xref: /openbmc/qemu/hw/ppc/spapr.c (revision 744c72a8)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/visitor.h"
33 #include "sysemu/sysemu.h"
34 #include "sysemu/hostmem.h"
35 #include "sysemu/numa.h"
36 #include "sysemu/qtest.h"
37 #include "sysemu/reset.h"
38 #include "sysemu/runstate.h"
39 #include "qemu/log.h"
40 #include "hw/fw-path-provider.h"
41 #include "elf.h"
42 #include "net/net.h"
43 #include "sysemu/device_tree.h"
44 #include "sysemu/cpus.h"
45 #include "sysemu/hw_accel.h"
46 #include "kvm_ppc.h"
47 #include "migration/misc.h"
48 #include "migration/qemu-file-types.h"
49 #include "migration/global_state.h"
50 #include "migration/register.h"
51 #include "migration/blocker.h"
52 #include "mmu-hash64.h"
53 #include "mmu-book3s-v3.h"
54 #include "cpu-models.h"
55 #include "hw/core/cpu.h"
56 
57 #include "hw/ppc/ppc.h"
58 #include "hw/loader.h"
59 
60 #include "hw/ppc/fdt.h"
61 #include "hw/ppc/spapr.h"
62 #include "hw/ppc/spapr_vio.h"
63 #include "hw/qdev-properties.h"
64 #include "hw/pci-host/spapr.h"
65 #include "hw/pci/msi.h"
66 
67 #include "hw/pci/pci.h"
68 #include "hw/scsi/scsi.h"
69 #include "hw/virtio/virtio-scsi.h"
70 #include "hw/virtio/vhost-scsi-common.h"
71 
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 #include "hw/ppc/pef.h"
86 
87 #include "monitor/monitor.h"
88 
89 #include <libfdt.h>
90 
91 /* SLOF memory layout:
92  *
93  * SLOF raw image loaded at 0, copies its romfs right below the flat
94  * device-tree, then position SLOF itself 31M below that
95  *
96  * So we set FW_OVERHEAD to 40MB which should account for all of that
97  * and more
98  *
99  * We load our kernel at 4M, leaving space for SLOF initial image
100  */
101 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
102 #define FW_MAX_SIZE             0x400000
103 #define FW_FILE_NAME            "slof.bin"
104 #define FW_OVERHEAD             0x2800000
105 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
106 
107 #define MIN_RMA_SLOF            (128 * MiB)
108 
109 #define PHANDLE_INTC            0x00001111
110 
111 /* These two functions implement the VCPU id numbering: one to compute them
112  * all and one to identify thread 0 of a VCORE. Any change to the first one
113  * is likely to have an impact on the second one, so let's keep them close.
114  */
115 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
116 {
117     MachineState *ms = MACHINE(spapr);
118     unsigned int smp_threads = ms->smp.threads;
119 
120     assert(spapr->vsmt);
121     return
122         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
123 }
124 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
125                                       PowerPCCPU *cpu)
126 {
127     assert(spapr->vsmt);
128     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
129 }
130 
131 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
132 {
133     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
134      * and newer QEMUs don't even have them. In both cases, we don't want
135      * to send anything on the wire.
136      */
137     return false;
138 }
139 
140 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
141     .name = "icp/server",
142     .version_id = 1,
143     .minimum_version_id = 1,
144     .needed = pre_2_10_vmstate_dummy_icp_needed,
145     .fields = (VMStateField[]) {
146         VMSTATE_UNUSED(4), /* uint32_t xirr */
147         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
148         VMSTATE_UNUSED(1), /* uint8_t mfrr */
149         VMSTATE_END_OF_LIST()
150     },
151 };
152 
153 static void pre_2_10_vmstate_register_dummy_icp(int i)
154 {
155     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
156                      (void *)(uintptr_t) i);
157 }
158 
159 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
160 {
161     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
162                        (void *)(uintptr_t) i);
163 }
164 
165 int spapr_max_server_number(SpaprMachineState *spapr)
166 {
167     MachineState *ms = MACHINE(spapr);
168 
169     assert(spapr->vsmt);
170     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
171 }
172 
173 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
174                                   int smt_threads)
175 {
176     int i, ret = 0;
177     uint32_t servers_prop[smt_threads];
178     uint32_t gservers_prop[smt_threads * 2];
179     int index = spapr_get_vcpu_id(cpu);
180 
181     if (cpu->compat_pvr) {
182         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
183         if (ret < 0) {
184             return ret;
185         }
186     }
187 
188     /* Build interrupt servers and gservers properties */
189     for (i = 0; i < smt_threads; i++) {
190         servers_prop[i] = cpu_to_be32(index + i);
191         /* Hack, direct the group queues back to cpu 0 */
192         gservers_prop[i*2] = cpu_to_be32(index + i);
193         gservers_prop[i*2 + 1] = 0;
194     }
195     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
196                       servers_prop, sizeof(servers_prop));
197     if (ret < 0) {
198         return ret;
199     }
200     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
201                       gservers_prop, sizeof(gservers_prop));
202 
203     return ret;
204 }
205 
206 static void spapr_dt_pa_features(SpaprMachineState *spapr,
207                                  PowerPCCPU *cpu,
208                                  void *fdt, int offset)
209 {
210     uint8_t pa_features_206[] = { 6, 0,
211         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
212     uint8_t pa_features_207[] = { 24, 0,
213         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
214         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
215         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
216         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
217     uint8_t pa_features_300[] = { 66, 0,
218         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
219         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
220         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
221         /* 6: DS207 */
222         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
223         /* 16: Vector */
224         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
225         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
226         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
227         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
228         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
229         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
230         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
231         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
233         /* 42: PM, 44: PC RA, 46: SC vec'd */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
235         /* 48: SIMD, 50: QP BFP, 52: String */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
237         /* 54: DecFP, 56: DecI, 58: SHA */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
239         /* 60: NM atomic, 62: RNG */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
241     };
242     uint8_t *pa_features = NULL;
243     size_t pa_size;
244 
245     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
246         pa_features = pa_features_206;
247         pa_size = sizeof(pa_features_206);
248     }
249     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
250         pa_features = pa_features_207;
251         pa_size = sizeof(pa_features_207);
252     }
253     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
254         pa_features = pa_features_300;
255         pa_size = sizeof(pa_features_300);
256     }
257     if (!pa_features) {
258         return;
259     }
260 
261     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
262         /*
263          * Note: we keep CI large pages off by default because a 64K capable
264          * guest provisioned with large pages might otherwise try to map a qemu
265          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
266          * even if that qemu runs on a 4k host.
267          * We dd this bit back here if we are confident this is not an issue
268          */
269         pa_features[3] |= 0x20;
270     }
271     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
272         pa_features[24] |= 0x80;    /* Transactional memory support */
273     }
274     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
275         /* Workaround for broken kernels that attempt (guest) radix
276          * mode when they can't handle it, if they see the radix bit set
277          * in pa-features. So hide it from them. */
278         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
279     }
280 
281     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
282 }
283 
284 static hwaddr spapr_node0_size(MachineState *machine)
285 {
286     if (machine->numa_state->num_nodes) {
287         int i;
288         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
289             if (machine->numa_state->nodes[i].node_mem) {
290                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
291                            machine->ram_size);
292             }
293         }
294     }
295     return machine->ram_size;
296 }
297 
298 static void add_str(GString *s, const gchar *s1)
299 {
300     g_string_append_len(s, s1, strlen(s1) + 1);
301 }
302 
303 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
304                                 hwaddr start, hwaddr size)
305 {
306     char mem_name[32];
307     uint64_t mem_reg_property[2];
308     int off;
309 
310     mem_reg_property[0] = cpu_to_be64(start);
311     mem_reg_property[1] = cpu_to_be64(size);
312 
313     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
314     off = fdt_add_subnode(fdt, 0, mem_name);
315     _FDT(off);
316     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
317     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
318                       sizeof(mem_reg_property))));
319     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
320     return off;
321 }
322 
323 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
324 {
325     MemoryDeviceInfoList *info;
326 
327     for (info = list; info; info = info->next) {
328         MemoryDeviceInfo *value = info->value;
329 
330         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
331             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
332 
333             if (addr >= pcdimm_info->addr &&
334                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
335                 return pcdimm_info->node;
336             }
337         }
338     }
339 
340     return -1;
341 }
342 
343 struct sPAPRDrconfCellV2 {
344      uint32_t seq_lmbs;
345      uint64_t base_addr;
346      uint32_t drc_index;
347      uint32_t aa_index;
348      uint32_t flags;
349 } QEMU_PACKED;
350 
351 typedef struct DrconfCellQueue {
352     struct sPAPRDrconfCellV2 cell;
353     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
354 } DrconfCellQueue;
355 
356 static DrconfCellQueue *
357 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
358                       uint32_t drc_index, uint32_t aa_index,
359                       uint32_t flags)
360 {
361     DrconfCellQueue *elem;
362 
363     elem = g_malloc0(sizeof(*elem));
364     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
365     elem->cell.base_addr = cpu_to_be64(base_addr);
366     elem->cell.drc_index = cpu_to_be32(drc_index);
367     elem->cell.aa_index = cpu_to_be32(aa_index);
368     elem->cell.flags = cpu_to_be32(flags);
369 
370     return elem;
371 }
372 
373 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
374                                       int offset, MemoryDeviceInfoList *dimms)
375 {
376     MachineState *machine = MACHINE(spapr);
377     uint8_t *int_buf, *cur_index;
378     int ret;
379     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
380     uint64_t addr, cur_addr, size;
381     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
382     uint64_t mem_end = machine->device_memory->base +
383                        memory_region_size(&machine->device_memory->mr);
384     uint32_t node, buf_len, nr_entries = 0;
385     SpaprDrc *drc;
386     DrconfCellQueue *elem, *next;
387     MemoryDeviceInfoList *info;
388     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
389         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
390 
391     /* Entry to cover RAM and the gap area */
392     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
393                                  SPAPR_LMB_FLAGS_RESERVED |
394                                  SPAPR_LMB_FLAGS_DRC_INVALID);
395     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
396     nr_entries++;
397 
398     cur_addr = machine->device_memory->base;
399     for (info = dimms; info; info = info->next) {
400         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
401 
402         addr = di->addr;
403         size = di->size;
404         node = di->node;
405 
406         /*
407          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
408          * area is marked hotpluggable in the next iteration for the bigger
409          * chunk including the NVDIMM occupied area.
410          */
411         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
412             continue;
413 
414         /* Entry for hot-pluggable area */
415         if (cur_addr < addr) {
416             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
417             g_assert(drc);
418             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
419                                          cur_addr, spapr_drc_index(drc), -1, 0);
420             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
421             nr_entries++;
422         }
423 
424         /* Entry for DIMM */
425         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
426         g_assert(drc);
427         elem = spapr_get_drconf_cell(size / lmb_size, addr,
428                                      spapr_drc_index(drc), node,
429                                      (SPAPR_LMB_FLAGS_ASSIGNED |
430                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
431         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
432         nr_entries++;
433         cur_addr = addr + size;
434     }
435 
436     /* Entry for remaining hotpluggable area */
437     if (cur_addr < mem_end) {
438         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
439         g_assert(drc);
440         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
441                                      cur_addr, spapr_drc_index(drc), -1, 0);
442         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
443         nr_entries++;
444     }
445 
446     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
447     int_buf = cur_index = g_malloc0(buf_len);
448     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
449     cur_index += sizeof(nr_entries);
450 
451     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
452         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
453         cur_index += sizeof(elem->cell);
454         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
455         g_free(elem);
456     }
457 
458     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
459     g_free(int_buf);
460     if (ret < 0) {
461         return -1;
462     }
463     return 0;
464 }
465 
466 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
467                                    int offset, MemoryDeviceInfoList *dimms)
468 {
469     MachineState *machine = MACHINE(spapr);
470     int i, ret;
471     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
472     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
473     uint32_t nr_lmbs = (machine->device_memory->base +
474                        memory_region_size(&machine->device_memory->mr)) /
475                        lmb_size;
476     uint32_t *int_buf, *cur_index, buf_len;
477 
478     /*
479      * Allocate enough buffer size to fit in ibm,dynamic-memory
480      */
481     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
482     cur_index = int_buf = g_malloc0(buf_len);
483     int_buf[0] = cpu_to_be32(nr_lmbs);
484     cur_index++;
485     for (i = 0; i < nr_lmbs; i++) {
486         uint64_t addr = i * lmb_size;
487         uint32_t *dynamic_memory = cur_index;
488 
489         if (i >= device_lmb_start) {
490             SpaprDrc *drc;
491 
492             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
493             g_assert(drc);
494 
495             dynamic_memory[0] = cpu_to_be32(addr >> 32);
496             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
497             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
498             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
499             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
500             if (memory_region_present(get_system_memory(), addr)) {
501                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
502             } else {
503                 dynamic_memory[5] = cpu_to_be32(0);
504             }
505         } else {
506             /*
507              * LMB information for RMA, boot time RAM and gap b/n RAM and
508              * device memory region -- all these are marked as reserved
509              * and as having no valid DRC.
510              */
511             dynamic_memory[0] = cpu_to_be32(addr >> 32);
512             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
513             dynamic_memory[2] = cpu_to_be32(0);
514             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
515             dynamic_memory[4] = cpu_to_be32(-1);
516             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
517                                             SPAPR_LMB_FLAGS_DRC_INVALID);
518         }
519 
520         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
521     }
522     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
523     g_free(int_buf);
524     if (ret < 0) {
525         return -1;
526     }
527     return 0;
528 }
529 
530 /*
531  * Adds ibm,dynamic-reconfiguration-memory node.
532  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
533  * of this device tree node.
534  */
535 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
536                                                    void *fdt)
537 {
538     MachineState *machine = MACHINE(spapr);
539     int ret, 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     MemoryDeviceInfoList *dimms = NULL;
544 
545     /*
546      * Don't create the node if there is no device memory
547      */
548     if (machine->ram_size == machine->maxram_size) {
549         return 0;
550     }
551 
552     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
553 
554     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
555                     sizeof(prop_lmb_size));
556     if (ret < 0) {
557         return ret;
558     }
559 
560     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
561     if (ret < 0) {
562         return ret;
563     }
564 
565     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
566     if (ret < 0) {
567         return ret;
568     }
569 
570     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
571     dimms = qmp_memory_device_list();
572     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
573         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
574     } else {
575         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
576     }
577     qapi_free_MemoryDeviceInfoList(dimms);
578 
579     if (ret < 0) {
580         return ret;
581     }
582 
583     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
584 
585     return ret;
586 }
587 
588 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
589 {
590     MachineState *machine = MACHINE(spapr);
591     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
592     hwaddr mem_start, node_size;
593     int i, nb_nodes = machine->numa_state->num_nodes;
594     NodeInfo *nodes = machine->numa_state->nodes;
595 
596     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
597         if (!nodes[i].node_mem) {
598             continue;
599         }
600         if (mem_start >= machine->ram_size) {
601             node_size = 0;
602         } else {
603             node_size = nodes[i].node_mem;
604             if (node_size > machine->ram_size - mem_start) {
605                 node_size = machine->ram_size - mem_start;
606             }
607         }
608         if (!mem_start) {
609             /* spapr_machine_init() checks for rma_size <= node0_size
610              * already */
611             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
612             mem_start += spapr->rma_size;
613             node_size -= spapr->rma_size;
614         }
615         for ( ; node_size; ) {
616             hwaddr sizetmp = pow2floor(node_size);
617 
618             /* mem_start != 0 here */
619             if (ctzl(mem_start) < ctzl(sizetmp)) {
620                 sizetmp = 1ULL << ctzl(mem_start);
621             }
622 
623             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
624             node_size -= sizetmp;
625             mem_start += sizetmp;
626         }
627     }
628 
629     /* Generate ibm,dynamic-reconfiguration-memory node if required */
630     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
631         int ret;
632 
633         g_assert(smc->dr_lmb_enabled);
634         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
635         if (ret) {
636             return ret;
637         }
638     }
639 
640     return 0;
641 }
642 
643 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
644                          SpaprMachineState *spapr)
645 {
646     MachineState *ms = MACHINE(spapr);
647     PowerPCCPU *cpu = POWERPC_CPU(cs);
648     CPUPPCState *env = &cpu->env;
649     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
650     int index = spapr_get_vcpu_id(cpu);
651     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
652                        0xffffffff, 0xffffffff};
653     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
654         : SPAPR_TIMEBASE_FREQ;
655     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
656     uint32_t page_sizes_prop[64];
657     size_t page_sizes_prop_size;
658     unsigned int smp_threads = ms->smp.threads;
659     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
660     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
661     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
662     SpaprDrc *drc;
663     int drc_index;
664     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
665     int i;
666 
667     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
668     if (drc) {
669         drc_index = spapr_drc_index(drc);
670         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
671     }
672 
673     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
674     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
675 
676     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
677     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
678                            env->dcache_line_size)));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
682                            env->icache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
684                            env->icache_line_size)));
685 
686     if (pcc->l1_dcache_size) {
687         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
688                                pcc->l1_dcache_size)));
689     } else {
690         warn_report("Unknown L1 dcache size for cpu");
691     }
692     if (pcc->l1_icache_size) {
693         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
694                                pcc->l1_icache_size)));
695     } else {
696         warn_report("Unknown L1 icache size for cpu");
697     }
698 
699     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
700     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
701     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
702     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
703     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
704     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
705 
706     if (ppc_has_spr(cpu, SPR_PURR)) {
707         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
708     }
709     if (ppc_has_spr(cpu, SPR_PURR)) {
710         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
711     }
712 
713     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
714         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
715                           segs, sizeof(segs))));
716     }
717 
718     /* Advertise VSX (vector extensions) if available
719      *   1               == VMX / Altivec available
720      *   2               == VSX available
721      *
722      * Only CPUs for which we create core types in spapr_cpu_core.c
723      * are possible, and all of those have VMX */
724     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
725         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
726     } else {
727         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
728     }
729 
730     /* Advertise DFP (Decimal Floating Point) if available
731      *   0 / no property == no DFP
732      *   1               == DFP available */
733     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
734         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
735     }
736 
737     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
738                                                       sizeof(page_sizes_prop));
739     if (page_sizes_prop_size) {
740         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
741                           page_sizes_prop, page_sizes_prop_size)));
742     }
743 
744     spapr_dt_pa_features(spapr, cpu, fdt, offset);
745 
746     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
747                            cs->cpu_index / vcpus_per_socket)));
748 
749     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
750                       pft_size_prop, sizeof(pft_size_prop))));
751 
752     if (ms->numa_state->num_nodes > 1) {
753         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
754     }
755 
756     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
757 
758     if (pcc->radix_page_info) {
759         for (i = 0; i < pcc->radix_page_info->count; i++) {
760             radix_AP_encodings[i] =
761                 cpu_to_be32(pcc->radix_page_info->entries[i]);
762         }
763         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
764                           radix_AP_encodings,
765                           pcc->radix_page_info->count *
766                           sizeof(radix_AP_encodings[0]))));
767     }
768 
769     /*
770      * We set this property to let the guest know that it can use the large
771      * decrementer and its width in bits.
772      */
773     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
774         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
775                               pcc->lrg_decr_bits)));
776 }
777 
778 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
779 {
780     CPUState **rev;
781     CPUState *cs;
782     int n_cpus;
783     int cpus_offset;
784     int i;
785 
786     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
787     _FDT(cpus_offset);
788     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
789     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
790 
791     /*
792      * We walk the CPUs in reverse order to ensure that CPU DT nodes
793      * created by fdt_add_subnode() end up in the right order in FDT
794      * for the guest kernel the enumerate the CPUs correctly.
795      *
796      * The CPU list cannot be traversed in reverse order, so we need
797      * to do extra work.
798      */
799     n_cpus = 0;
800     rev = NULL;
801     CPU_FOREACH(cs) {
802         rev = g_renew(CPUState *, rev, n_cpus + 1);
803         rev[n_cpus++] = cs;
804     }
805 
806     for (i = n_cpus - 1; i >= 0; i--) {
807         CPUState *cs = rev[i];
808         PowerPCCPU *cpu = POWERPC_CPU(cs);
809         int index = spapr_get_vcpu_id(cpu);
810         DeviceClass *dc = DEVICE_GET_CLASS(cs);
811         g_autofree char *nodename = NULL;
812         int offset;
813 
814         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
815             continue;
816         }
817 
818         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
819         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
820         _FDT(offset);
821         spapr_dt_cpu(cs, fdt, offset, spapr);
822     }
823 
824     g_free(rev);
825 }
826 
827 static int spapr_dt_rng(void *fdt)
828 {
829     int node;
830     int ret;
831 
832     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
833     if (node <= 0) {
834         return -1;
835     }
836     ret = fdt_setprop_string(fdt, node, "device_type",
837                              "ibm,platform-facilities");
838     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
839     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
840 
841     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
842     if (node <= 0) {
843         return -1;
844     }
845     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
846 
847     return ret ? -1 : 0;
848 }
849 
850 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
851 {
852     MachineState *ms = MACHINE(spapr);
853     int rtas;
854     GString *hypertas = g_string_sized_new(256);
855     GString *qemu_hypertas = g_string_sized_new(256);
856     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
857         memory_region_size(&MACHINE(spapr)->device_memory->mr);
858     uint32_t lrdr_capacity[] = {
859         cpu_to_be32(max_device_addr >> 32),
860         cpu_to_be32(max_device_addr & 0xffffffff),
861         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
862         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
863         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
864     };
865 
866     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
867 
868     /* hypertas */
869     add_str(hypertas, "hcall-pft");
870     add_str(hypertas, "hcall-term");
871     add_str(hypertas, "hcall-dabr");
872     add_str(hypertas, "hcall-interrupt");
873     add_str(hypertas, "hcall-tce");
874     add_str(hypertas, "hcall-vio");
875     add_str(hypertas, "hcall-splpar");
876     add_str(hypertas, "hcall-join");
877     add_str(hypertas, "hcall-bulk");
878     add_str(hypertas, "hcall-set-mode");
879     add_str(hypertas, "hcall-sprg0");
880     add_str(hypertas, "hcall-copy");
881     add_str(hypertas, "hcall-debug");
882     add_str(hypertas, "hcall-vphn");
883     add_str(qemu_hypertas, "hcall-memop1");
884 
885     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
886         add_str(hypertas, "hcall-multi-tce");
887     }
888 
889     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
890         add_str(hypertas, "hcall-hpt-resize");
891     }
892 
893     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
894                      hypertas->str, hypertas->len));
895     g_string_free(hypertas, TRUE);
896     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
897                      qemu_hypertas->str, qemu_hypertas->len));
898     g_string_free(qemu_hypertas, TRUE);
899 
900     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
901 
902     /*
903      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
904      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
905      *
906      * The system reset requirements are driven by existing Linux and PowerVM
907      * implementation which (contrary to PAPR) saves r3 in the error log
908      * structure like machine check, so Linux expects to find the saved r3
909      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
910      * does not look at the error value).
911      *
912      * System reset interrupts are not subject to interlock like machine
913      * check, so this memory area could be corrupted if the sreset is
914      * interrupted by a machine check (or vice versa) if it was shared. To
915      * prevent this, system reset uses per-CPU areas for the sreset save
916      * area. A system reset that interrupts a system reset handler could
917      * still overwrite this area, but Linux doesn't try to recover in that
918      * case anyway.
919      *
920      * The extra 8 bytes is required because Linux's FWNMI error log check
921      * is off-by-one.
922      */
923     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
924 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
925     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
926                           RTAS_ERROR_LOG_MAX));
927     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
928                           RTAS_EVENT_SCAN_RATE));
929 
930     g_assert(msi_nonbroken);
931     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
932 
933     /*
934      * According to PAPR, rtas ibm,os-term does not guarantee a return
935      * back to the guest cpu.
936      *
937      * While an additional ibm,extended-os-term property indicates
938      * that rtas call return will always occur. Set this property.
939      */
940     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
941 
942     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
943                      lrdr_capacity, sizeof(lrdr_capacity)));
944 
945     spapr_dt_rtas_tokens(fdt, rtas);
946 }
947 
948 /*
949  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
950  * and the XIVE features that the guest may request and thus the valid
951  * values for bytes 23..26 of option vector 5:
952  */
953 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
954                                           int chosen)
955 {
956     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
957 
958     char val[2 * 4] = {
959         23, 0x00, /* XICS / XIVE mode */
960         24, 0x00, /* Hash/Radix, filled in below. */
961         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
962         26, 0x40, /* Radix options: GTSE == yes. */
963     };
964 
965     if (spapr->irq->xics && spapr->irq->xive) {
966         val[1] = SPAPR_OV5_XIVE_BOTH;
967     } else if (spapr->irq->xive) {
968         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
969     } else {
970         assert(spapr->irq->xics);
971         val[1] = SPAPR_OV5_XIVE_LEGACY;
972     }
973 
974     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
975                           first_ppc_cpu->compat_pvr)) {
976         /*
977          * If we're in a pre POWER9 compat mode then the guest should
978          * do hash and use the legacy interrupt mode
979          */
980         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
981         val[3] = 0x00; /* Hash */
982         spapr_check_mmu_mode(false);
983     } else if (kvm_enabled()) {
984         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
985             val[3] = 0x80; /* OV5_MMU_BOTH */
986         } else if (kvmppc_has_cap_mmu_radix()) {
987             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
988         } else {
989             val[3] = 0x00; /* Hash */
990         }
991     } else {
992         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
993         val[3] = 0xC0;
994     }
995     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
996                      val, sizeof(val)));
997 }
998 
999 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1000 {
1001     MachineState *machine = MACHINE(spapr);
1002     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1003     int chosen;
1004 
1005     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1006 
1007     if (reset) {
1008         const char *boot_device = machine->boot_order;
1009         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1010         size_t cb = 0;
1011         char *bootlist = get_boot_devices_list(&cb);
1012 
1013         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1014             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1015                                     machine->kernel_cmdline));
1016         }
1017 
1018         if (spapr->initrd_size) {
1019             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1020                                   spapr->initrd_base));
1021             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1022                                   spapr->initrd_base + spapr->initrd_size));
1023         }
1024 
1025         if (spapr->kernel_size) {
1026             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1027                                   cpu_to_be64(spapr->kernel_size) };
1028 
1029             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1030                          &kprop, sizeof(kprop)));
1031             if (spapr->kernel_le) {
1032                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1033             }
1034         }
1035         if (boot_menu) {
1036             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1037         }
1038         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1039         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1040         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1041 
1042         if (cb && bootlist) {
1043             int i;
1044 
1045             for (i = 0; i < cb; i++) {
1046                 if (bootlist[i] == '\n') {
1047                     bootlist[i] = ' ';
1048                 }
1049             }
1050             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1051         }
1052 
1053         if (boot_device && strlen(boot_device)) {
1054             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1055         }
1056 
1057         if (!spapr->has_graphics && stdout_path) {
1058             /*
1059              * "linux,stdout-path" and "stdout" properties are
1060              * deprecated by linux kernel. New platforms should only
1061              * use the "stdout-path" property. Set the new property
1062              * and continue using older property to remain compatible
1063              * with the existing firmware.
1064              */
1065             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1066             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1067         }
1068 
1069         /*
1070          * We can deal with BAR reallocation just fine, advertise it
1071          * to the guest
1072          */
1073         if (smc->linux_pci_probe) {
1074             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1075         }
1076 
1077         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1078 
1079         g_free(stdout_path);
1080         g_free(bootlist);
1081     }
1082 
1083     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1084 }
1085 
1086 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1087 {
1088     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1089      * KVM to work under pHyp with some guest co-operation */
1090     int hypervisor;
1091     uint8_t hypercall[16];
1092 
1093     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1094     /* indicate KVM hypercall interface */
1095     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1096     if (kvmppc_has_cap_fixup_hcalls()) {
1097         /*
1098          * Older KVM versions with older guest kernels were broken
1099          * with the magic page, don't allow the guest to map it.
1100          */
1101         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1102                                   sizeof(hypercall))) {
1103             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1104                              hypercall, sizeof(hypercall)));
1105         }
1106     }
1107 }
1108 
1109 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1110 {
1111     MachineState *machine = MACHINE(spapr);
1112     MachineClass *mc = MACHINE_GET_CLASS(machine);
1113     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1114     uint32_t root_drc_type_mask = 0;
1115     int ret;
1116     void *fdt;
1117     SpaprPhbState *phb;
1118     char *buf;
1119 
1120     fdt = g_malloc0(space);
1121     _FDT((fdt_create_empty_tree(fdt, space)));
1122 
1123     /* Root node */
1124     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1125     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1126     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1127 
1128     /* Guest UUID & Name*/
1129     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1130     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1131     if (qemu_uuid_set) {
1132         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1133     }
1134     g_free(buf);
1135 
1136     if (qemu_get_vm_name()) {
1137         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1138                                 qemu_get_vm_name()));
1139     }
1140 
1141     /* Host Model & Serial Number */
1142     if (spapr->host_model) {
1143         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1144     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1145         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1146         g_free(buf);
1147     }
1148 
1149     if (spapr->host_serial) {
1150         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1151     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1153         g_free(buf);
1154     }
1155 
1156     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1157     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1158 
1159     /* /interrupt controller */
1160     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1161 
1162     ret = spapr_dt_memory(spapr, fdt);
1163     if (ret < 0) {
1164         error_report("couldn't setup memory nodes in fdt");
1165         exit(1);
1166     }
1167 
1168     /* /vdevice */
1169     spapr_dt_vdevice(spapr->vio_bus, fdt);
1170 
1171     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1172         ret = spapr_dt_rng(fdt);
1173         if (ret < 0) {
1174             error_report("could not set up rng device in the fdt");
1175             exit(1);
1176         }
1177     }
1178 
1179     QLIST_FOREACH(phb, &spapr->phbs, list) {
1180         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1181         if (ret < 0) {
1182             error_report("couldn't setup PCI devices in fdt");
1183             exit(1);
1184         }
1185     }
1186 
1187     spapr_dt_cpus(fdt, spapr);
1188 
1189     /* ibm,drc-indexes and friends */
1190     if (smc->dr_lmb_enabled) {
1191         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1192     }
1193     if (smc->dr_phb_enabled) {
1194         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1195     }
1196     if (mc->nvdimm_supported) {
1197         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1198     }
1199     if (root_drc_type_mask) {
1200         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1201     }
1202 
1203     if (mc->has_hotpluggable_cpus) {
1204         int offset = fdt_path_offset(fdt, "/cpus");
1205         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1206         if (ret < 0) {
1207             error_report("Couldn't set up CPU DR device tree properties");
1208             exit(1);
1209         }
1210     }
1211 
1212     /* /event-sources */
1213     spapr_dt_events(spapr, fdt);
1214 
1215     /* /rtas */
1216     spapr_dt_rtas(spapr, fdt);
1217 
1218     /* /chosen */
1219     spapr_dt_chosen(spapr, fdt, reset);
1220 
1221     /* /hypervisor */
1222     if (kvm_enabled()) {
1223         spapr_dt_hypervisor(spapr, fdt);
1224     }
1225 
1226     /* Build memory reserve map */
1227     if (reset) {
1228         if (spapr->kernel_size) {
1229             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1230                                   spapr->kernel_size)));
1231         }
1232         if (spapr->initrd_size) {
1233             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1234                                   spapr->initrd_size)));
1235         }
1236     }
1237 
1238     /* NVDIMM devices */
1239     if (mc->nvdimm_supported) {
1240         spapr_dt_persistent_memory(spapr, fdt);
1241     }
1242 
1243     return fdt;
1244 }
1245 
1246 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1247 {
1248     SpaprMachineState *spapr = opaque;
1249 
1250     return (addr & 0x0fffffff) + spapr->kernel_addr;
1251 }
1252 
1253 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1254                                     PowerPCCPU *cpu)
1255 {
1256     CPUPPCState *env = &cpu->env;
1257 
1258     /* The TCG path should also be holding the BQL at this point */
1259     g_assert(qemu_mutex_iothread_locked());
1260 
1261     if (msr_pr) {
1262         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1263         env->gpr[3] = H_PRIVILEGE;
1264     } else {
1265         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1266     }
1267 }
1268 
1269 struct LPCRSyncState {
1270     target_ulong value;
1271     target_ulong mask;
1272 };
1273 
1274 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1275 {
1276     struct LPCRSyncState *s = arg.host_ptr;
1277     PowerPCCPU *cpu = POWERPC_CPU(cs);
1278     CPUPPCState *env = &cpu->env;
1279     target_ulong lpcr;
1280 
1281     cpu_synchronize_state(cs);
1282     lpcr = env->spr[SPR_LPCR];
1283     lpcr &= ~s->mask;
1284     lpcr |= s->value;
1285     ppc_store_lpcr(cpu, lpcr);
1286 }
1287 
1288 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1289 {
1290     CPUState *cs;
1291     struct LPCRSyncState s = {
1292         .value = value,
1293         .mask = mask
1294     };
1295     CPU_FOREACH(cs) {
1296         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1297     }
1298 }
1299 
1300 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1301 {
1302     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1303 
1304     /* Copy PATE1:GR into PATE0:HR */
1305     entry->dw0 = spapr->patb_entry & PATE0_HR;
1306     entry->dw1 = spapr->patb_entry;
1307 }
1308 
1309 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1310 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1311 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1312 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1313 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1314 
1315 /*
1316  * Get the fd to access the kernel htab, re-opening it if necessary
1317  */
1318 static int get_htab_fd(SpaprMachineState *spapr)
1319 {
1320     Error *local_err = NULL;
1321 
1322     if (spapr->htab_fd >= 0) {
1323         return spapr->htab_fd;
1324     }
1325 
1326     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1327     if (spapr->htab_fd < 0) {
1328         error_report_err(local_err);
1329     }
1330 
1331     return spapr->htab_fd;
1332 }
1333 
1334 void close_htab_fd(SpaprMachineState *spapr)
1335 {
1336     if (spapr->htab_fd >= 0) {
1337         close(spapr->htab_fd);
1338     }
1339     spapr->htab_fd = -1;
1340 }
1341 
1342 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1343 {
1344     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1345 
1346     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1347 }
1348 
1349 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1350 {
1351     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1352 
1353     assert(kvm_enabled());
1354 
1355     if (!spapr->htab) {
1356         return 0;
1357     }
1358 
1359     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1360 }
1361 
1362 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1363                                                 hwaddr ptex, int n)
1364 {
1365     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1366     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1367 
1368     if (!spapr->htab) {
1369         /*
1370          * HTAB is controlled by KVM. Fetch into temporary buffer
1371          */
1372         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1373         kvmppc_read_hptes(hptes, ptex, n);
1374         return hptes;
1375     }
1376 
1377     /*
1378      * HTAB is controlled by QEMU. Just point to the internally
1379      * accessible PTEG.
1380      */
1381     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1382 }
1383 
1384 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1385                               const ppc_hash_pte64_t *hptes,
1386                               hwaddr ptex, int n)
1387 {
1388     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1389 
1390     if (!spapr->htab) {
1391         g_free((void *)hptes);
1392     }
1393 
1394     /* Nothing to do for qemu managed HPT */
1395 }
1396 
1397 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1398                       uint64_t pte0, uint64_t pte1)
1399 {
1400     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1401     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1402 
1403     if (!spapr->htab) {
1404         kvmppc_write_hpte(ptex, pte0, pte1);
1405     } else {
1406         if (pte0 & HPTE64_V_VALID) {
1407             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1408             /*
1409              * When setting valid, we write PTE1 first. This ensures
1410              * proper synchronization with the reading code in
1411              * ppc_hash64_pteg_search()
1412              */
1413             smp_wmb();
1414             stq_p(spapr->htab + offset, pte0);
1415         } else {
1416             stq_p(spapr->htab + offset, pte0);
1417             /*
1418              * When clearing it we set PTE0 first. This ensures proper
1419              * synchronization with the reading code in
1420              * ppc_hash64_pteg_search()
1421              */
1422             smp_wmb();
1423             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1424         }
1425     }
1426 }
1427 
1428 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1429                              uint64_t pte1)
1430 {
1431     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1432     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1433 
1434     if (!spapr->htab) {
1435         /* There should always be a hash table when this is called */
1436         error_report("spapr_hpte_set_c called with no hash table !");
1437         return;
1438     }
1439 
1440     /* The HW performs a non-atomic byte update */
1441     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1442 }
1443 
1444 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1445                              uint64_t pte1)
1446 {
1447     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1448     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1449 
1450     if (!spapr->htab) {
1451         /* There should always be a hash table when this is called */
1452         error_report("spapr_hpte_set_r called with no hash table !");
1453         return;
1454     }
1455 
1456     /* The HW performs a non-atomic byte update */
1457     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1458 }
1459 
1460 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1461 {
1462     int shift;
1463 
1464     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1465      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1466      * that's much more than is needed for Linux guests */
1467     shift = ctz64(pow2ceil(ramsize)) - 7;
1468     shift = MAX(shift, 18); /* Minimum architected size */
1469     shift = MIN(shift, 46); /* Maximum architected size */
1470     return shift;
1471 }
1472 
1473 void spapr_free_hpt(SpaprMachineState *spapr)
1474 {
1475     g_free(spapr->htab);
1476     spapr->htab = NULL;
1477     spapr->htab_shift = 0;
1478     close_htab_fd(spapr);
1479 }
1480 
1481 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1482 {
1483     ERRP_GUARD();
1484     long rc;
1485 
1486     /* Clean up any HPT info from a previous boot */
1487     spapr_free_hpt(spapr);
1488 
1489     rc = kvmppc_reset_htab(shift);
1490 
1491     if (rc == -EOPNOTSUPP) {
1492         error_setg(errp, "HPT not supported in nested guests");
1493         return -EOPNOTSUPP;
1494     }
1495 
1496     if (rc < 0) {
1497         /* kernel-side HPT needed, but couldn't allocate one */
1498         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1499                          shift);
1500         error_append_hint(errp, "Try smaller maxmem?\n");
1501         return -errno;
1502     } else if (rc > 0) {
1503         /* kernel-side HPT allocated */
1504         if (rc != shift) {
1505             error_setg(errp,
1506                        "Requested order %d HPT, but kernel allocated order %ld",
1507                        shift, rc);
1508             error_append_hint(errp, "Try smaller maxmem?\n");
1509             return -ENOSPC;
1510         }
1511 
1512         spapr->htab_shift = shift;
1513         spapr->htab = NULL;
1514     } else {
1515         /* kernel-side HPT not needed, allocate in userspace instead */
1516         size_t size = 1ULL << shift;
1517         int i;
1518 
1519         spapr->htab = qemu_memalign(size, size);
1520         memset(spapr->htab, 0, size);
1521         spapr->htab_shift = shift;
1522 
1523         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1524             DIRTY_HPTE(HPTE(spapr->htab, i));
1525         }
1526     }
1527     /* We're setting up a hash table, so that means we're not radix */
1528     spapr->patb_entry = 0;
1529     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1530     return 0;
1531 }
1532 
1533 void spapr_setup_hpt(SpaprMachineState *spapr)
1534 {
1535     int hpt_shift;
1536 
1537     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1538         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1539     } else {
1540         uint64_t current_ram_size;
1541 
1542         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1543         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1544     }
1545     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1546 
1547     if (kvm_enabled()) {
1548         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1549 
1550         /* Check our RMA fits in the possible VRMA */
1551         if (vrma_limit < spapr->rma_size) {
1552             error_report("Unable to create %" HWADDR_PRIu
1553                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1554                          spapr->rma_size / MiB, vrma_limit / MiB);
1555             exit(EXIT_FAILURE);
1556         }
1557     }
1558 }
1559 
1560 void spapr_check_mmu_mode(bool guest_radix)
1561 {
1562     if (guest_radix) {
1563         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1564             error_report("Guest requested unavailable MMU mode (radix).");
1565             exit(EXIT_FAILURE);
1566         }
1567     } else {
1568         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1569             && !kvmppc_has_cap_mmu_hash_v3()) {
1570             error_report("Guest requested unavailable MMU mode (hash).");
1571             exit(EXIT_FAILURE);
1572         }
1573     }
1574 }
1575 
1576 static void spapr_machine_reset(MachineState *machine)
1577 {
1578     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1579     PowerPCCPU *first_ppc_cpu;
1580     hwaddr fdt_addr;
1581     void *fdt;
1582     int rc;
1583 
1584     pef_kvm_reset(machine->cgs, &error_fatal);
1585     spapr_caps_apply(spapr);
1586 
1587     first_ppc_cpu = POWERPC_CPU(first_cpu);
1588     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1589         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1590                               spapr->max_compat_pvr)) {
1591         /*
1592          * If using KVM with radix mode available, VCPUs can be started
1593          * without a HPT because KVM will start them in radix mode.
1594          * Set the GR bit in PATE so that we know there is no HPT.
1595          */
1596         spapr->patb_entry = PATE1_GR;
1597         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1598     } else {
1599         spapr_setup_hpt(spapr);
1600     }
1601 
1602     qemu_devices_reset();
1603 
1604     spapr_ovec_cleanup(spapr->ov5_cas);
1605     spapr->ov5_cas = spapr_ovec_new();
1606 
1607     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1608 
1609     /*
1610      * This is fixing some of the default configuration of the XIVE
1611      * devices. To be called after the reset of the machine devices.
1612      */
1613     spapr_irq_reset(spapr, &error_fatal);
1614 
1615     /*
1616      * There is no CAS under qtest. Simulate one to please the code that
1617      * depends on spapr->ov5_cas. This is especially needed to test device
1618      * unplug, so we do that before resetting the DRCs.
1619      */
1620     if (qtest_enabled()) {
1621         spapr_ovec_cleanup(spapr->ov5_cas);
1622         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1623     }
1624 
1625     /* DRC reset may cause a device to be unplugged. This will cause troubles
1626      * if this device is used by another device (eg, a running vhost backend
1627      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1628      * situations, we reset DRCs after all devices have been reset.
1629      */
1630     spapr_drc_reset_all(spapr);
1631 
1632     spapr_clear_pending_events(spapr);
1633 
1634     /*
1635      * We place the device tree just below either the top of the RMA,
1636      * or just below 2GB, whichever is lower, so that it can be
1637      * processed with 32-bit real mode code if necessary
1638      */
1639     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1640 
1641     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1642 
1643     rc = fdt_pack(fdt);
1644 
1645     /* Should only fail if we've built a corrupted tree */
1646     assert(rc == 0);
1647 
1648     /* Load the fdt */
1649     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1650     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1651     g_free(spapr->fdt_blob);
1652     spapr->fdt_size = fdt_totalsize(fdt);
1653     spapr->fdt_initial_size = spapr->fdt_size;
1654     spapr->fdt_blob = fdt;
1655 
1656     /* Set up the entry state */
1657     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1658     first_ppc_cpu->env.gpr[5] = 0;
1659 
1660     spapr->fwnmi_system_reset_addr = -1;
1661     spapr->fwnmi_machine_check_addr = -1;
1662     spapr->fwnmi_machine_check_interlock = -1;
1663 
1664     /* Signal all vCPUs waiting on this condition */
1665     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1666 
1667     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1668 }
1669 
1670 static void spapr_create_nvram(SpaprMachineState *spapr)
1671 {
1672     DeviceState *dev = qdev_new("spapr-nvram");
1673     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1674 
1675     if (dinfo) {
1676         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1677                                 &error_fatal);
1678     }
1679 
1680     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1681 
1682     spapr->nvram = (struct SpaprNvram *)dev;
1683 }
1684 
1685 static void spapr_rtc_create(SpaprMachineState *spapr)
1686 {
1687     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1688                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1689                                        &error_fatal, NULL);
1690     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1691     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1692                               "date");
1693 }
1694 
1695 /* Returns whether we want to use VGA or not */
1696 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1697 {
1698     switch (vga_interface_type) {
1699     case VGA_NONE:
1700         return false;
1701     case VGA_DEVICE:
1702         return true;
1703     case VGA_STD:
1704     case VGA_VIRTIO:
1705     case VGA_CIRRUS:
1706         return pci_vga_init(pci_bus) != NULL;
1707     default:
1708         error_setg(errp,
1709                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1710         return false;
1711     }
1712 }
1713 
1714 static int spapr_pre_load(void *opaque)
1715 {
1716     int rc;
1717 
1718     rc = spapr_caps_pre_load(opaque);
1719     if (rc) {
1720         return rc;
1721     }
1722 
1723     return 0;
1724 }
1725 
1726 static int spapr_post_load(void *opaque, int version_id)
1727 {
1728     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1729     int err = 0;
1730 
1731     err = spapr_caps_post_migration(spapr);
1732     if (err) {
1733         return err;
1734     }
1735 
1736     /*
1737      * In earlier versions, there was no separate qdev for the PAPR
1738      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1739      * So when migrating from those versions, poke the incoming offset
1740      * value into the RTC device
1741      */
1742     if (version_id < 3) {
1743         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1744         if (err) {
1745             return err;
1746         }
1747     }
1748 
1749     if (kvm_enabled() && spapr->patb_entry) {
1750         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1751         bool radix = !!(spapr->patb_entry & PATE1_GR);
1752         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1753 
1754         /*
1755          * Update LPCR:HR and UPRT as they may not be set properly in
1756          * the stream
1757          */
1758         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1759                             LPCR_HR | LPCR_UPRT);
1760 
1761         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1762         if (err) {
1763             error_report("Process table config unsupported by the host");
1764             return -EINVAL;
1765         }
1766     }
1767 
1768     err = spapr_irq_post_load(spapr, version_id);
1769     if (err) {
1770         return err;
1771     }
1772 
1773     return err;
1774 }
1775 
1776 static int spapr_pre_save(void *opaque)
1777 {
1778     int rc;
1779 
1780     rc = spapr_caps_pre_save(opaque);
1781     if (rc) {
1782         return rc;
1783     }
1784 
1785     return 0;
1786 }
1787 
1788 static bool version_before_3(void *opaque, int version_id)
1789 {
1790     return version_id < 3;
1791 }
1792 
1793 static bool spapr_pending_events_needed(void *opaque)
1794 {
1795     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1796     return !QTAILQ_EMPTY(&spapr->pending_events);
1797 }
1798 
1799 static const VMStateDescription vmstate_spapr_event_entry = {
1800     .name = "spapr_event_log_entry",
1801     .version_id = 1,
1802     .minimum_version_id = 1,
1803     .fields = (VMStateField[]) {
1804         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1805         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1806         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1807                                      NULL, extended_length),
1808         VMSTATE_END_OF_LIST()
1809     },
1810 };
1811 
1812 static const VMStateDescription vmstate_spapr_pending_events = {
1813     .name = "spapr_pending_events",
1814     .version_id = 1,
1815     .minimum_version_id = 1,
1816     .needed = spapr_pending_events_needed,
1817     .fields = (VMStateField[]) {
1818         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1819                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1820         VMSTATE_END_OF_LIST()
1821     },
1822 };
1823 
1824 static bool spapr_ov5_cas_needed(void *opaque)
1825 {
1826     SpaprMachineState *spapr = opaque;
1827     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1828     bool cas_needed;
1829 
1830     /* Prior to the introduction of SpaprOptionVector, we had two option
1831      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1832      * Both of these options encode machine topology into the device-tree
1833      * in such a way that the now-booted OS should still be able to interact
1834      * appropriately with QEMU regardless of what options were actually
1835      * negotiatied on the source side.
1836      *
1837      * As such, we can avoid migrating the CAS-negotiated options if these
1838      * are the only options available on the current machine/platform.
1839      * Since these are the only options available for pseries-2.7 and
1840      * earlier, this allows us to maintain old->new/new->old migration
1841      * compatibility.
1842      *
1843      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1844      * via default pseries-2.8 machines and explicit command-line parameters.
1845      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1846      * of the actual CAS-negotiated values to continue working properly. For
1847      * example, availability of memory unplug depends on knowing whether
1848      * OV5_HP_EVT was negotiated via CAS.
1849      *
1850      * Thus, for any cases where the set of available CAS-negotiatable
1851      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1852      * include the CAS-negotiated options in the migration stream, unless
1853      * if they affect boot time behaviour only.
1854      */
1855     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1856     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1857     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1858 
1859     /* We need extra information if we have any bits outside the mask
1860      * defined above */
1861     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1862 
1863     spapr_ovec_cleanup(ov5_mask);
1864 
1865     return cas_needed;
1866 }
1867 
1868 static const VMStateDescription vmstate_spapr_ov5_cas = {
1869     .name = "spapr_option_vector_ov5_cas",
1870     .version_id = 1,
1871     .minimum_version_id = 1,
1872     .needed = spapr_ov5_cas_needed,
1873     .fields = (VMStateField[]) {
1874         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1875                                  vmstate_spapr_ovec, SpaprOptionVector),
1876         VMSTATE_END_OF_LIST()
1877     },
1878 };
1879 
1880 static bool spapr_patb_entry_needed(void *opaque)
1881 {
1882     SpaprMachineState *spapr = opaque;
1883 
1884     return !!spapr->patb_entry;
1885 }
1886 
1887 static const VMStateDescription vmstate_spapr_patb_entry = {
1888     .name = "spapr_patb_entry",
1889     .version_id = 1,
1890     .minimum_version_id = 1,
1891     .needed = spapr_patb_entry_needed,
1892     .fields = (VMStateField[]) {
1893         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1894         VMSTATE_END_OF_LIST()
1895     },
1896 };
1897 
1898 static bool spapr_irq_map_needed(void *opaque)
1899 {
1900     SpaprMachineState *spapr = opaque;
1901 
1902     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1903 }
1904 
1905 static const VMStateDescription vmstate_spapr_irq_map = {
1906     .name = "spapr_irq_map",
1907     .version_id = 1,
1908     .minimum_version_id = 1,
1909     .needed = spapr_irq_map_needed,
1910     .fields = (VMStateField[]) {
1911         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1912         VMSTATE_END_OF_LIST()
1913     },
1914 };
1915 
1916 static bool spapr_dtb_needed(void *opaque)
1917 {
1918     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1919 
1920     return smc->update_dt_enabled;
1921 }
1922 
1923 static int spapr_dtb_pre_load(void *opaque)
1924 {
1925     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1926 
1927     g_free(spapr->fdt_blob);
1928     spapr->fdt_blob = NULL;
1929     spapr->fdt_size = 0;
1930 
1931     return 0;
1932 }
1933 
1934 static const VMStateDescription vmstate_spapr_dtb = {
1935     .name = "spapr_dtb",
1936     .version_id = 1,
1937     .minimum_version_id = 1,
1938     .needed = spapr_dtb_needed,
1939     .pre_load = spapr_dtb_pre_load,
1940     .fields = (VMStateField[]) {
1941         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1942         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1943         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1944                                      fdt_size),
1945         VMSTATE_END_OF_LIST()
1946     },
1947 };
1948 
1949 static bool spapr_fwnmi_needed(void *opaque)
1950 {
1951     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1952 
1953     return spapr->fwnmi_machine_check_addr != -1;
1954 }
1955 
1956 static int spapr_fwnmi_pre_save(void *opaque)
1957 {
1958     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1959 
1960     /*
1961      * Check if machine check handling is in progress and print a
1962      * warning message.
1963      */
1964     if (spapr->fwnmi_machine_check_interlock != -1) {
1965         warn_report("A machine check is being handled during migration. The"
1966                 "handler may run and log hardware error on the destination");
1967     }
1968 
1969     return 0;
1970 }
1971 
1972 static const VMStateDescription vmstate_spapr_fwnmi = {
1973     .name = "spapr_fwnmi",
1974     .version_id = 1,
1975     .minimum_version_id = 1,
1976     .needed = spapr_fwnmi_needed,
1977     .pre_save = spapr_fwnmi_pre_save,
1978     .fields = (VMStateField[]) {
1979         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1980         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1981         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1982         VMSTATE_END_OF_LIST()
1983     },
1984 };
1985 
1986 static const VMStateDescription vmstate_spapr = {
1987     .name = "spapr",
1988     .version_id = 3,
1989     .minimum_version_id = 1,
1990     .pre_load = spapr_pre_load,
1991     .post_load = spapr_post_load,
1992     .pre_save = spapr_pre_save,
1993     .fields = (VMStateField[]) {
1994         /* used to be @next_irq */
1995         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1996 
1997         /* RTC offset */
1998         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1999 
2000         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2001         VMSTATE_END_OF_LIST()
2002     },
2003     .subsections = (const VMStateDescription*[]) {
2004         &vmstate_spapr_ov5_cas,
2005         &vmstate_spapr_patb_entry,
2006         &vmstate_spapr_pending_events,
2007         &vmstate_spapr_cap_htm,
2008         &vmstate_spapr_cap_vsx,
2009         &vmstate_spapr_cap_dfp,
2010         &vmstate_spapr_cap_cfpc,
2011         &vmstate_spapr_cap_sbbc,
2012         &vmstate_spapr_cap_ibs,
2013         &vmstate_spapr_cap_hpt_maxpagesize,
2014         &vmstate_spapr_irq_map,
2015         &vmstate_spapr_cap_nested_kvm_hv,
2016         &vmstate_spapr_dtb,
2017         &vmstate_spapr_cap_large_decr,
2018         &vmstate_spapr_cap_ccf_assist,
2019         &vmstate_spapr_cap_fwnmi,
2020         &vmstate_spapr_fwnmi,
2021         NULL
2022     }
2023 };
2024 
2025 static int htab_save_setup(QEMUFile *f, void *opaque)
2026 {
2027     SpaprMachineState *spapr = opaque;
2028 
2029     /* "Iteration" header */
2030     if (!spapr->htab_shift) {
2031         qemu_put_be32(f, -1);
2032     } else {
2033         qemu_put_be32(f, spapr->htab_shift);
2034     }
2035 
2036     if (spapr->htab) {
2037         spapr->htab_save_index = 0;
2038         spapr->htab_first_pass = true;
2039     } else {
2040         if (spapr->htab_shift) {
2041             assert(kvm_enabled());
2042         }
2043     }
2044 
2045 
2046     return 0;
2047 }
2048 
2049 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2050                             int chunkstart, int n_valid, int n_invalid)
2051 {
2052     qemu_put_be32(f, chunkstart);
2053     qemu_put_be16(f, n_valid);
2054     qemu_put_be16(f, n_invalid);
2055     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2056                     HASH_PTE_SIZE_64 * n_valid);
2057 }
2058 
2059 static void htab_save_end_marker(QEMUFile *f)
2060 {
2061     qemu_put_be32(f, 0);
2062     qemu_put_be16(f, 0);
2063     qemu_put_be16(f, 0);
2064 }
2065 
2066 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2067                                  int64_t max_ns)
2068 {
2069     bool has_timeout = max_ns != -1;
2070     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2071     int index = spapr->htab_save_index;
2072     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2073 
2074     assert(spapr->htab_first_pass);
2075 
2076     do {
2077         int chunkstart;
2078 
2079         /* Consume invalid HPTEs */
2080         while ((index < htabslots)
2081                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2082             CLEAN_HPTE(HPTE(spapr->htab, index));
2083             index++;
2084         }
2085 
2086         /* Consume valid HPTEs */
2087         chunkstart = index;
2088         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2089                && HPTE_VALID(HPTE(spapr->htab, index))) {
2090             CLEAN_HPTE(HPTE(spapr->htab, index));
2091             index++;
2092         }
2093 
2094         if (index > chunkstart) {
2095             int n_valid = index - chunkstart;
2096 
2097             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2098 
2099             if (has_timeout &&
2100                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2101                 break;
2102             }
2103         }
2104     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2105 
2106     if (index >= htabslots) {
2107         assert(index == htabslots);
2108         index = 0;
2109         spapr->htab_first_pass = false;
2110     }
2111     spapr->htab_save_index = index;
2112 }
2113 
2114 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2115                                 int64_t max_ns)
2116 {
2117     bool final = max_ns < 0;
2118     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2119     int examined = 0, sent = 0;
2120     int index = spapr->htab_save_index;
2121     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2122 
2123     assert(!spapr->htab_first_pass);
2124 
2125     do {
2126         int chunkstart, invalidstart;
2127 
2128         /* Consume non-dirty HPTEs */
2129         while ((index < htabslots)
2130                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2131             index++;
2132             examined++;
2133         }
2134 
2135         chunkstart = index;
2136         /* Consume valid dirty HPTEs */
2137         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2138                && HPTE_DIRTY(HPTE(spapr->htab, index))
2139                && HPTE_VALID(HPTE(spapr->htab, index))) {
2140             CLEAN_HPTE(HPTE(spapr->htab, index));
2141             index++;
2142             examined++;
2143         }
2144 
2145         invalidstart = index;
2146         /* Consume invalid dirty HPTEs */
2147         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2148                && HPTE_DIRTY(HPTE(spapr->htab, index))
2149                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2150             CLEAN_HPTE(HPTE(spapr->htab, index));
2151             index++;
2152             examined++;
2153         }
2154 
2155         if (index > chunkstart) {
2156             int n_valid = invalidstart - chunkstart;
2157             int n_invalid = index - invalidstart;
2158 
2159             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2160             sent += index - chunkstart;
2161 
2162             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2163                 break;
2164             }
2165         }
2166 
2167         if (examined >= htabslots) {
2168             break;
2169         }
2170 
2171         if (index >= htabslots) {
2172             assert(index == htabslots);
2173             index = 0;
2174         }
2175     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2176 
2177     if (index >= htabslots) {
2178         assert(index == htabslots);
2179         index = 0;
2180     }
2181 
2182     spapr->htab_save_index = index;
2183 
2184     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2185 }
2186 
2187 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2188 #define MAX_KVM_BUF_SIZE    2048
2189 
2190 static int htab_save_iterate(QEMUFile *f, void *opaque)
2191 {
2192     SpaprMachineState *spapr = opaque;
2193     int fd;
2194     int rc = 0;
2195 
2196     /* Iteration header */
2197     if (!spapr->htab_shift) {
2198         qemu_put_be32(f, -1);
2199         return 1;
2200     } else {
2201         qemu_put_be32(f, 0);
2202     }
2203 
2204     if (!spapr->htab) {
2205         assert(kvm_enabled());
2206 
2207         fd = get_htab_fd(spapr);
2208         if (fd < 0) {
2209             return fd;
2210         }
2211 
2212         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2213         if (rc < 0) {
2214             return rc;
2215         }
2216     } else  if (spapr->htab_first_pass) {
2217         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2218     } else {
2219         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2220     }
2221 
2222     htab_save_end_marker(f);
2223 
2224     return rc;
2225 }
2226 
2227 static int htab_save_complete(QEMUFile *f, void *opaque)
2228 {
2229     SpaprMachineState *spapr = opaque;
2230     int fd;
2231 
2232     /* Iteration header */
2233     if (!spapr->htab_shift) {
2234         qemu_put_be32(f, -1);
2235         return 0;
2236     } else {
2237         qemu_put_be32(f, 0);
2238     }
2239 
2240     if (!spapr->htab) {
2241         int rc;
2242 
2243         assert(kvm_enabled());
2244 
2245         fd = get_htab_fd(spapr);
2246         if (fd < 0) {
2247             return fd;
2248         }
2249 
2250         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2251         if (rc < 0) {
2252             return rc;
2253         }
2254     } else {
2255         if (spapr->htab_first_pass) {
2256             htab_save_first_pass(f, spapr, -1);
2257         }
2258         htab_save_later_pass(f, spapr, -1);
2259     }
2260 
2261     /* End marker */
2262     htab_save_end_marker(f);
2263 
2264     return 0;
2265 }
2266 
2267 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2268 {
2269     SpaprMachineState *spapr = opaque;
2270     uint32_t section_hdr;
2271     int fd = -1;
2272     Error *local_err = NULL;
2273 
2274     if (version_id < 1 || version_id > 1) {
2275         error_report("htab_load() bad version");
2276         return -EINVAL;
2277     }
2278 
2279     section_hdr = qemu_get_be32(f);
2280 
2281     if (section_hdr == -1) {
2282         spapr_free_hpt(spapr);
2283         return 0;
2284     }
2285 
2286     if (section_hdr) {
2287         int ret;
2288 
2289         /* First section gives the htab size */
2290         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2291         if (ret < 0) {
2292             error_report_err(local_err);
2293             return ret;
2294         }
2295         return 0;
2296     }
2297 
2298     if (!spapr->htab) {
2299         assert(kvm_enabled());
2300 
2301         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2302         if (fd < 0) {
2303             error_report_err(local_err);
2304             return fd;
2305         }
2306     }
2307 
2308     while (true) {
2309         uint32_t index;
2310         uint16_t n_valid, n_invalid;
2311 
2312         index = qemu_get_be32(f);
2313         n_valid = qemu_get_be16(f);
2314         n_invalid = qemu_get_be16(f);
2315 
2316         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2317             /* End of Stream */
2318             break;
2319         }
2320 
2321         if ((index + n_valid + n_invalid) >
2322             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2323             /* Bad index in stream */
2324             error_report(
2325                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2326                 index, n_valid, n_invalid, spapr->htab_shift);
2327             return -EINVAL;
2328         }
2329 
2330         if (spapr->htab) {
2331             if (n_valid) {
2332                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2333                                 HASH_PTE_SIZE_64 * n_valid);
2334             }
2335             if (n_invalid) {
2336                 memset(HPTE(spapr->htab, index + n_valid), 0,
2337                        HASH_PTE_SIZE_64 * n_invalid);
2338             }
2339         } else {
2340             int rc;
2341 
2342             assert(fd >= 0);
2343 
2344             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2345                                         &local_err);
2346             if (rc < 0) {
2347                 error_report_err(local_err);
2348                 return rc;
2349             }
2350         }
2351     }
2352 
2353     if (!spapr->htab) {
2354         assert(fd >= 0);
2355         close(fd);
2356     }
2357 
2358     return 0;
2359 }
2360 
2361 static void htab_save_cleanup(void *opaque)
2362 {
2363     SpaprMachineState *spapr = opaque;
2364 
2365     close_htab_fd(spapr);
2366 }
2367 
2368 static SaveVMHandlers savevm_htab_handlers = {
2369     .save_setup = htab_save_setup,
2370     .save_live_iterate = htab_save_iterate,
2371     .save_live_complete_precopy = htab_save_complete,
2372     .save_cleanup = htab_save_cleanup,
2373     .load_state = htab_load,
2374 };
2375 
2376 static void spapr_boot_set(void *opaque, const char *boot_device,
2377                            Error **errp)
2378 {
2379     MachineState *machine = MACHINE(opaque);
2380     machine->boot_order = g_strdup(boot_device);
2381 }
2382 
2383 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2384 {
2385     MachineState *machine = MACHINE(spapr);
2386     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2387     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2388     int i;
2389 
2390     for (i = 0; i < nr_lmbs; i++) {
2391         uint64_t addr;
2392 
2393         addr = i * lmb_size + machine->device_memory->base;
2394         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2395                                addr / lmb_size);
2396     }
2397 }
2398 
2399 /*
2400  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2401  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2402  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2403  */
2404 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2405 {
2406     int i;
2407 
2408     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2409         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2410                    " is not aligned to %" PRIu64 " MiB",
2411                    machine->ram_size,
2412                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2413         return;
2414     }
2415 
2416     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2417         error_setg(errp, "Maximum 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     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2425         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2426             error_setg(errp,
2427                        "Node %d memory size 0x%" PRIx64
2428                        " is not aligned to %" PRIu64 " MiB",
2429                        i, machine->numa_state->nodes[i].node_mem,
2430                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2431             return;
2432         }
2433     }
2434 }
2435 
2436 /* find cpu slot in machine->possible_cpus by core_id */
2437 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2438 {
2439     int index = id / ms->smp.threads;
2440 
2441     if (index >= ms->possible_cpus->len) {
2442         return NULL;
2443     }
2444     if (idx) {
2445         *idx = index;
2446     }
2447     return &ms->possible_cpus->cpus[index];
2448 }
2449 
2450 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2451 {
2452     MachineState *ms = MACHINE(spapr);
2453     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2454     Error *local_err = NULL;
2455     bool vsmt_user = !!spapr->vsmt;
2456     int kvm_smt = kvmppc_smt_threads();
2457     int ret;
2458     unsigned int smp_threads = ms->smp.threads;
2459 
2460     if (!kvm_enabled() && (smp_threads > 1)) {
2461         error_setg(errp, "TCG cannot support more than 1 thread/core "
2462                    "on a pseries machine");
2463         return;
2464     }
2465     if (!is_power_of_2(smp_threads)) {
2466         error_setg(errp, "Cannot support %d threads/core on a pseries "
2467                    "machine because it must be a power of 2", smp_threads);
2468         return;
2469     }
2470 
2471     /* Detemine the VSMT mode to use: */
2472     if (vsmt_user) {
2473         if (spapr->vsmt < smp_threads) {
2474             error_setg(errp, "Cannot support VSMT mode %d"
2475                        " because it must be >= threads/core (%d)",
2476                        spapr->vsmt, smp_threads);
2477             return;
2478         }
2479         /* In this case, spapr->vsmt has been set by the command line */
2480     } else if (!smc->smp_threads_vsmt) {
2481         /*
2482          * Default VSMT value is tricky, because we need it to be as
2483          * consistent as possible (for migration), but this requires
2484          * changing it for at least some existing cases.  We pick 8 as
2485          * the value that we'd get with KVM on POWER8, the
2486          * overwhelmingly common case in production systems.
2487          */
2488         spapr->vsmt = MAX(8, smp_threads);
2489     } else {
2490         spapr->vsmt = smp_threads;
2491     }
2492 
2493     /* KVM: If necessary, set the SMT mode: */
2494     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2495         ret = kvmppc_set_smt_threads(spapr->vsmt);
2496         if (ret) {
2497             /* Looks like KVM isn't able to change VSMT mode */
2498             error_setg(&local_err,
2499                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2500                        spapr->vsmt, ret);
2501             /* We can live with that if the default one is big enough
2502              * for the number of threads, and a submultiple of the one
2503              * we want.  In this case we'll waste some vcpu ids, but
2504              * behaviour will be correct */
2505             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2506                 warn_report_err(local_err);
2507             } else {
2508                 if (!vsmt_user) {
2509                     error_append_hint(&local_err,
2510                                       "On PPC, a VM with %d threads/core"
2511                                       " on a host with %d threads/core"
2512                                       " requires the use of VSMT mode %d.\n",
2513                                       smp_threads, kvm_smt, spapr->vsmt);
2514                 }
2515                 kvmppc_error_append_smt_possible_hint(&local_err);
2516                 error_propagate(errp, local_err);
2517             }
2518         }
2519     }
2520     /* else TCG: nothing to do currently */
2521 }
2522 
2523 static void spapr_init_cpus(SpaprMachineState *spapr)
2524 {
2525     MachineState *machine = MACHINE(spapr);
2526     MachineClass *mc = MACHINE_GET_CLASS(machine);
2527     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2528     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2529     const CPUArchIdList *possible_cpus;
2530     unsigned int smp_cpus = machine->smp.cpus;
2531     unsigned int smp_threads = machine->smp.threads;
2532     unsigned int max_cpus = machine->smp.max_cpus;
2533     int boot_cores_nr = smp_cpus / smp_threads;
2534     int i;
2535 
2536     possible_cpus = mc->possible_cpu_arch_ids(machine);
2537     if (mc->has_hotpluggable_cpus) {
2538         if (smp_cpus % smp_threads) {
2539             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2540                          smp_cpus, smp_threads);
2541             exit(1);
2542         }
2543         if (max_cpus % smp_threads) {
2544             error_report("max_cpus (%u) must be multiple of threads (%u)",
2545                          max_cpus, smp_threads);
2546             exit(1);
2547         }
2548     } else {
2549         if (max_cpus != smp_cpus) {
2550             error_report("This machine version does not support CPU hotplug");
2551             exit(1);
2552         }
2553         boot_cores_nr = possible_cpus->len;
2554     }
2555 
2556     if (smc->pre_2_10_has_unused_icps) {
2557         int i;
2558 
2559         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2560             /* Dummy entries get deregistered when real ICPState objects
2561              * are registered during CPU core hotplug.
2562              */
2563             pre_2_10_vmstate_register_dummy_icp(i);
2564         }
2565     }
2566 
2567     for (i = 0; i < possible_cpus->len; i++) {
2568         int core_id = i * smp_threads;
2569 
2570         if (mc->has_hotpluggable_cpus) {
2571             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2572                                    spapr_vcpu_id(spapr, core_id));
2573         }
2574 
2575         if (i < boot_cores_nr) {
2576             Object *core  = object_new(type);
2577             int nr_threads = smp_threads;
2578 
2579             /* Handle the partially filled core for older machine types */
2580             if ((i + 1) * smp_threads >= smp_cpus) {
2581                 nr_threads = smp_cpus - i * smp_threads;
2582             }
2583 
2584             object_property_set_int(core, "nr-threads", nr_threads,
2585                                     &error_fatal);
2586             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2587                                     &error_fatal);
2588             qdev_realize(DEVICE(core), NULL, &error_fatal);
2589 
2590             object_unref(core);
2591         }
2592     }
2593 }
2594 
2595 static PCIHostState *spapr_create_default_phb(void)
2596 {
2597     DeviceState *dev;
2598 
2599     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2600     qdev_prop_set_uint32(dev, "index", 0);
2601     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2602 
2603     return PCI_HOST_BRIDGE(dev);
2604 }
2605 
2606 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2607 {
2608     MachineState *machine = MACHINE(spapr);
2609     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2610     hwaddr rma_size = machine->ram_size;
2611     hwaddr node0_size = spapr_node0_size(machine);
2612 
2613     /* RMA has to fit in the first NUMA node */
2614     rma_size = MIN(rma_size, node0_size);
2615 
2616     /*
2617      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2618      * never exceed that
2619      */
2620     rma_size = MIN(rma_size, 1 * TiB);
2621 
2622     /*
2623      * Clamp the RMA size based on machine type.  This is for
2624      * migration compatibility with older qemu versions, which limited
2625      * the RMA size for complicated and mostly bad reasons.
2626      */
2627     if (smc->rma_limit) {
2628         rma_size = MIN(rma_size, smc->rma_limit);
2629     }
2630 
2631     if (rma_size < MIN_RMA_SLOF) {
2632         error_setg(errp,
2633                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2634                    "ldMiB guest RMA (Real Mode Area memory)",
2635                    MIN_RMA_SLOF / MiB);
2636         return 0;
2637     }
2638 
2639     return rma_size;
2640 }
2641 
2642 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2643 {
2644     MachineState *machine = MACHINE(spapr);
2645     int i;
2646 
2647     for (i = 0; i < machine->ram_slots; i++) {
2648         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2649     }
2650 }
2651 
2652 /* pSeries LPAR / sPAPR hardware init */
2653 static void spapr_machine_init(MachineState *machine)
2654 {
2655     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2656     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2657     MachineClass *mc = MACHINE_GET_CLASS(machine);
2658     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2659     const char *kernel_filename = machine->kernel_filename;
2660     const char *initrd_filename = machine->initrd_filename;
2661     PCIHostState *phb;
2662     int i;
2663     MemoryRegion *sysmem = get_system_memory();
2664     long load_limit, fw_size;
2665     char *filename;
2666     Error *resize_hpt_err = NULL;
2667 
2668     /*
2669      * if Secure VM (PEF) support is configured, then initialize it
2670      */
2671     pef_kvm_init(machine->cgs, &error_fatal);
2672 
2673     msi_nonbroken = true;
2674 
2675     QLIST_INIT(&spapr->phbs);
2676     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2677 
2678     /* Determine capabilities to run with */
2679     spapr_caps_init(spapr);
2680 
2681     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2682     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2683         /*
2684          * If the user explicitly requested a mode we should either
2685          * supply it, or fail completely (which we do below).  But if
2686          * it's not set explicitly, we reset our mode to something
2687          * that works
2688          */
2689         if (resize_hpt_err) {
2690             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2691             error_free(resize_hpt_err);
2692             resize_hpt_err = NULL;
2693         } else {
2694             spapr->resize_hpt = smc->resize_hpt_default;
2695         }
2696     }
2697 
2698     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2699 
2700     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2701         /*
2702          * User requested HPT resize, but this host can't supply it.  Bail out
2703          */
2704         error_report_err(resize_hpt_err);
2705         exit(1);
2706     }
2707     error_free(resize_hpt_err);
2708 
2709     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2710 
2711     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2712     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2713 
2714     /*
2715      * VSMT must be set in order to be able to compute VCPU ids, ie to
2716      * call spapr_max_server_number() or spapr_vcpu_id().
2717      */
2718     spapr_set_vsmt_mode(spapr, &error_fatal);
2719 
2720     /* Set up Interrupt Controller before we create the VCPUs */
2721     spapr_irq_init(spapr, &error_fatal);
2722 
2723     /* Set up containers for ibm,client-architecture-support negotiated options
2724      */
2725     spapr->ov5 = spapr_ovec_new();
2726     spapr->ov5_cas = spapr_ovec_new();
2727 
2728     if (smc->dr_lmb_enabled) {
2729         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2730         spapr_validate_node_memory(machine, &error_fatal);
2731     }
2732 
2733     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2734 
2735     /* advertise support for dedicated HP event source to guests */
2736     if (spapr->use_hotplug_event_source) {
2737         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2738     }
2739 
2740     /* advertise support for HPT resizing */
2741     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2742         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2743     }
2744 
2745     /* advertise support for ibm,dyamic-memory-v2 */
2746     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2747 
2748     /* advertise XIVE on POWER9 machines */
2749     if (spapr->irq->xive) {
2750         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2751     }
2752 
2753     /* init CPUs */
2754     spapr_init_cpus(spapr);
2755 
2756     /*
2757      * check we don't have a memory-less/cpu-less NUMA node
2758      * Firmware relies on the existing memory/cpu topology to provide the
2759      * NUMA topology to the kernel.
2760      * And the linux kernel needs to know the NUMA topology at start
2761      * to be able to hotplug CPUs later.
2762      */
2763     if (machine->numa_state->num_nodes) {
2764         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2765             /* check for memory-less node */
2766             if (machine->numa_state->nodes[i].node_mem == 0) {
2767                 CPUState *cs;
2768                 int found = 0;
2769                 /* check for cpu-less node */
2770                 CPU_FOREACH(cs) {
2771                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2772                     if (cpu->node_id == i) {
2773                         found = 1;
2774                         break;
2775                     }
2776                 }
2777                 /* memory-less and cpu-less node */
2778                 if (!found) {
2779                     error_report(
2780                        "Memory-less/cpu-less nodes are not supported (node %d)",
2781                                  i);
2782                     exit(1);
2783                 }
2784             }
2785         }
2786 
2787     }
2788 
2789     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2790 
2791     /* Init numa_assoc_array */
2792     spapr_numa_associativity_init(spapr, machine);
2793 
2794     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2795         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2796                               spapr->max_compat_pvr)) {
2797         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2798         /* KVM and TCG always allow GTSE with radix... */
2799         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2800     }
2801     /* ... but not with hash (currently). */
2802 
2803     if (kvm_enabled()) {
2804         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2805         kvmppc_enable_logical_ci_hcalls();
2806         kvmppc_enable_set_mode_hcall();
2807 
2808         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2809         kvmppc_enable_clear_ref_mod_hcalls();
2810 
2811         /* Enable H_PAGE_INIT */
2812         kvmppc_enable_h_page_init();
2813     }
2814 
2815     /* map RAM */
2816     memory_region_add_subregion(sysmem, 0, machine->ram);
2817 
2818     /* always allocate the device memory information */
2819     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2820 
2821     /* initialize hotplug memory address space */
2822     if (machine->ram_size < machine->maxram_size) {
2823         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2824         /*
2825          * Limit the number of hotpluggable memory slots to half the number
2826          * slots that KVM supports, leaving the other half for PCI and other
2827          * devices. However ensure that number of slots doesn't drop below 32.
2828          */
2829         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2830                            SPAPR_MAX_RAM_SLOTS;
2831 
2832         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2833             max_memslots = SPAPR_MAX_RAM_SLOTS;
2834         }
2835         if (machine->ram_slots > max_memslots) {
2836             error_report("Specified number of memory slots %"
2837                          PRIu64" exceeds max supported %d",
2838                          machine->ram_slots, max_memslots);
2839             exit(1);
2840         }
2841 
2842         machine->device_memory->base = ROUND_UP(machine->ram_size,
2843                                                 SPAPR_DEVICE_MEM_ALIGN);
2844         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2845                            "device-memory", device_mem_size);
2846         memory_region_add_subregion(sysmem, machine->device_memory->base,
2847                                     &machine->device_memory->mr);
2848     }
2849 
2850     if (smc->dr_lmb_enabled) {
2851         spapr_create_lmb_dr_connectors(spapr);
2852     }
2853 
2854     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2855         /* Create the error string for live migration blocker */
2856         error_setg(&spapr->fwnmi_migration_blocker,
2857             "A machine check is being handled during migration. The handler"
2858             "may run and log hardware error on the destination");
2859     }
2860 
2861     if (mc->nvdimm_supported) {
2862         spapr_create_nvdimm_dr_connectors(spapr);
2863     }
2864 
2865     /* Set up RTAS event infrastructure */
2866     spapr_events_init(spapr);
2867 
2868     /* Set up the RTC RTAS interfaces */
2869     spapr_rtc_create(spapr);
2870 
2871     /* Set up VIO bus */
2872     spapr->vio_bus = spapr_vio_bus_init();
2873 
2874     for (i = 0; serial_hd(i); i++) {
2875         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2876     }
2877 
2878     /* We always have at least the nvram device on VIO */
2879     spapr_create_nvram(spapr);
2880 
2881     /*
2882      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2883      * connectors (described in root DT node's "ibm,drc-types" property)
2884      * are pre-initialized here. additional child connectors (such as
2885      * connectors for a PHBs PCI slots) are added as needed during their
2886      * parent's realization.
2887      */
2888     if (smc->dr_phb_enabled) {
2889         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2890             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2891         }
2892     }
2893 
2894     /* Set up PCI */
2895     spapr_pci_rtas_init();
2896 
2897     phb = spapr_create_default_phb();
2898 
2899     for (i = 0; i < nb_nics; i++) {
2900         NICInfo *nd = &nd_table[i];
2901 
2902         if (!nd->model) {
2903             nd->model = g_strdup("spapr-vlan");
2904         }
2905 
2906         if (g_str_equal(nd->model, "spapr-vlan") ||
2907             g_str_equal(nd->model, "ibmveth")) {
2908             spapr_vlan_create(spapr->vio_bus, nd);
2909         } else {
2910             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2911         }
2912     }
2913 
2914     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2915         spapr_vscsi_create(spapr->vio_bus);
2916     }
2917 
2918     /* Graphics */
2919     if (spapr_vga_init(phb->bus, &error_fatal)) {
2920         spapr->has_graphics = true;
2921         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2922     }
2923 
2924     if (machine->usb) {
2925         if (smc->use_ohci_by_default) {
2926             pci_create_simple(phb->bus, -1, "pci-ohci");
2927         } else {
2928             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2929         }
2930 
2931         if (spapr->has_graphics) {
2932             USBBus *usb_bus = usb_bus_find(-1);
2933 
2934             usb_create_simple(usb_bus, "usb-kbd");
2935             usb_create_simple(usb_bus, "usb-mouse");
2936         }
2937     }
2938 
2939     if (kernel_filename) {
2940         spapr->kernel_size = load_elf(kernel_filename, NULL,
2941                                       translate_kernel_address, spapr,
2942                                       NULL, NULL, NULL, NULL, 1,
2943                                       PPC_ELF_MACHINE, 0, 0);
2944         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2945             spapr->kernel_size = load_elf(kernel_filename, NULL,
2946                                           translate_kernel_address, spapr,
2947                                           NULL, NULL, NULL, NULL, 0,
2948                                           PPC_ELF_MACHINE, 0, 0);
2949             spapr->kernel_le = spapr->kernel_size > 0;
2950         }
2951         if (spapr->kernel_size < 0) {
2952             error_report("error loading %s: %s", kernel_filename,
2953                          load_elf_strerror(spapr->kernel_size));
2954             exit(1);
2955         }
2956 
2957         /* load initrd */
2958         if (initrd_filename) {
2959             /* Try to locate the initrd in the gap between the kernel
2960              * and the firmware. Add a bit of space just in case
2961              */
2962             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2963                                   + 0x1ffff) & ~0xffff;
2964             spapr->initrd_size = load_image_targphys(initrd_filename,
2965                                                      spapr->initrd_base,
2966                                                      load_limit
2967                                                      - spapr->initrd_base);
2968             if (spapr->initrd_size < 0) {
2969                 error_report("could not load initial ram disk '%s'",
2970                              initrd_filename);
2971                 exit(1);
2972             }
2973         }
2974     }
2975 
2976     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2977     if (!filename) {
2978         error_report("Could not find LPAR firmware '%s'", bios_name);
2979         exit(1);
2980     }
2981     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2982     if (fw_size <= 0) {
2983         error_report("Could not load LPAR firmware '%s'", filename);
2984         exit(1);
2985     }
2986     g_free(filename);
2987 
2988     /* FIXME: Should register things through the MachineState's qdev
2989      * interface, this is a legacy from the sPAPREnvironment structure
2990      * which predated MachineState but had a similar function */
2991     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2992     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2993                          &savevm_htab_handlers, spapr);
2994 
2995     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2996 
2997     qemu_register_boot_set(spapr_boot_set, spapr);
2998 
2999     /*
3000      * Nothing needs to be done to resume a suspended guest because
3001      * suspending does not change the machine state, so no need for
3002      * a ->wakeup method.
3003      */
3004     qemu_register_wakeup_support();
3005 
3006     if (kvm_enabled()) {
3007         /* to stop and start vmclock */
3008         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3009                                          &spapr->tb);
3010 
3011         kvmppc_spapr_enable_inkernel_multitce();
3012     }
3013 
3014     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3015 }
3016 
3017 #define DEFAULT_KVM_TYPE "auto"
3018 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3019 {
3020     /*
3021      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3022      * accomodate the 'HV' and 'PV' formats that exists in the
3023      * wild. The 'auto' mode is being introduced already as
3024      * lower-case, thus we don't need to bother checking for
3025      * "AUTO".
3026      */
3027     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3028         return 0;
3029     }
3030 
3031     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3032         return 1;
3033     }
3034 
3035     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3036         return 2;
3037     }
3038 
3039     error_report("Unknown kvm-type specified '%s'", vm_type);
3040     exit(1);
3041 }
3042 
3043 /*
3044  * Implementation of an interface to adjust firmware path
3045  * for the bootindex property handling.
3046  */
3047 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3048                                    DeviceState *dev)
3049 {
3050 #define CAST(type, obj, name) \
3051     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3052     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3053     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3054     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3055     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
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     if (pcidev) {
3130         return spapr_pci_fw_dev_name(pcidev);
3131     }
3132 
3133     return NULL;
3134 }
3135 
3136 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3137 {
3138     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3139 
3140     return g_strdup(spapr->kvm_type);
3141 }
3142 
3143 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3144 {
3145     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3146 
3147     g_free(spapr->kvm_type);
3148     spapr->kvm_type = g_strdup(value);
3149 }
3150 
3151 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3152 {
3153     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3154 
3155     return spapr->use_hotplug_event_source;
3156 }
3157 
3158 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3159                                             Error **errp)
3160 {
3161     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3162 
3163     spapr->use_hotplug_event_source = value;
3164 }
3165 
3166 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3167 {
3168     return true;
3169 }
3170 
3171 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3172 {
3173     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3174 
3175     switch (spapr->resize_hpt) {
3176     case SPAPR_RESIZE_HPT_DEFAULT:
3177         return g_strdup("default");
3178     case SPAPR_RESIZE_HPT_DISABLED:
3179         return g_strdup("disabled");
3180     case SPAPR_RESIZE_HPT_ENABLED:
3181         return g_strdup("enabled");
3182     case SPAPR_RESIZE_HPT_REQUIRED:
3183         return g_strdup("required");
3184     }
3185     g_assert_not_reached();
3186 }
3187 
3188 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3189 {
3190     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3191 
3192     if (strcmp(value, "default") == 0) {
3193         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3194     } else if (strcmp(value, "disabled") == 0) {
3195         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3196     } else if (strcmp(value, "enabled") == 0) {
3197         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3198     } else if (strcmp(value, "required") == 0) {
3199         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3200     } else {
3201         error_setg(errp, "Bad value for \"resize-hpt\" property");
3202     }
3203 }
3204 
3205 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3206 {
3207     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3208 
3209     if (spapr->irq == &spapr_irq_xics_legacy) {
3210         return g_strdup("legacy");
3211     } else if (spapr->irq == &spapr_irq_xics) {
3212         return g_strdup("xics");
3213     } else if (spapr->irq == &spapr_irq_xive) {
3214         return g_strdup("xive");
3215     } else if (spapr->irq == &spapr_irq_dual) {
3216         return g_strdup("dual");
3217     }
3218     g_assert_not_reached();
3219 }
3220 
3221 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3222 {
3223     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3224 
3225     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3226         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3227         return;
3228     }
3229 
3230     /* The legacy IRQ backend can not be set */
3231     if (strcmp(value, "xics") == 0) {
3232         spapr->irq = &spapr_irq_xics;
3233     } else if (strcmp(value, "xive") == 0) {
3234         spapr->irq = &spapr_irq_xive;
3235     } else if (strcmp(value, "dual") == 0) {
3236         spapr->irq = &spapr_irq_dual;
3237     } else {
3238         error_setg(errp, "Bad value for \"ic-mode\" property");
3239     }
3240 }
3241 
3242 static char *spapr_get_host_model(Object *obj, Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     return g_strdup(spapr->host_model);
3247 }
3248 
3249 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3250 {
3251     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3252 
3253     g_free(spapr->host_model);
3254     spapr->host_model = g_strdup(value);
3255 }
3256 
3257 static char *spapr_get_host_serial(Object *obj, Error **errp)
3258 {
3259     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3260 
3261     return g_strdup(spapr->host_serial);
3262 }
3263 
3264 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3265 {
3266     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3267 
3268     g_free(spapr->host_serial);
3269     spapr->host_serial = g_strdup(value);
3270 }
3271 
3272 static void spapr_instance_init(Object *obj)
3273 {
3274     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3275     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3276     MachineState *ms = MACHINE(spapr);
3277     MachineClass *mc = MACHINE_GET_CLASS(ms);
3278 
3279     /*
3280      * NVDIMM support went live in 5.1 without considering that, in
3281      * other archs, the user needs to enable NVDIMM support with the
3282      * 'nvdimm' machine option and the default behavior is NVDIMM
3283      * support disabled. It is too late to roll back to the standard
3284      * behavior without breaking 5.1 guests.
3285      */
3286     if (mc->nvdimm_supported) {
3287         ms->nvdimms_state->is_enabled = true;
3288     }
3289 
3290     spapr->htab_fd = -1;
3291     spapr->use_hotplug_event_source = true;
3292     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3293     object_property_add_str(obj, "kvm-type",
3294                             spapr_get_kvm_type, spapr_set_kvm_type);
3295     object_property_set_description(obj, "kvm-type",
3296                                     "Specifies the KVM virtualization mode (auto,"
3297                                     " hv, pr). Defaults to 'auto'. This mode will use"
3298                                     " any available KVM module loaded in the host,"
3299                                     " where kvm_hv takes precedence if both kvm_hv and"
3300                                     " kvm_pr are loaded.");
3301     object_property_add_bool(obj, "modern-hotplug-events",
3302                             spapr_get_modern_hotplug_events,
3303                             spapr_set_modern_hotplug_events);
3304     object_property_set_description(obj, "modern-hotplug-events",
3305                                     "Use dedicated hotplug event mechanism in"
3306                                     " place of standard EPOW events when possible"
3307                                     " (required for memory hot-unplug support)");
3308     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3309                             "Maximum permitted CPU compatibility mode");
3310 
3311     object_property_add_str(obj, "resize-hpt",
3312                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3313     object_property_set_description(obj, "resize-hpt",
3314                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3315     object_property_add_uint32_ptr(obj, "vsmt",
3316                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3317     object_property_set_description(obj, "vsmt",
3318                                     "Virtual SMT: KVM behaves as if this were"
3319                                     " the host's SMT mode");
3320 
3321     object_property_add_bool(obj, "vfio-no-msix-emulation",
3322                              spapr_get_msix_emulation, NULL);
3323 
3324     object_property_add_uint64_ptr(obj, "kernel-addr",
3325                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3326     object_property_set_description(obj, "kernel-addr",
3327                                     stringify(KERNEL_LOAD_ADDR)
3328                                     " for -kernel is the default");
3329     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3330     /* The machine class defines the default interrupt controller mode */
3331     spapr->irq = smc->irq;
3332     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3333                             spapr_set_ic_mode);
3334     object_property_set_description(obj, "ic-mode",
3335                  "Specifies the interrupt controller mode (xics, xive, dual)");
3336 
3337     object_property_add_str(obj, "host-model",
3338         spapr_get_host_model, spapr_set_host_model);
3339     object_property_set_description(obj, "host-model",
3340         "Host model to advertise in guest device tree");
3341     object_property_add_str(obj, "host-serial",
3342         spapr_get_host_serial, spapr_set_host_serial);
3343     object_property_set_description(obj, "host-serial",
3344         "Host serial number to advertise in guest device tree");
3345 }
3346 
3347 static void spapr_machine_finalizefn(Object *obj)
3348 {
3349     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3350 
3351     g_free(spapr->kvm_type);
3352 }
3353 
3354 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3355 {
3356     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3357     PowerPCCPU *cpu = POWERPC_CPU(cs);
3358     CPUPPCState *env = &cpu->env;
3359 
3360     cpu_synchronize_state(cs);
3361     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3362     if (spapr->fwnmi_system_reset_addr != -1) {
3363         uint64_t rtas_addr, addr;
3364 
3365         /* get rtas addr from fdt */
3366         rtas_addr = spapr_get_rtas_addr();
3367         if (!rtas_addr) {
3368             qemu_system_guest_panicked(NULL);
3369             return;
3370         }
3371 
3372         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3373         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3374         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3375         env->gpr[3] = addr;
3376     }
3377     ppc_cpu_do_system_reset(cs);
3378     if (spapr->fwnmi_system_reset_addr != -1) {
3379         env->nip = spapr->fwnmi_system_reset_addr;
3380     }
3381 }
3382 
3383 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3384 {
3385     CPUState *cs;
3386 
3387     CPU_FOREACH(cs) {
3388         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3389     }
3390 }
3391 
3392 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3393                           void *fdt, int *fdt_start_offset, Error **errp)
3394 {
3395     uint64_t addr;
3396     uint32_t node;
3397 
3398     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3399     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3400                                     &error_abort);
3401     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3402                                              SPAPR_MEMORY_BLOCK_SIZE);
3403     return 0;
3404 }
3405 
3406 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3407                            bool dedicated_hp_event_source)
3408 {
3409     SpaprDrc *drc;
3410     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3411     int i;
3412     uint64_t addr = addr_start;
3413     bool hotplugged = spapr_drc_hotplugged(dev);
3414 
3415     for (i = 0; i < nr_lmbs; i++) {
3416         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3417                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3418         g_assert(drc);
3419 
3420         /*
3421          * memory_device_get_free_addr() provided a range of free addresses
3422          * that doesn't overlap with any existing mapping at pre-plug. The
3423          * corresponding LMB DRCs are thus assumed to be all attachable.
3424          */
3425         spapr_drc_attach(drc, dev);
3426         if (!hotplugged) {
3427             spapr_drc_reset(drc);
3428         }
3429         addr += SPAPR_MEMORY_BLOCK_SIZE;
3430     }
3431     /* send hotplug notification to the
3432      * guest only in case of hotplugged memory
3433      */
3434     if (hotplugged) {
3435         if (dedicated_hp_event_source) {
3436             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3437                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3438             g_assert(drc);
3439             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3440                                                    nr_lmbs,
3441                                                    spapr_drc_index(drc));
3442         } else {
3443             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3444                                            nr_lmbs);
3445         }
3446     }
3447 }
3448 
3449 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3450 {
3451     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3452     PCDIMMDevice *dimm = PC_DIMM(dev);
3453     uint64_t size, addr;
3454     int64_t slot;
3455     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3456 
3457     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3458 
3459     pc_dimm_plug(dimm, MACHINE(ms));
3460 
3461     if (!is_nvdimm) {
3462         addr = object_property_get_uint(OBJECT(dimm),
3463                                         PC_DIMM_ADDR_PROP, &error_abort);
3464         spapr_add_lmbs(dev, addr, size,
3465                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3466     } else {
3467         slot = object_property_get_int(OBJECT(dimm),
3468                                        PC_DIMM_SLOT_PROP, &error_abort);
3469         /* We should have valid slot number at this point */
3470         g_assert(slot >= 0);
3471         spapr_add_nvdimm(dev, slot);
3472     }
3473 }
3474 
3475 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3476                                   Error **errp)
3477 {
3478     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3479     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3480     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3481     PCDIMMDevice *dimm = PC_DIMM(dev);
3482     Error *local_err = NULL;
3483     uint64_t size;
3484     Object *memdev;
3485     hwaddr pagesize;
3486 
3487     if (!smc->dr_lmb_enabled) {
3488         error_setg(errp, "Memory hotplug not supported for this machine");
3489         return;
3490     }
3491 
3492     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3493     if (local_err) {
3494         error_propagate(errp, local_err);
3495         return;
3496     }
3497 
3498     if (is_nvdimm) {
3499         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3500             return;
3501         }
3502     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3503         error_setg(errp, "Hotplugged memory size must be a multiple of "
3504                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3505         return;
3506     }
3507 
3508     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3509                                       &error_abort);
3510     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3511     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3512         return;
3513     }
3514 
3515     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3516 }
3517 
3518 struct SpaprDimmState {
3519     PCDIMMDevice *dimm;
3520     uint32_t nr_lmbs;
3521     QTAILQ_ENTRY(SpaprDimmState) next;
3522 };
3523 
3524 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3525                                                        PCDIMMDevice *dimm)
3526 {
3527     SpaprDimmState *dimm_state = NULL;
3528 
3529     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3530         if (dimm_state->dimm == dimm) {
3531             break;
3532         }
3533     }
3534     return dimm_state;
3535 }
3536 
3537 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3538                                                       uint32_t nr_lmbs,
3539                                                       PCDIMMDevice *dimm)
3540 {
3541     SpaprDimmState *ds = NULL;
3542 
3543     /*
3544      * If this request is for a DIMM whose removal had failed earlier
3545      * (due to guest's refusal to remove the LMBs), we would have this
3546      * dimm already in the pending_dimm_unplugs list. In that
3547      * case don't add again.
3548      */
3549     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3550     if (!ds) {
3551         ds = g_malloc0(sizeof(SpaprDimmState));
3552         ds->nr_lmbs = nr_lmbs;
3553         ds->dimm = dimm;
3554         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3555     }
3556     return ds;
3557 }
3558 
3559 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3560                                               SpaprDimmState *dimm_state)
3561 {
3562     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3563     g_free(dimm_state);
3564 }
3565 
3566 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3567                                                         PCDIMMDevice *dimm)
3568 {
3569     SpaprDrc *drc;
3570     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3571                                                   &error_abort);
3572     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3573     uint32_t avail_lmbs = 0;
3574     uint64_t addr_start, addr;
3575     int i;
3576 
3577     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3578                                           &error_abort);
3579 
3580     addr = addr_start;
3581     for (i = 0; i < nr_lmbs; i++) {
3582         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3583                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3584         g_assert(drc);
3585         if (drc->dev) {
3586             avail_lmbs++;
3587         }
3588         addr += SPAPR_MEMORY_BLOCK_SIZE;
3589     }
3590 
3591     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3592 }
3593 
3594 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3595 {
3596     SpaprDimmState *ds;
3597     PCDIMMDevice *dimm;
3598     SpaprDrc *drc;
3599     uint32_t nr_lmbs;
3600     uint64_t size, addr_start, addr;
3601     g_autofree char *qapi_error = NULL;
3602     int i;
3603 
3604     if (!dev) {
3605         return;
3606     }
3607 
3608     dimm = PC_DIMM(dev);
3609     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3610 
3611     /*
3612      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3613      * unplug state, but one of its DRC is marked as unplug_requested.
3614      * This is bad and weird enough to g_assert() out.
3615      */
3616     g_assert(ds);
3617 
3618     spapr_pending_dimm_unplugs_remove(spapr, ds);
3619 
3620     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3621     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3622 
3623     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3624                                           &error_abort);
3625 
3626     addr = addr_start;
3627     for (i = 0; i < nr_lmbs; i++) {
3628         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3629                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3630         g_assert(drc);
3631 
3632         drc->unplug_requested = false;
3633         addr += SPAPR_MEMORY_BLOCK_SIZE;
3634     }
3635 
3636     /*
3637      * Tell QAPI that something happened and the memory
3638      * hotunplug wasn't successful.
3639      */
3640     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3641                                  "for device %s", dev->id);
3642     qapi_event_send_mem_unplug_error(dev->id, qapi_error);
3643 }
3644 
3645 /* Callback to be called during DRC release. */
3646 void spapr_lmb_release(DeviceState *dev)
3647 {
3648     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3649     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3650     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3651 
3652     /* This information will get lost if a migration occurs
3653      * during the unplug process. In this case recover it. */
3654     if (ds == NULL) {
3655         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3656         g_assert(ds);
3657         /* The DRC being examined by the caller at least must be counted */
3658         g_assert(ds->nr_lmbs);
3659     }
3660 
3661     if (--ds->nr_lmbs) {
3662         return;
3663     }
3664 
3665     /*
3666      * Now that all the LMBs have been removed by the guest, call the
3667      * unplug handler chain. This can never fail.
3668      */
3669     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3670     object_unparent(OBJECT(dev));
3671 }
3672 
3673 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3674 {
3675     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3676     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3677 
3678     /* We really shouldn't get this far without anything to unplug */
3679     g_assert(ds);
3680 
3681     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3682     qdev_unrealize(dev);
3683     spapr_pending_dimm_unplugs_remove(spapr, ds);
3684 }
3685 
3686 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3687                                         DeviceState *dev, Error **errp)
3688 {
3689     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3690     PCDIMMDevice *dimm = PC_DIMM(dev);
3691     uint32_t nr_lmbs;
3692     uint64_t size, addr_start, addr;
3693     int i;
3694     SpaprDrc *drc;
3695 
3696     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3697         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3698         return;
3699     }
3700 
3701     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3702     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3703 
3704     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3705                                           &error_abort);
3706 
3707     /*
3708      * An existing pending dimm state for this DIMM means that there is an
3709      * unplug operation in progress, waiting for the spapr_lmb_release
3710      * callback to complete the job (BQL can't cover that far). In this case,
3711      * bail out to avoid detaching DRCs that were already released.
3712      */
3713     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3714         error_setg(errp, "Memory unplug already in progress for device %s",
3715                    dev->id);
3716         return;
3717     }
3718 
3719     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3720 
3721     addr = addr_start;
3722     for (i = 0; i < nr_lmbs; i++) {
3723         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3724                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3725         g_assert(drc);
3726 
3727         spapr_drc_unplug_request(drc);
3728         addr += SPAPR_MEMORY_BLOCK_SIZE;
3729     }
3730 
3731     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3732                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3733     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3734                                               nr_lmbs, spapr_drc_index(drc));
3735 }
3736 
3737 /* Callback to be called during DRC release. */
3738 void spapr_core_release(DeviceState *dev)
3739 {
3740     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3741 
3742     /* Call the unplug handler chain. This can never fail. */
3743     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3744     object_unparent(OBJECT(dev));
3745 }
3746 
3747 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3748 {
3749     MachineState *ms = MACHINE(hotplug_dev);
3750     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3751     CPUCore *cc = CPU_CORE(dev);
3752     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3753 
3754     if (smc->pre_2_10_has_unused_icps) {
3755         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3756         int i;
3757 
3758         for (i = 0; i < cc->nr_threads; i++) {
3759             CPUState *cs = CPU(sc->threads[i]);
3760 
3761             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3762         }
3763     }
3764 
3765     assert(core_slot);
3766     core_slot->cpu = NULL;
3767     qdev_unrealize(dev);
3768 }
3769 
3770 static
3771 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3772                                Error **errp)
3773 {
3774     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3775     int index;
3776     SpaprDrc *drc;
3777     CPUCore *cc = CPU_CORE(dev);
3778 
3779     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3780         error_setg(errp, "Unable to find CPU core with core-id: %d",
3781                    cc->core_id);
3782         return;
3783     }
3784     if (index == 0) {
3785         error_setg(errp, "Boot CPU core may not be unplugged");
3786         return;
3787     }
3788 
3789     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3790                           spapr_vcpu_id(spapr, cc->core_id));
3791     g_assert(drc);
3792 
3793     if (!spapr_drc_unplug_requested(drc)) {
3794         spapr_drc_unplug_request(drc);
3795     }
3796 
3797     /*
3798      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3799      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3800      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3801      * attempt (e.g. the kernel will refuse to remove the last online
3802      * CPU), we will never attempt it again because unplug_requested
3803      * will still be 'true' in that case.
3804      */
3805     spapr_hotplug_req_remove_by_index(drc);
3806 }
3807 
3808 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3809                            void *fdt, int *fdt_start_offset, Error **errp)
3810 {
3811     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3812     CPUState *cs = CPU(core->threads[0]);
3813     PowerPCCPU *cpu = POWERPC_CPU(cs);
3814     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3815     int id = spapr_get_vcpu_id(cpu);
3816     g_autofree char *nodename = NULL;
3817     int offset;
3818 
3819     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3820     offset = fdt_add_subnode(fdt, 0, nodename);
3821 
3822     spapr_dt_cpu(cs, fdt, offset, spapr);
3823 
3824     /*
3825      * spapr_dt_cpu() does not fill the 'name' property in the
3826      * CPU node. The function is called during boot process, before
3827      * and after CAS, and overwriting the 'name' property written
3828      * by SLOF is not allowed.
3829      *
3830      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3831      * CPUs more compatible with the coldplugged ones, which have
3832      * the 'name' property. Linux Kernel also relies on this
3833      * property to identify CPU nodes.
3834      */
3835     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3836 
3837     *fdt_start_offset = offset;
3838     return 0;
3839 }
3840 
3841 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3842 {
3843     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3844     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3845     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3846     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3847     CPUCore *cc = CPU_CORE(dev);
3848     CPUState *cs;
3849     SpaprDrc *drc;
3850     CPUArchId *core_slot;
3851     int index;
3852     bool hotplugged = spapr_drc_hotplugged(dev);
3853     int i;
3854 
3855     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3856     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3857 
3858     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3859                           spapr_vcpu_id(spapr, cc->core_id));
3860 
3861     g_assert(drc || !mc->has_hotpluggable_cpus);
3862 
3863     if (drc) {
3864         /*
3865          * spapr_core_pre_plug() already buys us this is a brand new
3866          * core being plugged into a free slot. Nothing should already
3867          * be attached to the corresponding DRC.
3868          */
3869         spapr_drc_attach(drc, dev);
3870 
3871         if (hotplugged) {
3872             /*
3873              * Send hotplug notification interrupt to the guest only
3874              * in case of hotplugged CPUs.
3875              */
3876             spapr_hotplug_req_add_by_index(drc);
3877         } else {
3878             spapr_drc_reset(drc);
3879         }
3880     }
3881 
3882     core_slot->cpu = OBJECT(dev);
3883 
3884     /*
3885      * Set compatibility mode to match the boot CPU, which was either set
3886      * by the machine reset code or by CAS. This really shouldn't fail at
3887      * this point.
3888      */
3889     if (hotplugged) {
3890         for (i = 0; i < cc->nr_threads; i++) {
3891             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3892                            &error_abort);
3893         }
3894     }
3895 
3896     if (smc->pre_2_10_has_unused_icps) {
3897         for (i = 0; i < cc->nr_threads; i++) {
3898             cs = CPU(core->threads[i]);
3899             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3900         }
3901     }
3902 }
3903 
3904 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3905                                 Error **errp)
3906 {
3907     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3908     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3909     CPUCore *cc = CPU_CORE(dev);
3910     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3911     const char *type = object_get_typename(OBJECT(dev));
3912     CPUArchId *core_slot;
3913     int index;
3914     unsigned int smp_threads = machine->smp.threads;
3915 
3916     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3917         error_setg(errp, "CPU hotplug not supported for this machine");
3918         return;
3919     }
3920 
3921     if (strcmp(base_core_type, type)) {
3922         error_setg(errp, "CPU core type should be %s", base_core_type);
3923         return;
3924     }
3925 
3926     if (cc->core_id % smp_threads) {
3927         error_setg(errp, "invalid core id %d", cc->core_id);
3928         return;
3929     }
3930 
3931     /*
3932      * In general we should have homogeneous threads-per-core, but old
3933      * (pre hotplug support) machine types allow the last core to have
3934      * reduced threads as a compatibility hack for when we allowed
3935      * total vcpus not a multiple of threads-per-core.
3936      */
3937     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3938         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3939                    smp_threads);
3940         return;
3941     }
3942 
3943     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3944     if (!core_slot) {
3945         error_setg(errp, "core id %d out of range", cc->core_id);
3946         return;
3947     }
3948 
3949     if (core_slot->cpu) {
3950         error_setg(errp, "core %d already populated", cc->core_id);
3951         return;
3952     }
3953 
3954     numa_cpu_pre_plug(core_slot, dev, errp);
3955 }
3956 
3957 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3958                           void *fdt, int *fdt_start_offset, Error **errp)
3959 {
3960     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3961     int intc_phandle;
3962 
3963     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3964     if (intc_phandle <= 0) {
3965         return -1;
3966     }
3967 
3968     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3969         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3970         return -1;
3971     }
3972 
3973     /* generally SLOF creates these, for hotplug it's up to QEMU */
3974     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3975 
3976     return 0;
3977 }
3978 
3979 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3980                                Error **errp)
3981 {
3982     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3983     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3984     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3985     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3986     SpaprDrc *drc;
3987 
3988     if (dev->hotplugged && !smc->dr_phb_enabled) {
3989         error_setg(errp, "PHB hotplug not supported for this machine");
3990         return false;
3991     }
3992 
3993     if (sphb->index == (uint32_t)-1) {
3994         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3995         return false;
3996     }
3997 
3998     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3999     if (drc && drc->dev) {
4000         error_setg(errp, "PHB %d already attached", sphb->index);
4001         return false;
4002     }
4003 
4004     /*
4005      * This will check that sphb->index doesn't exceed the maximum number of
4006      * PHBs for the current machine type.
4007      */
4008     return
4009         smc->phb_placement(spapr, sphb->index,
4010                            &sphb->buid, &sphb->io_win_addr,
4011                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4012                            windows_supported, sphb->dma_liobn,
4013                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4014                            errp);
4015 }
4016 
4017 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4018 {
4019     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4020     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4021     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4022     SpaprDrc *drc;
4023     bool hotplugged = spapr_drc_hotplugged(dev);
4024 
4025     if (!smc->dr_phb_enabled) {
4026         return;
4027     }
4028 
4029     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4030     /* hotplug hooks should check it's enabled before getting this far */
4031     assert(drc);
4032 
4033     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4034     spapr_drc_attach(drc, dev);
4035 
4036     if (hotplugged) {
4037         spapr_hotplug_req_add_by_index(drc);
4038     } else {
4039         spapr_drc_reset(drc);
4040     }
4041 }
4042 
4043 void spapr_phb_release(DeviceState *dev)
4044 {
4045     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4046 
4047     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4048     object_unparent(OBJECT(dev));
4049 }
4050 
4051 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4052 {
4053     qdev_unrealize(dev);
4054 }
4055 
4056 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4057                                      DeviceState *dev, Error **errp)
4058 {
4059     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4060     SpaprDrc *drc;
4061 
4062     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4063     assert(drc);
4064 
4065     if (!spapr_drc_unplug_requested(drc)) {
4066         spapr_drc_unplug_request(drc);
4067         spapr_hotplug_req_remove_by_index(drc);
4068     } else {
4069         error_setg(errp,
4070                    "PCI Host Bridge unplug already in progress for device %s",
4071                    dev->id);
4072     }
4073 }
4074 
4075 static
4076 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4077                               Error **errp)
4078 {
4079     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4080 
4081     if (spapr->tpm_proxy != NULL) {
4082         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4083         return false;
4084     }
4085 
4086     return true;
4087 }
4088 
4089 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4090 {
4091     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4092     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4093 
4094     /* Already checked in spapr_tpm_proxy_pre_plug() */
4095     g_assert(spapr->tpm_proxy == NULL);
4096 
4097     spapr->tpm_proxy = tpm_proxy;
4098 }
4099 
4100 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4101 {
4102     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4103 
4104     qdev_unrealize(dev);
4105     object_unparent(OBJECT(dev));
4106     spapr->tpm_proxy = NULL;
4107 }
4108 
4109 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4110                                       DeviceState *dev, Error **errp)
4111 {
4112     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4113         spapr_memory_plug(hotplug_dev, dev);
4114     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4115         spapr_core_plug(hotplug_dev, dev);
4116     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4117         spapr_phb_plug(hotplug_dev, dev);
4118     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4119         spapr_tpm_proxy_plug(hotplug_dev, dev);
4120     }
4121 }
4122 
4123 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4124                                         DeviceState *dev, Error **errp)
4125 {
4126     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4127         spapr_memory_unplug(hotplug_dev, dev);
4128     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4129         spapr_core_unplug(hotplug_dev, dev);
4130     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4131         spapr_phb_unplug(hotplug_dev, dev);
4132     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4133         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4134     }
4135 }
4136 
4137 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4138 {
4139     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4140         /*
4141          * CAS will process all pending unplug requests.
4142          *
4143          * HACK: a guest could theoretically have cleared all bits in OV5,
4144          * but none of the guests we care for do.
4145          */
4146         spapr_ovec_empty(spapr->ov5_cas);
4147 }
4148 
4149 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4150                                                 DeviceState *dev, Error **errp)
4151 {
4152     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4153     MachineClass *mc = MACHINE_GET_CLASS(sms);
4154     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4155 
4156     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4157         if (spapr_memory_hot_unplug_supported(sms)) {
4158             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4159         } else {
4160             error_setg(errp, "Memory hot unplug not supported for this guest");
4161         }
4162     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4163         if (!mc->has_hotpluggable_cpus) {
4164             error_setg(errp, "CPU hot unplug not supported on this machine");
4165             return;
4166         }
4167         spapr_core_unplug_request(hotplug_dev, dev, errp);
4168     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4169         if (!smc->dr_phb_enabled) {
4170             error_setg(errp, "PHB hot unplug not supported on this machine");
4171             return;
4172         }
4173         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4174     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4175         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4176     }
4177 }
4178 
4179 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4180                                           DeviceState *dev, Error **errp)
4181 {
4182     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4183         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4184     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4185         spapr_core_pre_plug(hotplug_dev, dev, errp);
4186     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4187         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4188     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4189         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4190     }
4191 }
4192 
4193 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4194                                                  DeviceState *dev)
4195 {
4196     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4197         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4198         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4199         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4200         return HOTPLUG_HANDLER(machine);
4201     }
4202     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4203         PCIDevice *pcidev = PCI_DEVICE(dev);
4204         PCIBus *root = pci_device_root_bus(pcidev);
4205         SpaprPhbState *phb =
4206             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4207                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4208 
4209         if (phb) {
4210             return HOTPLUG_HANDLER(phb);
4211         }
4212     }
4213     return NULL;
4214 }
4215 
4216 static CpuInstanceProperties
4217 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4218 {
4219     CPUArchId *core_slot;
4220     MachineClass *mc = MACHINE_GET_CLASS(machine);
4221 
4222     /* make sure possible_cpu are intialized */
4223     mc->possible_cpu_arch_ids(machine);
4224     /* get CPU core slot containing thread that matches cpu_index */
4225     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4226     assert(core_slot);
4227     return core_slot->props;
4228 }
4229 
4230 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4231 {
4232     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4233 }
4234 
4235 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4236 {
4237     int i;
4238     unsigned int smp_threads = machine->smp.threads;
4239     unsigned int smp_cpus = machine->smp.cpus;
4240     const char *core_type;
4241     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4242     MachineClass *mc = MACHINE_GET_CLASS(machine);
4243 
4244     if (!mc->has_hotpluggable_cpus) {
4245         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4246     }
4247     if (machine->possible_cpus) {
4248         assert(machine->possible_cpus->len == spapr_max_cores);
4249         return machine->possible_cpus;
4250     }
4251 
4252     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4253     if (!core_type) {
4254         error_report("Unable to find sPAPR CPU Core definition");
4255         exit(1);
4256     }
4257 
4258     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4259                              sizeof(CPUArchId) * spapr_max_cores);
4260     machine->possible_cpus->len = spapr_max_cores;
4261     for (i = 0; i < machine->possible_cpus->len; i++) {
4262         int core_id = i * smp_threads;
4263 
4264         machine->possible_cpus->cpus[i].type = core_type;
4265         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4266         machine->possible_cpus->cpus[i].arch_id = core_id;
4267         machine->possible_cpus->cpus[i].props.has_core_id = true;
4268         machine->possible_cpus->cpus[i].props.core_id = core_id;
4269     }
4270     return machine->possible_cpus;
4271 }
4272 
4273 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4274                                 uint64_t *buid, hwaddr *pio,
4275                                 hwaddr *mmio32, hwaddr *mmio64,
4276                                 unsigned n_dma, uint32_t *liobns,
4277                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4278 {
4279     /*
4280      * New-style PHB window placement.
4281      *
4282      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4283      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4284      * windows.
4285      *
4286      * Some guest kernels can't work with MMIO windows above 1<<46
4287      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4288      *
4289      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4290      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4291      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4292      * 1TiB 64-bit MMIO windows for each PHB.
4293      */
4294     const uint64_t base_buid = 0x800000020000000ULL;
4295     int i;
4296 
4297     /* Sanity check natural alignments */
4298     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4299     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4300     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4301     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4302     /* Sanity check bounds */
4303     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4304                       SPAPR_PCI_MEM32_WIN_SIZE);
4305     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4306                       SPAPR_PCI_MEM64_WIN_SIZE);
4307 
4308     if (index >= SPAPR_MAX_PHBS) {
4309         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4310                    SPAPR_MAX_PHBS - 1);
4311         return false;
4312     }
4313 
4314     *buid = base_buid + index;
4315     for (i = 0; i < n_dma; ++i) {
4316         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4317     }
4318 
4319     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4320     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4321     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4322 
4323     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4324     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4325     return true;
4326 }
4327 
4328 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4329 {
4330     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4331 
4332     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4333 }
4334 
4335 static void spapr_ics_resend(XICSFabric *dev)
4336 {
4337     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4338 
4339     ics_resend(spapr->ics);
4340 }
4341 
4342 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4343 {
4344     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4345 
4346     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4347 }
4348 
4349 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4350                                  Monitor *mon)
4351 {
4352     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4353 
4354     spapr_irq_print_info(spapr, mon);
4355     monitor_printf(mon, "irqchip: %s\n",
4356                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4357 }
4358 
4359 /*
4360  * This is a XIVE only operation
4361  */
4362 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4363                            uint8_t nvt_blk, uint32_t nvt_idx,
4364                            bool cam_ignore, uint8_t priority,
4365                            uint32_t logic_serv, XiveTCTXMatch *match)
4366 {
4367     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4368     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4369     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4370     int count;
4371 
4372     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4373                            priority, logic_serv, match);
4374     if (count < 0) {
4375         return count;
4376     }
4377 
4378     /*
4379      * When we implement the save and restore of the thread interrupt
4380      * contexts in the enter/exit CPU handlers of the machine and the
4381      * escalations in QEMU, we should be able to handle non dispatched
4382      * vCPUs.
4383      *
4384      * Until this is done, the sPAPR machine should find at least one
4385      * matching context always.
4386      */
4387     if (count == 0) {
4388         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4389                       nvt_blk, nvt_idx);
4390     }
4391 
4392     return count;
4393 }
4394 
4395 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4396 {
4397     return cpu->vcpu_id;
4398 }
4399 
4400 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4401 {
4402     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4403     MachineState *ms = MACHINE(spapr);
4404     int vcpu_id;
4405 
4406     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4407 
4408     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4409         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4410         error_append_hint(errp, "Adjust the number of cpus to %d "
4411                           "or try to raise the number of threads per core\n",
4412                           vcpu_id * ms->smp.threads / spapr->vsmt);
4413         return false;
4414     }
4415 
4416     cpu->vcpu_id = vcpu_id;
4417     return true;
4418 }
4419 
4420 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4421 {
4422     CPUState *cs;
4423 
4424     CPU_FOREACH(cs) {
4425         PowerPCCPU *cpu = POWERPC_CPU(cs);
4426 
4427         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4428             return cpu;
4429         }
4430     }
4431 
4432     return NULL;
4433 }
4434 
4435 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4436 {
4437     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4438 
4439     /* These are only called by TCG, KVM maintains dispatch state */
4440 
4441     spapr_cpu->prod = false;
4442     if (spapr_cpu->vpa_addr) {
4443         CPUState *cs = CPU(cpu);
4444         uint32_t dispatch;
4445 
4446         dispatch = ldl_be_phys(cs->as,
4447                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4448         dispatch++;
4449         if ((dispatch & 1) != 0) {
4450             qemu_log_mask(LOG_GUEST_ERROR,
4451                           "VPA: incorrect dispatch counter value for "
4452                           "dispatched partition %u, correcting.\n", dispatch);
4453             dispatch++;
4454         }
4455         stl_be_phys(cs->as,
4456                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4457     }
4458 }
4459 
4460 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4461 {
4462     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4463 
4464     if (spapr_cpu->vpa_addr) {
4465         CPUState *cs = CPU(cpu);
4466         uint32_t dispatch;
4467 
4468         dispatch = ldl_be_phys(cs->as,
4469                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4470         dispatch++;
4471         if ((dispatch & 1) != 1) {
4472             qemu_log_mask(LOG_GUEST_ERROR,
4473                           "VPA: incorrect dispatch counter value for "
4474                           "preempted partition %u, correcting.\n", dispatch);
4475             dispatch++;
4476         }
4477         stl_be_phys(cs->as,
4478                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4479     }
4480 }
4481 
4482 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4483 {
4484     MachineClass *mc = MACHINE_CLASS(oc);
4485     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4486     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4487     NMIClass *nc = NMI_CLASS(oc);
4488     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4489     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4490     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4491     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4492     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4493 
4494     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4495     mc->ignore_boot_device_suffixes = true;
4496 
4497     /*
4498      * We set up the default / latest behaviour here.  The class_init
4499      * functions for the specific versioned machine types can override
4500      * these details for backwards compatibility
4501      */
4502     mc->init = spapr_machine_init;
4503     mc->reset = spapr_machine_reset;
4504     mc->block_default_type = IF_SCSI;
4505 
4506     /*
4507      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4508      * should be limited by the host capability instead of hardcoded.
4509      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4510      * guests are welcome to have as many CPUs as the host are capable
4511      * of emulate.
4512      */
4513     mc->max_cpus = INT32_MAX;
4514 
4515     mc->no_parallel = 1;
4516     mc->default_boot_order = "";
4517     mc->default_ram_size = 512 * MiB;
4518     mc->default_ram_id = "ppc_spapr.ram";
4519     mc->default_display = "std";
4520     mc->kvm_type = spapr_kvm_type;
4521     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4522     mc->pci_allow_0_address = true;
4523     assert(!mc->get_hotplug_handler);
4524     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4525     hc->pre_plug = spapr_machine_device_pre_plug;
4526     hc->plug = spapr_machine_device_plug;
4527     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4528     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4529     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4530     hc->unplug_request = spapr_machine_device_unplug_request;
4531     hc->unplug = spapr_machine_device_unplug;
4532 
4533     smc->dr_lmb_enabled = true;
4534     smc->update_dt_enabled = true;
4535     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4536     mc->has_hotpluggable_cpus = true;
4537     mc->nvdimm_supported = true;
4538     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4539     fwc->get_dev_path = spapr_get_fw_dev_path;
4540     nc->nmi_monitor_handler = spapr_nmi;
4541     smc->phb_placement = spapr_phb_placement;
4542     vhc->hypercall = emulate_spapr_hypercall;
4543     vhc->hpt_mask = spapr_hpt_mask;
4544     vhc->map_hptes = spapr_map_hptes;
4545     vhc->unmap_hptes = spapr_unmap_hptes;
4546     vhc->hpte_set_c = spapr_hpte_set_c;
4547     vhc->hpte_set_r = spapr_hpte_set_r;
4548     vhc->get_pate = spapr_get_pate;
4549     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4550     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4551     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4552     xic->ics_get = spapr_ics_get;
4553     xic->ics_resend = spapr_ics_resend;
4554     xic->icp_get = spapr_icp_get;
4555     ispc->print_info = spapr_pic_print_info;
4556     /* Force NUMA node memory size to be a multiple of
4557      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4558      * in which LMBs are represented and hot-added
4559      */
4560     mc->numa_mem_align_shift = 28;
4561     mc->auto_enable_numa = true;
4562 
4563     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4564     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4565     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4566     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4567     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4568     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4569     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4570     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4571     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4572     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4573     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4574     spapr_caps_add_properties(smc);
4575     smc->irq = &spapr_irq_dual;
4576     smc->dr_phb_enabled = true;
4577     smc->linux_pci_probe = true;
4578     smc->smp_threads_vsmt = true;
4579     smc->nr_xirqs = SPAPR_NR_XIRQS;
4580     xfc->match_nvt = spapr_match_nvt;
4581 }
4582 
4583 static const TypeInfo spapr_machine_info = {
4584     .name          = TYPE_SPAPR_MACHINE,
4585     .parent        = TYPE_MACHINE,
4586     .abstract      = true,
4587     .instance_size = sizeof(SpaprMachineState),
4588     .instance_init = spapr_instance_init,
4589     .instance_finalize = spapr_machine_finalizefn,
4590     .class_size    = sizeof(SpaprMachineClass),
4591     .class_init    = spapr_machine_class_init,
4592     .interfaces = (InterfaceInfo[]) {
4593         { TYPE_FW_PATH_PROVIDER },
4594         { TYPE_NMI },
4595         { TYPE_HOTPLUG_HANDLER },
4596         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4597         { TYPE_XICS_FABRIC },
4598         { TYPE_INTERRUPT_STATS_PROVIDER },
4599         { TYPE_XIVE_FABRIC },
4600         { }
4601     },
4602 };
4603 
4604 static void spapr_machine_latest_class_options(MachineClass *mc)
4605 {
4606     mc->alias = "pseries";
4607     mc->is_default = true;
4608 }
4609 
4610 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4611     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4612                                                     void *data)      \
4613     {                                                                \
4614         MachineClass *mc = MACHINE_CLASS(oc);                        \
4615         spapr_machine_##suffix##_class_options(mc);                  \
4616         if (latest) {                                                \
4617             spapr_machine_latest_class_options(mc);                  \
4618         }                                                            \
4619     }                                                                \
4620     static const TypeInfo spapr_machine_##suffix##_info = {          \
4621         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4622         .parent = TYPE_SPAPR_MACHINE,                                \
4623         .class_init = spapr_machine_##suffix##_class_init,           \
4624     };                                                               \
4625     static void spapr_machine_register_##suffix(void)                \
4626     {                                                                \
4627         type_register(&spapr_machine_##suffix##_info);               \
4628     }                                                                \
4629     type_init(spapr_machine_register_##suffix)
4630 
4631 /*
4632  * pseries-6.1
4633  */
4634 static void spapr_machine_6_1_class_options(MachineClass *mc)
4635 {
4636     /* Defaults for the latest behaviour inherited from the base class */
4637 }
4638 
4639 DEFINE_SPAPR_MACHINE(6_1, "6.1", true);
4640 
4641 /*
4642  * pseries-6.0
4643  */
4644 static void spapr_machine_6_0_class_options(MachineClass *mc)
4645 {
4646     spapr_machine_6_1_class_options(mc);
4647     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4648 }
4649 
4650 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4651 
4652 /*
4653  * pseries-5.2
4654  */
4655 static void spapr_machine_5_2_class_options(MachineClass *mc)
4656 {
4657     spapr_machine_6_0_class_options(mc);
4658     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4659 }
4660 
4661 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4662 
4663 /*
4664  * pseries-5.1
4665  */
4666 static void spapr_machine_5_1_class_options(MachineClass *mc)
4667 {
4668     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4669 
4670     spapr_machine_5_2_class_options(mc);
4671     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4672     smc->pre_5_2_numa_associativity = true;
4673 }
4674 
4675 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4676 
4677 /*
4678  * pseries-5.0
4679  */
4680 static void spapr_machine_5_0_class_options(MachineClass *mc)
4681 {
4682     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4683     static GlobalProperty compat[] = {
4684         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4685     };
4686 
4687     spapr_machine_5_1_class_options(mc);
4688     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4689     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4690     mc->numa_mem_supported = true;
4691     smc->pre_5_1_assoc_refpoints = true;
4692 }
4693 
4694 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4695 
4696 /*
4697  * pseries-4.2
4698  */
4699 static void spapr_machine_4_2_class_options(MachineClass *mc)
4700 {
4701     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4702 
4703     spapr_machine_5_0_class_options(mc);
4704     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4705     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4706     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4707     smc->rma_limit = 16 * GiB;
4708     mc->nvdimm_supported = false;
4709 }
4710 
4711 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4712 
4713 /*
4714  * pseries-4.1
4715  */
4716 static void spapr_machine_4_1_class_options(MachineClass *mc)
4717 {
4718     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4719     static GlobalProperty compat[] = {
4720         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4721         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4722     };
4723 
4724     spapr_machine_4_2_class_options(mc);
4725     smc->linux_pci_probe = false;
4726     smc->smp_threads_vsmt = false;
4727     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4728     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4729 }
4730 
4731 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4732 
4733 /*
4734  * pseries-4.0
4735  */
4736 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4737                               uint64_t *buid, hwaddr *pio,
4738                               hwaddr *mmio32, hwaddr *mmio64,
4739                               unsigned n_dma, uint32_t *liobns,
4740                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4741 {
4742     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4743                              liobns, nv2gpa, nv2atsd, errp)) {
4744         return false;
4745     }
4746 
4747     *nv2gpa = 0;
4748     *nv2atsd = 0;
4749     return true;
4750 }
4751 static void spapr_machine_4_0_class_options(MachineClass *mc)
4752 {
4753     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4754 
4755     spapr_machine_4_1_class_options(mc);
4756     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4757     smc->phb_placement = phb_placement_4_0;
4758     smc->irq = &spapr_irq_xics;
4759     smc->pre_4_1_migration = true;
4760 }
4761 
4762 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4763 
4764 /*
4765  * pseries-3.1
4766  */
4767 static void spapr_machine_3_1_class_options(MachineClass *mc)
4768 {
4769     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4770 
4771     spapr_machine_4_0_class_options(mc);
4772     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4773 
4774     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4775     smc->update_dt_enabled = false;
4776     smc->dr_phb_enabled = false;
4777     smc->broken_host_serial_model = true;
4778     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4779     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4780     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4781     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4785 
4786 /*
4787  * pseries-3.0
4788  */
4789 
4790 static void spapr_machine_3_0_class_options(MachineClass *mc)
4791 {
4792     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4793 
4794     spapr_machine_3_1_class_options(mc);
4795     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4796 
4797     smc->legacy_irq_allocation = true;
4798     smc->nr_xirqs = 0x400;
4799     smc->irq = &spapr_irq_xics_legacy;
4800 }
4801 
4802 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4803 
4804 /*
4805  * pseries-2.12
4806  */
4807 static void spapr_machine_2_12_class_options(MachineClass *mc)
4808 {
4809     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4810     static GlobalProperty compat[] = {
4811         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4812         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4813     };
4814 
4815     spapr_machine_3_0_class_options(mc);
4816     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4817     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4818 
4819     /* We depend on kvm_enabled() to choose a default value for the
4820      * hpt-max-page-size capability. Of course we can't do it here
4821      * because this is too early and the HW accelerator isn't initialzed
4822      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4823      */
4824     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4825 }
4826 
4827 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4828 
4829 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4830 {
4831     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4832 
4833     spapr_machine_2_12_class_options(mc);
4834     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4835     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4836     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4837 }
4838 
4839 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4840 
4841 /*
4842  * pseries-2.11
4843  */
4844 
4845 static void spapr_machine_2_11_class_options(MachineClass *mc)
4846 {
4847     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4848 
4849     spapr_machine_2_12_class_options(mc);
4850     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4851     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4852 }
4853 
4854 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4855 
4856 /*
4857  * pseries-2.10
4858  */
4859 
4860 static void spapr_machine_2_10_class_options(MachineClass *mc)
4861 {
4862     spapr_machine_2_11_class_options(mc);
4863     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4864 }
4865 
4866 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4867 
4868 /*
4869  * pseries-2.9
4870  */
4871 
4872 static void spapr_machine_2_9_class_options(MachineClass *mc)
4873 {
4874     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4875     static GlobalProperty compat[] = {
4876         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4877     };
4878 
4879     spapr_machine_2_10_class_options(mc);
4880     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4881     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4882     smc->pre_2_10_has_unused_icps = true;
4883     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4884 }
4885 
4886 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4887 
4888 /*
4889  * pseries-2.8
4890  */
4891 
4892 static void spapr_machine_2_8_class_options(MachineClass *mc)
4893 {
4894     static GlobalProperty compat[] = {
4895         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4896     };
4897 
4898     spapr_machine_2_9_class_options(mc);
4899     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4900     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4901     mc->numa_mem_align_shift = 23;
4902 }
4903 
4904 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4905 
4906 /*
4907  * pseries-2.7
4908  */
4909 
4910 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4911                               uint64_t *buid, hwaddr *pio,
4912                               hwaddr *mmio32, hwaddr *mmio64,
4913                               unsigned n_dma, uint32_t *liobns,
4914                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4915 {
4916     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4917     const uint64_t base_buid = 0x800000020000000ULL;
4918     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4919     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4920     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4921     const uint32_t max_index = 255;
4922     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4923 
4924     uint64_t ram_top = MACHINE(spapr)->ram_size;
4925     hwaddr phb0_base, phb_base;
4926     int i;
4927 
4928     /* Do we have device memory? */
4929     if (MACHINE(spapr)->maxram_size > ram_top) {
4930         /* Can't just use maxram_size, because there may be an
4931          * alignment gap between normal and device memory regions
4932          */
4933         ram_top = MACHINE(spapr)->device_memory->base +
4934             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4935     }
4936 
4937     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4938 
4939     if (index > max_index) {
4940         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4941                    max_index);
4942         return false;
4943     }
4944 
4945     *buid = base_buid + index;
4946     for (i = 0; i < n_dma; ++i) {
4947         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4948     }
4949 
4950     phb_base = phb0_base + index * phb_spacing;
4951     *pio = phb_base + pio_offset;
4952     *mmio32 = phb_base + mmio_offset;
4953     /*
4954      * We don't set the 64-bit MMIO window, relying on the PHB's
4955      * fallback behaviour of automatically splitting a large "32-bit"
4956      * window into contiguous 32-bit and 64-bit windows
4957      */
4958 
4959     *nv2gpa = 0;
4960     *nv2atsd = 0;
4961     return true;
4962 }
4963 
4964 static void spapr_machine_2_7_class_options(MachineClass *mc)
4965 {
4966     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4967     static GlobalProperty compat[] = {
4968         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4969         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4970         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4971         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4972     };
4973 
4974     spapr_machine_2_8_class_options(mc);
4975     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4976     mc->default_machine_opts = "modern-hotplug-events=off";
4977     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4978     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4979     smc->phb_placement = phb_placement_2_7;
4980 }
4981 
4982 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4983 
4984 /*
4985  * pseries-2.6
4986  */
4987 
4988 static void spapr_machine_2_6_class_options(MachineClass *mc)
4989 {
4990     static GlobalProperty compat[] = {
4991         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4992     };
4993 
4994     spapr_machine_2_7_class_options(mc);
4995     mc->has_hotpluggable_cpus = false;
4996     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4997     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4998 }
4999 
5000 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5001 
5002 /*
5003  * pseries-2.5
5004  */
5005 
5006 static void spapr_machine_2_5_class_options(MachineClass *mc)
5007 {
5008     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5009     static GlobalProperty compat[] = {
5010         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5011     };
5012 
5013     spapr_machine_2_6_class_options(mc);
5014     smc->use_ohci_by_default = true;
5015     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5016     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5017 }
5018 
5019 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5020 
5021 /*
5022  * pseries-2.4
5023  */
5024 
5025 static void spapr_machine_2_4_class_options(MachineClass *mc)
5026 {
5027     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5028 
5029     spapr_machine_2_5_class_options(mc);
5030     smc->dr_lmb_enabled = false;
5031     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5032 }
5033 
5034 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5035 
5036 /*
5037  * pseries-2.3
5038  */
5039 
5040 static void spapr_machine_2_3_class_options(MachineClass *mc)
5041 {
5042     static GlobalProperty compat[] = {
5043         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5044     };
5045     spapr_machine_2_4_class_options(mc);
5046     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5047     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5048 }
5049 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5050 
5051 /*
5052  * pseries-2.2
5053  */
5054 
5055 static void spapr_machine_2_2_class_options(MachineClass *mc)
5056 {
5057     static GlobalProperty compat[] = {
5058         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5059     };
5060 
5061     spapr_machine_2_3_class_options(mc);
5062     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5063     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5064     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5065 }
5066 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5067 
5068 /*
5069  * pseries-2.1
5070  */
5071 
5072 static void spapr_machine_2_1_class_options(MachineClass *mc)
5073 {
5074     spapr_machine_2_2_class_options(mc);
5075     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5076 }
5077 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5078 
5079 static void spapr_machine_register_types(void)
5080 {
5081     type_register_static(&spapr_machine_info);
5082 }
5083 
5084 type_init(spapr_machine_register_types)
5085