     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4255 
4256     ics_resend(spapr->ics);
4257 }
4258 
4259 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4260 {
4261     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4262 
4263     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4264 }
4265 
4266 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4267                                  Monitor *mon)
4268 {
4269     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4270 
4271     spapr_irq_print_info(spapr, mon);
4272     monitor_printf(mon, "irqchip: %s\n",
4273                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4274 }
4275 
4276 /*
4277  * This is a XIVE only operation
4278  */
4279 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4280                            uint8_t nvt_blk, uint32_t nvt_idx,
4281                            bool cam_ignore, uint8_t priority,
4282                            uint32_t logic_serv, XiveTCTXMatch *match)
4283 {
4284     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4285     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4286     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4287     int count;
4288 
4289     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4290                            priority, logic_serv, match);
4291     if (count < 0) {
4292         return count;
4293     }
4294 
4295     /*
4296      * When we implement the save and restore of the thread interrupt
4297      * contexts in the enter/exit CPU handlers of the machine and the
4298      * escalations in QEMU, we should be able to handle non dispatched
4299      * vCPUs.
4300      *
4301      * Until this is done, the sPAPR machine should find at least one
4302      * matching context always.
4303      */
4304     if (count == 0) {
4305         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4306                       nvt_blk, nvt_idx);
4307     }
4308 
4309     return count;
4310 }
4311 
4312 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4313 {
4314     return cpu->vcpu_id;
4315 }
4316 
4317 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4318 {
4319     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4320     MachineState *ms = MACHINE(spapr);
4321     int vcpu_id;
4322 
4323     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4324 
4325     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4326         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4327         error_append_hint(errp, "Adjust the number of cpus to %d "
4328                           "or try to raise the number of threads per core\n",
4329                           vcpu_id * ms->smp.threads / spapr->vsmt);
4330         return;
4331     }
4332 
4333     cpu->vcpu_id = vcpu_id;
4334 }
4335 
4336 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4337 {
4338     CPUState *cs;
4339 
4340     CPU_FOREACH(cs) {
4341         PowerPCCPU *cpu = POWERPC_CPU(cs);
4342 
4343         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4344             return cpu;
4345         }
4346     }
4347 
4348     return NULL;
4349 }
4350 
4351 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4352 {
4353     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4354 
4355     /* These are only called by TCG, KVM maintains dispatch state */
4356 
4357     spapr_cpu->prod = false;
4358     if (spapr_cpu->vpa_addr) {
4359         CPUState *cs = CPU(cpu);
4360         uint32_t dispatch;
4361 
4362         dispatch = ldl_be_phys(cs->as,
4363                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4364         dispatch++;
4365         if ((dispatch & 1) != 0) {
4366             qemu_log_mask(LOG_GUEST_ERROR,
4367                           "VPA: incorrect dispatch counter value for "
4368                           "dispatched partition %u, correcting.\n", dispatch);
4369             dispatch++;
4370         }
4371         stl_be_phys(cs->as,
4372                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4373     }
4374 }
4375 
4376 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4377 {
4378     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4379 
4380     if (spapr_cpu->vpa_addr) {
4381         CPUState *cs = CPU(cpu);
4382         uint32_t dispatch;
4383 
4384         dispatch = ldl_be_phys(cs->as,
4385                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4386         dispatch++;
4387         if ((dispatch & 1) != 1) {
4388             qemu_log_mask(LOG_GUEST_ERROR,
4389                           "VPA: incorrect dispatch counter value for "
4390                           "preempted partition %u, correcting.\n", dispatch);
4391             dispatch++;
4392         }
4393         stl_be_phys(cs->as,
4394                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4395     }
4396 }
4397 
4398 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4399 {
4400     MachineClass *mc = MACHINE_CLASS(oc);
4401     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4402     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4403     NMIClass *nc = NMI_CLASS(oc);
4404     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4405     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4406     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4407     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4408     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4409 
4410     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4411     mc->ignore_boot_device_suffixes = true;
4412 
4413     /*
4414      * We set up the default / latest behaviour here.  The class_init
4415      * functions for the specific versioned machine types can override
4416      * these details for backwards compatibility
4417      */
4418     mc->init = spapr_machine_init;
4419     mc->reset = spapr_machine_reset;
4420     mc->block_default_type = IF_SCSI;
4421     mc->max_cpus = 1024;
4422     mc->no_parallel = 1;
4423     mc->default_boot_order = "";
4424     mc->default_ram_size = 512 * MiB;
4425     mc->default_ram_id = "ppc_spapr.ram";
4426     mc->default_display = "std";
4427     mc->kvm_type = spapr_kvm_type;
4428     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4429     mc->pci_allow_0_address = true;
4430     assert(!mc->get_hotplug_handler);
4431     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4432     hc->pre_plug = spapr_machine_device_pre_plug;
4433     hc->plug = spapr_machine_device_plug;
4434     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4435     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4436     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4437     hc->unplug_request = spapr_machine_device_unplug_request;
4438     hc->unplug = spapr_machine_device_unplug;
4439 
4440     smc->dr_lmb_enabled = true;
4441     smc->update_dt_enabled = true;
4442     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4443     mc->has_hotpluggable_cpus = true;
4444     mc->nvdimm_supported = true;
4445     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4446     fwc->get_dev_path = spapr_get_fw_dev_path;
4447     nc->nmi_monitor_handler = spapr_nmi;
4448     smc->phb_placement = spapr_phb_placement;
4449     vhc->hypercall = emulate_spapr_hypercall;
4450     vhc->hpt_mask = spapr_hpt_mask;
4451     vhc->map_hptes = spapr_map_hptes;
4452     vhc->unmap_hptes = spapr_unmap_hptes;
4453     vhc->hpte_set_c = spapr_hpte_set_c;
4454     vhc->hpte_set_r = spapr_hpte_set_r;
4455     vhc->get_pate = spapr_get_pate;
4456     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4457     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4458     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4459     xic->ics_get = spapr_ics_get;
4460     xic->ics_resend = spapr_ics_resend;
4461     xic->icp_get = spapr_icp_get;
4462     ispc->print_info = spapr_pic_print_info;
4463     /* Force NUMA node memory size to be a multiple of
4464      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4465      * in which LMBs are represented and hot-added
4466      */
4467     mc->numa_mem_align_shift = 28;
4468     mc->auto_enable_numa = true;
4469 
4470     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4471     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4472     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4473     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4474     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4475     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4476     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4477     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4478     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4479     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4480     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4481     spapr_caps_add_properties(smc);
4482     smc->irq = &spapr_irq_dual;
4483     smc->dr_phb_enabled = true;
4484     smc->linux_pci_probe = true;
4485     smc->smp_threads_vsmt = true;
4486     smc->nr_xirqs = SPAPR_NR_XIRQS;
4487     xfc->match_nvt = spapr_match_nvt;
4488 }
4489 
4490 static const TypeInfo spapr_machine_info = {
4491     .name          = TYPE_SPAPR_MACHINE,
4492     .parent        = TYPE_MACHINE,
4493     .abstract      = true,
4494     .instance_size = sizeof(SpaprMachineState),
4495     .instance_init = spapr_instance_init,
4496     .instance_finalize = spapr_machine_finalizefn,
4497     .class_size    = sizeof(SpaprMachineClass),
4498     .class_init    = spapr_machine_class_init,
4499     .interfaces = (InterfaceInfo[]) {
4500         { TYPE_FW_PATH_PROVIDER },
4501         { TYPE_NMI },
4502         { TYPE_HOTPLUG_HANDLER },
4503         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4504         { TYPE_XICS_FABRIC },
4505         { TYPE_INTERRUPT_STATS_PROVIDER },
4506         { TYPE_XIVE_FABRIC },
4507         { }
4508     },
4509 };
4510 
4511 static void spapr_machine_latest_class_options(MachineClass *mc)
4512 {
4513     mc->alias = "pseries";
4514     mc->is_default = true;
4515 }
4516 
4517 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4518     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4519                                                     void *data)      \
4520     {                                                                \
4521         MachineClass *mc = MACHINE_CLASS(oc);                        \
4522         spapr_machine_##suffix##_class_options(mc);                  \
4523         if (latest) {                                                \
4524             spapr_machine_latest_class_options(mc);                  \
4525         }                                                            \
4526     }                                                                \
4527     static const TypeInfo spapr_machine_##suffix##_info = {          \
4528         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4529         .parent = TYPE_SPAPR_MACHINE,                                \
4530         .class_init = spapr_machine_##suffix##_class_init,           \
4531     };                                                               \
4532     static void spapr_machine_register_##suffix(void)                \
4533     {                                                                \
4534         type_register(&spapr_machine_##suffix##_info);               \
4535     }                                                                \
4536     type_init(spapr_machine_register_##suffix)
4537 
4538 /*
4539  * pseries-5.2
4540  */
4541 static void spapr_machine_5_2_class_options(MachineClass *mc)
4542 {
4543     /* Defaults for the latest behaviour inherited from the base class */
4544 }
4545 
4546 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4547 
4548 /*
4549  * pseries-5.1
4550  */
4551 static void spapr_machine_5_1_class_options(MachineClass *mc)
4552 {
4553     spapr_machine_5_2_class_options(mc);
4554     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4555 }
4556 
4557 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4558 
4559 /*
4560  * pseries-5.0
4561  */
4562 static void spapr_machine_5_0_class_options(MachineClass *mc)
4563 {
4564     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4565     static GlobalProperty compat[] = {
4566         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4567     };
4568 
4569     spapr_machine_5_1_class_options(mc);
4570     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4571     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4572     mc->numa_mem_supported = true;
4573     smc->pre_5_1_assoc_refpoints = true;
4574 }
4575 
4576 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4577 
4578 /*
4579  * pseries-4.2
4580  */
4581 static void spapr_machine_4_2_class_options(MachineClass *mc)
4582 {
4583     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4584 
4585     spapr_machine_5_0_class_options(mc);
4586     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4587     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4588     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4589     smc->rma_limit = 16 * GiB;
4590     mc->nvdimm_supported = false;
4591 }
4592 
4593 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4594 
4595 /*
4596  * pseries-4.1
4597  */
4598 static void spapr_machine_4_1_class_options(MachineClass *mc)
4599 {
4600     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4601     static GlobalProperty compat[] = {
4602         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4603         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4604     };
4605 
4606     spapr_machine_4_2_class_options(mc);
4607     smc->linux_pci_probe = false;
4608     smc->smp_threads_vsmt = false;
4609     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4610     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4611 }
4612 
4613 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4614 
4615 /*
4616  * pseries-4.0
4617  */
4618 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4619                               uint64_t *buid, hwaddr *pio,
4620                               hwaddr *mmio32, hwaddr *mmio64,
4621                               unsigned n_dma, uint32_t *liobns,
4622                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4623 {
4624     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4625                         nv2gpa, nv2atsd, errp);
4626     *nv2gpa = 0;
4627     *nv2atsd = 0;
4628 }
4629 
4630 static void spapr_machine_4_0_class_options(MachineClass *mc)
4631 {
4632     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4633 
4634     spapr_machine_4_1_class_options(mc);
4635     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4636     smc->phb_placement = phb_placement_4_0;
4637     smc->irq = &spapr_irq_xics;
4638     smc->pre_4_1_migration = true;
4639 }
4640 
4641 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4642 
4643 /*
4644  * pseries-3.1
4645  */
4646 static void spapr_machine_3_1_class_options(MachineClass *mc)
4647 {
4648     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4649 
4650     spapr_machine_4_0_class_options(mc);
4651     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4652 
4653     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4654     smc->update_dt_enabled = false;
4655     smc->dr_phb_enabled = false;
4656     smc->broken_host_serial_model = true;
4657     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4658     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4659     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4660     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4661 }
4662 
4663 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4664 
4665 /*
4666  * pseries-3.0
4667  */
4668 
4669 static void spapr_machine_3_0_class_options(MachineClass *mc)
4670 {
4671     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4672 
4673     spapr_machine_3_1_class_options(mc);
4674     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4675 
4676     smc->legacy_irq_allocation = true;
4677     smc->nr_xirqs = 0x400;
4678     smc->irq = &spapr_irq_xics_legacy;
4679 }
4680 
4681 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4682 
4683 /*
4684  * pseries-2.12
4685  */
4686 static void spapr_machine_2_12_class_options(MachineClass *mc)
4687 {
4688     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4689     static GlobalProperty compat[] = {
4690         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4691         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4692     };
4693 
4694     spapr_machine_3_0_class_options(mc);
4695     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4696     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4697 
4698     /* We depend on kvm_enabled() to choose a default value for the
4699      * hpt-max-page-size capability. Of course we can't do it here
4700      * because this is too early and the HW accelerator isn't initialzed
4701      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4702      */
4703     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4704 }
4705 
4706 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4707 
4708 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4709 {
4710     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4711 
4712     spapr_machine_2_12_class_options(mc);
4713     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4714     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4715     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4716 }
4717 
4718 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4719 
4720 /*
4721  * pseries-2.11
4722  */
4723 
4724 static void spapr_machine_2_11_class_options(MachineClass *mc)
4725 {
4726     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4727 
4728     spapr_machine_2_12_class_options(mc);
4729     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4730     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4731 }
4732 
4733 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4734 
4735 /*
4736  * pseries-2.10
4737  */
4738 
4739 static void spapr_machine_2_10_class_options(MachineClass *mc)
4740 {
4741     spapr_machine_2_11_class_options(mc);
4742     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4743 }
4744 
4745 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4746 
4747 /*
4748  * pseries-2.9
4749  */
4750 
4751 static void spapr_machine_2_9_class_options(MachineClass *mc)
4752 {
4753     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4754     static GlobalProperty compat[] = {
4755         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4756     };
4757 
4758     spapr_machine_2_10_class_options(mc);
4759     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4760     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4761     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4762     smc->pre_2_10_has_unused_icps = true;
4763     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4764 }
4765 
4766 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4767 
4768 /*
4769  * pseries-2.8
4770  */
4771 
4772 static void spapr_machine_2_8_class_options(MachineClass *mc)
4773 {
4774     static GlobalProperty compat[] = {
4775         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4776     };
4777 
4778     spapr_machine_2_9_class_options(mc);
4779     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4780     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4781     mc->numa_mem_align_shift = 23;
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4785 
4786 /*
4787  * pseries-2.7
4788  */
4789 
4790 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4791                               uint64_t *buid, hwaddr *pio,
4792                               hwaddr *mmio32, hwaddr *mmio64,
4793                               unsigned n_dma, uint32_t *liobns,
4794                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4795 {
4796     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4797     const uint64_t base_buid = 0x800000020000000ULL;
4798     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4799     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4800     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4801     const uint32_t max_index = 255;
4802     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4803 
4804     uint64_t ram_top = MACHINE(spapr)->ram_size;
4805     hwaddr phb0_base, phb_base;
4806     int i;
4807 
4808     /* Do we have device memory? */
4809     if (MACHINE(spapr)->maxram_size > ram_top) {
4810         /* Can't just use maxram_size, because there may be an
4811          * alignment gap between normal and device memory regions
4812          */
4813         ram_top = MACHINE(spapr)->device_memory->base +
4814             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4815     }
4816 
4817     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4818 
4819     if (index > max_index) {
4820         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4821                    max_index);
4822         return;
4823     }
4824 
4825     *buid = base_buid + index;
4826     for (i = 0; i < n_dma; ++i) {
4827         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4828     }
4829 
4830     phb_base = phb0_base + index * phb_spacing;
4831     *pio = phb_base + pio_offset;
4832     *mmio32 = phb_base + mmio_offset;
4833     /*
4834      * We don't set the 64-bit MMIO window, relying on the PHB's
4835      * fallback behaviour of automatically splitting a large "32-bit"
4836      * window into contiguous 32-bit and 64-bit windows
4837      */
4838 
4839     *nv2gpa = 0;
4840     *nv2atsd = 0;
4841 }
4842 
4843 static void spapr_machine_2_7_class_options(MachineClass *mc)
4844 {
4845     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4846     static GlobalProperty compat[] = {
4847         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4848         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4849         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4850         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4851     };
4852 
4853     spapr_machine_2_8_class_options(mc);
4854     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4855     mc->default_machine_opts = "modern-hotplug-events=off";
4856     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4857     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4858     smc->phb_placement = phb_placement_2_7;
4859 }
4860 
4861 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4862 
4863 /*
4864  * pseries-2.6
4865  */
4866 
4867 static void spapr_machine_2_6_class_options(MachineClass *mc)
4868 {
4869     static GlobalProperty compat[] = {
4870         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4871     };
4872 
4873     spapr_machine_2_7_class_options(mc);
4874     mc->has_hotpluggable_cpus = false;
4875     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4876     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4877 }
4878 
4879 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4880 
4881 /*
4882  * pseries-2.5
4883  */
4884 
4885 static void spapr_machine_2_5_class_options(MachineClass *mc)
4886 {
4887     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4888     static GlobalProperty compat[] = {
4889         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4890     };
4891 
4892     spapr_machine_2_6_class_options(mc);
4893     smc->use_ohci_by_default = true;
4894     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4895     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4896 }
4897 
4898 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4899 
4900 /*
4901  * pseries-2.4
4902  */
4903 
4904 static void spapr_machine_2_4_class_options(MachineClass *mc)
4905 {
4906     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4907 
4908     spapr_machine_2_5_class_options(mc);
4909     smc->dr_lmb_enabled = false;
4910     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4911 }
4912 
4913 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4914 
4915 /*
4916  * pseries-2.3
4917  */
4918 
4919 static void spapr_machine_2_3_class_options(MachineClass *mc)
4920 {
4921     static GlobalProperty compat[] = {
4922         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4923     };
4924     spapr_machine_2_4_class_options(mc);
4925     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4926     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4927 }
4928 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4929 
4930 /*
4931  * pseries-2.2
4932  */
4933 
4934 static void spapr_machine_2_2_class_options(MachineClass *mc)
4935 {
4936     static GlobalProperty compat[] = {
4937         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4938     };
4939 
4940     spapr_machine_2_3_class_options(mc);
4941     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4942     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4943     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4944 }
4945 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4946 
4947 /*
4948  * pseries-2.1
4949  */
4950 
4951 static void spapr_machine_2_1_class_options(MachineClass *mc)
4952 {
4953     spapr_machine_2_2_class_options(mc);
4954     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4955 }
4956 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4957 
4958 static void spapr_machine_register_types(void)
4959 {
4960     type_register_static(&spapr_machine_info);
4961 }
4962 
4963 type_init(spapr_machine_register_types)
4964