xref: /qemu/hw/ppc/spapr.c (revision 6026fdbdbdf5f30edc906de0ae287e95c1b6892b)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu/datadir.h"
29 #include "qemu/memalign.h"
30 #include "qemu/guest-random.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-events-machine.h"
33 #include "qapi/qapi-events-qdev.h"
34 #include "qapi/visitor.h"
35 #include "sysemu/sysemu.h"
36 #include "sysemu/hostmem.h"
37 #include "sysemu/numa.h"
38 #include "sysemu/qtest.h"
39 #include "sysemu/reset.h"
40 #include "sysemu/runstate.h"
41 #include "qemu/log.h"
42 #include "hw/fw-path-provider.h"
43 #include "elf.h"
44 #include "net/net.h"
45 #include "sysemu/device_tree.h"
46 #include "sysemu/cpus.h"
47 #include "sysemu/hw_accel.h"
48 #include "kvm_ppc.h"
49 #include "migration/misc.h"
50 #include "migration/qemu-file-types.h"
51 #include "migration/global_state.h"
52 #include "migration/register.h"
53 #include "migration/blocker.h"
54 #include "mmu-hash64.h"
55 #include "mmu-book3s-v3.h"
56 #include "cpu-models.h"
57 #include "hw/core/cpu.h"
58 
59 #include "hw/ppc/ppc.h"
60 #include "hw/loader.h"
61 
62 #include "hw/ppc/fdt.h"
63 #include "hw/ppc/spapr.h"
64 #include "hw/ppc/spapr_nested.h"
65 #include "hw/ppc/spapr_vio.h"
66 #include "hw/ppc/vof.h"
67 #include "hw/qdev-properties.h"
68 #include "hw/pci-host/spapr.h"
69 #include "hw/pci/msi.h"
70 
71 #include "hw/pci/pci.h"
72 #include "hw/scsi/scsi.h"
73 #include "hw/virtio/virtio-scsi.h"
74 #include "hw/virtio/vhost-scsi-common.h"
75 
76 #include "exec/ram_addr.h"
77 #include "hw/usb.h"
78 #include "qemu/config-file.h"
79 #include "qemu/error-report.h"
80 #include "trace.h"
81 #include "hw/nmi.h"
82 #include "hw/intc/intc.h"
83 
84 #include "hw/ppc/spapr_cpu_core.h"
85 #include "hw/mem/memory-device.h"
86 #include "hw/ppc/spapr_tpm_proxy.h"
87 #include "hw/ppc/spapr_nvdimm.h"
88 #include "hw/ppc/spapr_numa.h"
89 #include "hw/ppc/pef.h"
90 
91 #include "monitor/monitor.h"
92 
93 #include <libfdt.h>
94 
95 /* SLOF memory layout:
96  *
97  * SLOF raw image loaded at 0, copies its romfs right below the flat
98  * device-tree, then position SLOF itself 31M below that
99  *
100  * So we set FW_OVERHEAD to 40MB which should account for all of that
101  * and more
102  *
103  * We load our kernel at 4M, leaving space for SLOF initial image
104  */
105 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
106 #define FW_MAX_SIZE             0x400000
107 #define FW_FILE_NAME            "slof.bin"
108 #define FW_FILE_NAME_VOF        "vof.bin"
109 #define FW_OVERHEAD             0x2800000
110 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
111 
112 #define MIN_RMA_SLOF            (128 * MiB)
113 
114 #define PHANDLE_INTC            0x00001111
115 
116 /* These two functions implement the VCPU id numbering: one to compute them
117  * all and one to identify thread 0 of a VCORE. Any change to the first one
118  * is likely to have an impact on the second one, so let's keep them close.
119  */
120 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
121 {
122     MachineState *ms = MACHINE(spapr);
123     unsigned int smp_threads = ms->smp.threads;
124 
125     assert(spapr->vsmt);
126     return
127         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
128 }
129 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
130                                       PowerPCCPU *cpu)
131 {
132     assert(spapr->vsmt);
133     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
134 }
135 
136 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
137 {
138     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
139      * and newer QEMUs don't even have them. In both cases, we don't want
140      * to send anything on the wire.
141      */
142     return false;
143 }
144 
145 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
146     /*
147      * Hack ahead.  We can't have two devices with the same name and
148      * instance id.  So I rename this to pass make check.
149      * Real help from people who knows the hardware is needed.
150      */
151     .name = "icp/server",
152     .version_id = 1,
153     .minimum_version_id = 1,
154     .needed = pre_2_10_vmstate_dummy_icp_needed,
155     .fields = (const VMStateField[]) {
156         VMSTATE_UNUSED(4), /* uint32_t xirr */
157         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
158         VMSTATE_UNUSED(1), /* uint8_t mfrr */
159         VMSTATE_END_OF_LIST()
160     },
161 };
162 
163 /*
164  * See comment in hw/intc/xics.c:icp_realize()
165  *
166  * You have to remove vmstate_replace_hack_for_ppc() when you remove
167  * the machine types that need the following function.
168  */
169 static void pre_2_10_vmstate_register_dummy_icp(int i)
170 {
171     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
172                      (void *)(uintptr_t) i);
173 }
174 
175 /*
176  * See comment in hw/intc/xics.c:icp_realize()
177  *
178  * You have to remove vmstate_replace_hack_for_ppc() when you remove
179  * the machine types that need the following function.
180  */
181 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
182 {
183     /*
184      * This used to be:
185      *
186      *    vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
187      *                      (void *)(uintptr_t) i);
188      */
189 }
190 
191 int spapr_max_server_number(SpaprMachineState *spapr)
192 {
193     MachineState *ms = MACHINE(spapr);
194 
195     assert(spapr->vsmt);
196     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
197 }
198 
199 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
200                                   int smt_threads)
201 {
202     int i, ret = 0;
203     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
204     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
205     int index = spapr_get_vcpu_id(cpu);
206 
207     if (cpu->compat_pvr) {
208         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
209         if (ret < 0) {
210             return ret;
211         }
212     }
213 
214     /* Build interrupt servers and gservers properties */
215     for (i = 0; i < smt_threads; i++) {
216         servers_prop[i] = cpu_to_be32(index + i);
217         /* Hack, direct the group queues back to cpu 0 */
218         gservers_prop[i*2] = cpu_to_be32(index + i);
219         gservers_prop[i*2 + 1] = 0;
220     }
221     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
222                       servers_prop, sizeof(*servers_prop) * smt_threads);
223     if (ret < 0) {
224         return ret;
225     }
226     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
227                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
228 
229     return ret;
230 }
231 
232 static void spapr_dt_pa_features(SpaprMachineState *spapr,
233                                  PowerPCCPU *cpu,
234                                  void *fdt, int offset)
235 {
236     /*
237      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
238      * but not MTTCG, so disable it. To advertise it, a cap would have
239      * to be added, or support implemented for MTTCG.
240      *
241      * Copy/paste is not supported by TCG, so it is not advertised. KVM
242      * can execute them but it has no accelerator drivers which are usable,
243      * so there isn't much need for it anyway.
244      */
245 
246     /* These should be kept in sync with pnv */
247     uint8_t pa_features_206[] = { 6, 0,
248         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
249     uint8_t pa_features_207[] = { 24, 0,
250         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
251         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
252         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
254     uint8_t pa_features_300[] = { 66, 0,
255         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
256         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
257         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
258         /* 6: DS207 */
259         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
260         /* 16: Vector */
261         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
262         /* 18: Vec. Scalar, 20: Vec. XOR */
263         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
264         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
265         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
266         /* 32: LE atomic, 34: EBB + ext EBB */
267         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
268         /* 40: Radix MMU */
269         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
270         /* 42: PM, 44: PC RA, 46: SC vec'd */
271         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
272         /* 48: SIMD, 50: QP BFP, 52: String */
273         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
274         /* 54: DecFP, 56: DecI, 58: SHA */
275         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
276         /* 60: NM atomic, 62: RNG */
277         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
278     };
279     /* 3.1 removes SAO, HTM support */
280     uint8_t pa_features_31[] = { 74, 0,
281         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
282         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
283         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
284         /* 6: DS207 */
285         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
286         /* 16: Vector */
287         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
288         /* 18: Vec. Scalar, 20: Vec. XOR */
289         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
290         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
291         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
292         /* 32: LE atomic, 34: EBB + ext EBB */
293         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
294         /* 40: Radix MMU */
295         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
296         /* 42: PM, 44: PC RA, 46: SC vec'd */
297         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
298         /* 48: SIMD, 50: QP BFP, 52: String */
299         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
300         /* 54: DecFP, 56: DecI, 58: SHA */
301         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
302         /* 60: NM atomic, 62: RNG */
303         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
304         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
305         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
306         /* 72: [P]HASHST/[P]HASHCHK */
307         0x80, 0x00,                         /* 72 - 73 */
308     };
309     uint8_t *pa_features = NULL;
310     size_t pa_size;
311 
312     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
313         pa_features = pa_features_206;
314         pa_size = sizeof(pa_features_206);
315     }
316     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
317         pa_features = pa_features_207;
318         pa_size = sizeof(pa_features_207);
319     }
320     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
321         pa_features = pa_features_300;
322         pa_size = sizeof(pa_features_300);
323     }
324     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
325         pa_features = pa_features_31;
326         pa_size = sizeof(pa_features_31);
327     }
328     if (!pa_features) {
329         return;
330     }
331 
332     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
333         /*
334          * Note: we keep CI large pages off by default because a 64K capable
335          * guest provisioned with large pages might otherwise try to map a qemu
336          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
337          * even if that qemu runs on a 4k host.
338          * We dd this bit back here if we are confident this is not an issue
339          */
340         pa_features[3] |= 0x20;
341     }
342     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
343         pa_features[24] |= 0x80;    /* Transactional memory support */
344     }
345     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
346         /* Workaround for broken kernels that attempt (guest) radix
347          * mode when they can't handle it, if they see the radix bit set
348          * in pa-features. So hide it from them. */
349         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
350     }
351 
352     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
353 }
354 
355 static hwaddr spapr_node0_size(MachineState *machine)
356 {
357     if (machine->numa_state->num_nodes) {
358         int i;
359         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
360             if (machine->numa_state->nodes[i].node_mem) {
361                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
362                            machine->ram_size);
363             }
364         }
365     }
366     return machine->ram_size;
367 }
368 
369 static void add_str(GString *s, const gchar *s1)
370 {
371     g_string_append_len(s, s1, strlen(s1) + 1);
372 }
373 
374 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
375                                 hwaddr start, hwaddr size)
376 {
377     char mem_name[32];
378     uint64_t mem_reg_property[2];
379     int off;
380 
381     mem_reg_property[0] = cpu_to_be64(start);
382     mem_reg_property[1] = cpu_to_be64(size);
383 
384     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
385     off = fdt_add_subnode(fdt, 0, mem_name);
386     _FDT(off);
387     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
388     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
389                       sizeof(mem_reg_property))));
390     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
391     return off;
392 }
393 
394 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
395 {
396     MemoryDeviceInfoList *info;
397 
398     for (info = list; info; info = info->next) {
399         MemoryDeviceInfo *value = info->value;
400 
401         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
402             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
403 
404             if (addr >= pcdimm_info->addr &&
405                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
406                 return pcdimm_info->node;
407             }
408         }
409     }
410 
411     return -1;
412 }
413 
414 struct sPAPRDrconfCellV2 {
415      uint32_t seq_lmbs;
416      uint64_t base_addr;
417      uint32_t drc_index;
418      uint32_t aa_index;
419      uint32_t flags;
420 } QEMU_PACKED;
421 
422 typedef struct DrconfCellQueue {
423     struct sPAPRDrconfCellV2 cell;
424     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
425 } DrconfCellQueue;
426 
427 static DrconfCellQueue *
428 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
429                       uint32_t drc_index, uint32_t aa_index,
430                       uint32_t flags)
431 {
432     DrconfCellQueue *elem;
433 
434     elem = g_malloc0(sizeof(*elem));
435     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
436     elem->cell.base_addr = cpu_to_be64(base_addr);
437     elem->cell.drc_index = cpu_to_be32(drc_index);
438     elem->cell.aa_index = cpu_to_be32(aa_index);
439     elem->cell.flags = cpu_to_be32(flags);
440 
441     return elem;
442 }
443 
444 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
445                                       int offset, MemoryDeviceInfoList *dimms)
446 {
447     MachineState *machine = MACHINE(spapr);
448     uint8_t *int_buf, *cur_index;
449     int ret;
450     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
451     uint64_t addr, cur_addr, size;
452     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
453     uint64_t mem_end = machine->device_memory->base +
454                        memory_region_size(&machine->device_memory->mr);
455     uint32_t node, buf_len, nr_entries = 0;
456     SpaprDrc *drc;
457     DrconfCellQueue *elem, *next;
458     MemoryDeviceInfoList *info;
459     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
460         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
461 
462     /* Entry to cover RAM and the gap area */
463     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
464                                  SPAPR_LMB_FLAGS_RESERVED |
465                                  SPAPR_LMB_FLAGS_DRC_INVALID);
466     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
467     nr_entries++;
468 
469     cur_addr = machine->device_memory->base;
470     for (info = dimms; info; info = info->next) {
471         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
472 
473         addr = di->addr;
474         size = di->size;
475         node = di->node;
476 
477         /*
478          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
479          * area is marked hotpluggable in the next iteration for the bigger
480          * chunk including the NVDIMM occupied area.
481          */
482         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
483             continue;
484 
485         /* Entry for hot-pluggable area */
486         if (cur_addr < addr) {
487             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
488             g_assert(drc);
489             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
490                                          cur_addr, spapr_drc_index(drc), -1, 0);
491             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
492             nr_entries++;
493         }
494 
495         /* Entry for DIMM */
496         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
497         g_assert(drc);
498         elem = spapr_get_drconf_cell(size / lmb_size, addr,
499                                      spapr_drc_index(drc), node,
500                                      (SPAPR_LMB_FLAGS_ASSIGNED |
501                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
502         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
503         nr_entries++;
504         cur_addr = addr + size;
505     }
506 
507     /* Entry for remaining hotpluggable area */
508     if (cur_addr < mem_end) {
509         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
510         g_assert(drc);
511         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
512                                      cur_addr, spapr_drc_index(drc), -1, 0);
513         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
514         nr_entries++;
515     }
516 
517     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
518     int_buf = cur_index = g_malloc0(buf_len);
519     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
520     cur_index += sizeof(nr_entries);
521 
522     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
523         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
524         cur_index += sizeof(elem->cell);
525         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
526         g_free(elem);
527     }
528 
529     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
530     g_free(int_buf);
531     if (ret < 0) {
532         return -1;
533     }
534     return 0;
535 }
536 
537 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
538                                    int offset, MemoryDeviceInfoList *dimms)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int i, ret;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
544     uint32_t nr_lmbs = (machine->device_memory->base +
545                        memory_region_size(&machine->device_memory->mr)) /
546                        lmb_size;
547     uint32_t *int_buf, *cur_index, buf_len;
548 
549     /*
550      * Allocate enough buffer size to fit in ibm,dynamic-memory
551      */
552     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
553     cur_index = int_buf = g_malloc0(buf_len);
554     int_buf[0] = cpu_to_be32(nr_lmbs);
555     cur_index++;
556     for (i = 0; i < nr_lmbs; i++) {
557         uint64_t addr = i * lmb_size;
558         uint32_t *dynamic_memory = cur_index;
559 
560         if (i >= device_lmb_start) {
561             SpaprDrc *drc;
562 
563             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
564             g_assert(drc);
565 
566             dynamic_memory[0] = cpu_to_be32(addr >> 32);
567             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
568             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
569             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
570             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
571             if (memory_region_present(get_system_memory(), addr)) {
572                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
573             } else {
574                 dynamic_memory[5] = cpu_to_be32(0);
575             }
576         } else {
577             /*
578              * LMB information for RMA, boot time RAM and gap b/n RAM and
579              * device memory region -- all these are marked as reserved
580              * and as having no valid DRC.
581              */
582             dynamic_memory[0] = cpu_to_be32(addr >> 32);
583             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
584             dynamic_memory[2] = cpu_to_be32(0);
585             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
586             dynamic_memory[4] = cpu_to_be32(-1);
587             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
588                                             SPAPR_LMB_FLAGS_DRC_INVALID);
589         }
590 
591         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
592     }
593     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
594     g_free(int_buf);
595     if (ret < 0) {
596         return -1;
597     }
598     return 0;
599 }
600 
601 /*
602  * Adds ibm,dynamic-reconfiguration-memory node.
603  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
604  * of this device tree node.
605  */
606 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
607                                                    void *fdt)
608 {
609     MachineState *machine = MACHINE(spapr);
610     int ret, offset;
611     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
612     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
613                                 cpu_to_be32(lmb_size & 0xffffffff)};
614     MemoryDeviceInfoList *dimms = NULL;
615 
616     /* Don't create the node if there is no device memory. */
617     if (!machine->device_memory) {
618         return 0;
619     }
620 
621     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
622 
623     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
624                     sizeof(prop_lmb_size));
625     if (ret < 0) {
626         return ret;
627     }
628 
629     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
630     if (ret < 0) {
631         return ret;
632     }
633 
634     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
635     if (ret < 0) {
636         return ret;
637     }
638 
639     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
640     dimms = qmp_memory_device_list();
641     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
642         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
643     } else {
644         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
645     }
646     qapi_free_MemoryDeviceInfoList(dimms);
647 
648     if (ret < 0) {
649         return ret;
650     }
651 
652     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
653 
654     return ret;
655 }
656 
657 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
658 {
659     MachineState *machine = MACHINE(spapr);
660     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
661     hwaddr mem_start, node_size;
662     int i, nb_nodes = machine->numa_state->num_nodes;
663     NodeInfo *nodes = machine->numa_state->nodes;
664 
665     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
666         if (!nodes[i].node_mem) {
667             continue;
668         }
669         if (mem_start >= machine->ram_size) {
670             node_size = 0;
671         } else {
672             node_size = nodes[i].node_mem;
673             if (node_size > machine->ram_size - mem_start) {
674                 node_size = machine->ram_size - mem_start;
675             }
676         }
677         if (!mem_start) {
678             /* spapr_machine_init() checks for rma_size <= node0_size
679              * already */
680             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
681             mem_start += spapr->rma_size;
682             node_size -= spapr->rma_size;
683         }
684         for ( ; node_size; ) {
685             hwaddr sizetmp = pow2floor(node_size);
686 
687             /* mem_start != 0 here */
688             if (ctzl(mem_start) < ctzl(sizetmp)) {
689                 sizetmp = 1ULL << ctzl(mem_start);
690             }
691 
692             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
693             node_size -= sizetmp;
694             mem_start += sizetmp;
695         }
696     }
697 
698     /* Generate ibm,dynamic-reconfiguration-memory node if required */
699     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
700         int ret;
701 
702         g_assert(smc->dr_lmb_enabled);
703         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
704         if (ret) {
705             return ret;
706         }
707     }
708 
709     return 0;
710 }
711 
712 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
713                          SpaprMachineState *spapr)
714 {
715     MachineState *ms = MACHINE(spapr);
716     PowerPCCPU *cpu = POWERPC_CPU(cs);
717     CPUPPCState *env = &cpu->env;
718     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
719     int index = spapr_get_vcpu_id(cpu);
720     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
721                        0xffffffff, 0xffffffff};
722     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
723         : SPAPR_TIMEBASE_FREQ;
724     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
725     uint32_t page_sizes_prop[64];
726     size_t page_sizes_prop_size;
727     unsigned int smp_threads = ms->smp.threads;
728     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
729     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
730     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
731     SpaprDrc *drc;
732     int drc_index;
733     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
734     int i;
735 
736     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
737     if (drc) {
738         drc_index = spapr_drc_index(drc);
739         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
740     }
741 
742     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
743     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
744 
745     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
746     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
747                            env->dcache_line_size)));
748     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
749                            env->dcache_line_size)));
750     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
751                            env->icache_line_size)));
752     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
753                            env->icache_line_size)));
754 
755     if (pcc->l1_dcache_size) {
756         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
757                                pcc->l1_dcache_size)));
758     } else {
759         warn_report("Unknown L1 dcache size for cpu");
760     }
761     if (pcc->l1_icache_size) {
762         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
763                                pcc->l1_icache_size)));
764     } else {
765         warn_report("Unknown L1 icache size for cpu");
766     }
767 
768     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
769     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
770     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
771     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
772     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
773     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
774 
775     if (ppc_has_spr(cpu, SPR_PURR)) {
776         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
777     }
778     if (ppc_has_spr(cpu, SPR_PURR)) {
779         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
780     }
781 
782     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
783         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
784                           segs, sizeof(segs))));
785     }
786 
787     /* Advertise VSX (vector extensions) if available
788      *   1               == VMX / Altivec available
789      *   2               == VSX available
790      *
791      * Only CPUs for which we create core types in spapr_cpu_core.c
792      * are possible, and all of those have VMX */
793     if (env->insns_flags & PPC_ALTIVEC) {
794         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
795             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
796         } else {
797             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
798         }
799     }
800 
801     /* Advertise DFP (Decimal Floating Point) if available
802      *   0 / no property == no DFP
803      *   1               == DFP available */
804     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
805         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
806     }
807 
808     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
809                                                       sizeof(page_sizes_prop));
810     if (page_sizes_prop_size) {
811         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
812                           page_sizes_prop, page_sizes_prop_size)));
813     }
814 
815     spapr_dt_pa_features(spapr, cpu, fdt, offset);
816 
817     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
818                            cs->cpu_index / vcpus_per_socket)));
819 
820     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
821                       pft_size_prop, sizeof(pft_size_prop))));
822 
823     if (ms->numa_state->num_nodes > 1) {
824         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
825     }
826 
827     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
828 
829     if (pcc->radix_page_info) {
830         for (i = 0; i < pcc->radix_page_info->count; i++) {
831             radix_AP_encodings[i] =
832                 cpu_to_be32(pcc->radix_page_info->entries[i]);
833         }
834         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
835                           radix_AP_encodings,
836                           pcc->radix_page_info->count *
837                           sizeof(radix_AP_encodings[0]))));
838     }
839 
840     /*
841      * We set this property to let the guest know that it can use the large
842      * decrementer and its width in bits.
843      */
844     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
845         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
846                               pcc->lrg_decr_bits)));
847 }
848 
849 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
850                              int cpus_offset)
851 {
852     PowerPCCPU *cpu = POWERPC_CPU(cs);
853     int index = spapr_get_vcpu_id(cpu);
854     DeviceClass *dc = DEVICE_GET_CLASS(cs);
855     g_autofree char *nodename = NULL;
856     int offset;
857 
858     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
859         return;
860     }
861 
862     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
863     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
864     _FDT(offset);
865     spapr_dt_cpu(cs, fdt, offset, spapr);
866 }
867 
868 
869 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
870 {
871     CPUState **rev;
872     CPUState *cs;
873     int n_cpus;
874     int cpus_offset;
875     int i;
876 
877     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
878     _FDT(cpus_offset);
879     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
880     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
881 
882     /*
883      * We walk the CPUs in reverse order to ensure that CPU DT nodes
884      * created by fdt_add_subnode() end up in the right order in FDT
885      * for the guest kernel the enumerate the CPUs correctly.
886      *
887      * The CPU list cannot be traversed in reverse order, so we need
888      * to do extra work.
889      */
890     n_cpus = 0;
891     rev = NULL;
892     CPU_FOREACH(cs) {
893         rev = g_renew(CPUState *, rev, n_cpus + 1);
894         rev[n_cpus++] = cs;
895     }
896 
897     for (i = n_cpus - 1; i >= 0; i--) {
898         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
899     }
900 
901     g_free(rev);
902 }
903 
904 static int spapr_dt_rng(void *fdt)
905 {
906     int node;
907     int ret;
908 
909     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
910     if (node <= 0) {
911         return -1;
912     }
913     ret = fdt_setprop_string(fdt, node, "device_type",
914                              "ibm,platform-facilities");
915     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
916     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
917 
918     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
919     if (node <= 0) {
920         return -1;
921     }
922     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
923 
924     return ret ? -1 : 0;
925 }
926 
927 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
928 {
929     MachineState *ms = MACHINE(spapr);
930     int rtas;
931     GString *hypertas = g_string_sized_new(256);
932     GString *qemu_hypertas = g_string_sized_new(256);
933     uint32_t lrdr_capacity[] = {
934         0,
935         0,
936         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
937         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
938         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
939     };
940 
941     /* Do we have device memory? */
942     if (MACHINE(spapr)->device_memory) {
943         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
944             memory_region_size(&MACHINE(spapr)->device_memory->mr);
945 
946         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
947         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
948     }
949 
950     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
951 
952     /* hypertas */
953     add_str(hypertas, "hcall-pft");
954     add_str(hypertas, "hcall-term");
955     add_str(hypertas, "hcall-dabr");
956     add_str(hypertas, "hcall-interrupt");
957     add_str(hypertas, "hcall-tce");
958     add_str(hypertas, "hcall-vio");
959     add_str(hypertas, "hcall-splpar");
960     add_str(hypertas, "hcall-join");
961     add_str(hypertas, "hcall-bulk");
962     add_str(hypertas, "hcall-set-mode");
963     add_str(hypertas, "hcall-sprg0");
964     add_str(hypertas, "hcall-copy");
965     add_str(hypertas, "hcall-debug");
966     add_str(hypertas, "hcall-vphn");
967     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
968         add_str(hypertas, "hcall-rpt-invalidate");
969     }
970 
971     add_str(qemu_hypertas, "hcall-memop1");
972 
973     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
974         add_str(hypertas, "hcall-multi-tce");
975     }
976 
977     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
978         add_str(hypertas, "hcall-hpt-resize");
979     }
980 
981     add_str(hypertas, "hcall-watchdog");
982 
983     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
984                      hypertas->str, hypertas->len));
985     g_string_free(hypertas, TRUE);
986     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
987                      qemu_hypertas->str, qemu_hypertas->len));
988     g_string_free(qemu_hypertas, TRUE);
989 
990     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
991 
992     /*
993      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
994      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
995      *
996      * The system reset requirements are driven by existing Linux and PowerVM
997      * implementation which (contrary to PAPR) saves r3 in the error log
998      * structure like machine check, so Linux expects to find the saved r3
999      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
1000      * does not look at the error value).
1001      *
1002      * System reset interrupts are not subject to interlock like machine
1003      * check, so this memory area could be corrupted if the sreset is
1004      * interrupted by a machine check (or vice versa) if it was shared. To
1005      * prevent this, system reset uses per-CPU areas for the sreset save
1006      * area. A system reset that interrupts a system reset handler could
1007      * still overwrite this area, but Linux doesn't try to recover in that
1008      * case anyway.
1009      *
1010      * The extra 8 bytes is required because Linux's FWNMI error log check
1011      * is off-by-one.
1012      *
1013      * RTAS_MIN_SIZE is required for the RTAS blob itself.
1014      */
1015     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
1016                           RTAS_ERROR_LOG_MAX +
1017                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
1018                           sizeof(uint64_t)));
1019     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1020                           RTAS_ERROR_LOG_MAX));
1021     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1022                           RTAS_EVENT_SCAN_RATE));
1023 
1024     g_assert(msi_nonbroken);
1025     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1026 
1027     /*
1028      * According to PAPR, rtas ibm,os-term does not guarantee a return
1029      * back to the guest cpu.
1030      *
1031      * While an additional ibm,extended-os-term property indicates
1032      * that rtas call return will always occur. Set this property.
1033      */
1034     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1035 
1036     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1037                      lrdr_capacity, sizeof(lrdr_capacity)));
1038 
1039     spapr_dt_rtas_tokens(fdt, rtas);
1040 }
1041 
1042 /*
1043  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1044  * and the XIVE features that the guest may request and thus the valid
1045  * values for bytes 23..26 of option vector 5:
1046  */
1047 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1048                                           int chosen)
1049 {
1050     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1051 
1052     char val[2 * 4] = {
1053         23, 0x00, /* XICS / XIVE mode */
1054         24, 0x00, /* Hash/Radix, filled in below. */
1055         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1056         26, 0x40, /* Radix options: GTSE == yes. */
1057     };
1058 
1059     if (spapr->irq->xics && spapr->irq->xive) {
1060         val[1] = SPAPR_OV5_XIVE_BOTH;
1061     } else if (spapr->irq->xive) {
1062         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1063     } else {
1064         assert(spapr->irq->xics);
1065         val[1] = SPAPR_OV5_XIVE_LEGACY;
1066     }
1067 
1068     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1069                           first_ppc_cpu->compat_pvr)) {
1070         /*
1071          * If we're in a pre POWER9 compat mode then the guest should
1072          * do hash and use the legacy interrupt mode
1073          */
1074         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1075         val[3] = 0x00; /* Hash */
1076         spapr_check_mmu_mode(false);
1077     } else if (kvm_enabled()) {
1078         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1079             val[3] = 0x80; /* OV5_MMU_BOTH */
1080         } else if (kvmppc_has_cap_mmu_radix()) {
1081             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1082         } else {
1083             val[3] = 0x00; /* Hash */
1084         }
1085     } else {
1086         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1087         val[3] = 0xC0;
1088     }
1089     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1090                      val, sizeof(val)));
1091 }
1092 
1093 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1094 {
1095     MachineState *machine = MACHINE(spapr);
1096     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1097     int chosen;
1098 
1099     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1100 
1101     if (reset) {
1102         const char *boot_device = spapr->boot_device;
1103         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1104         size_t cb = 0;
1105         g_autofree char *bootlist = get_boot_devices_list(&cb);
1106 
1107         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1108             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1109                                     machine->kernel_cmdline));
1110         }
1111 
1112         if (spapr->initrd_size) {
1113             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1114                                   spapr->initrd_base));
1115             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1116                                   spapr->initrd_base + spapr->initrd_size));
1117         }
1118 
1119         if (spapr->kernel_size) {
1120             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1121                                   cpu_to_be64(spapr->kernel_size) };
1122 
1123             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1124                          &kprop, sizeof(kprop)));
1125             if (spapr->kernel_le) {
1126                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1127             }
1128         }
1129         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1130             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1131         }
1132         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1133         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1134         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1135 
1136         if (cb && bootlist) {
1137             int i;
1138 
1139             for (i = 0; i < cb; i++) {
1140                 if (bootlist[i] == '\n') {
1141                     bootlist[i] = ' ';
1142                 }
1143             }
1144             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1145         }
1146 
1147         if (boot_device && strlen(boot_device)) {
1148             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1149         }
1150 
1151         if (spapr->want_stdout_path && stdout_path) {
1152             /*
1153              * "linux,stdout-path" and "stdout" properties are
1154              * deprecated by linux kernel. New platforms should only
1155              * use the "stdout-path" property. Set the new property
1156              * and continue using older property to remain compatible
1157              * with the existing firmware.
1158              */
1159             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1160             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1161         }
1162 
1163         /*
1164          * We can deal with BAR reallocation just fine, advertise it
1165          * to the guest
1166          */
1167         if (smc->linux_pci_probe) {
1168             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1169         }
1170 
1171         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1172     }
1173 
1174     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1175 
1176     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1177 }
1178 
1179 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1180 {
1181     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1182      * KVM to work under pHyp with some guest co-operation */
1183     int hypervisor;
1184     uint8_t hypercall[16];
1185 
1186     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1187     /* indicate KVM hypercall interface */
1188     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1189     if (kvmppc_has_cap_fixup_hcalls()) {
1190         /*
1191          * Older KVM versions with older guest kernels were broken
1192          * with the magic page, don't allow the guest to map it.
1193          */
1194         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1195                                   sizeof(hypercall))) {
1196             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1197                              hypercall, sizeof(hypercall)));
1198         }
1199     }
1200 }
1201 
1202 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1203 {
1204     MachineState *machine = MACHINE(spapr);
1205     MachineClass *mc = MACHINE_GET_CLASS(machine);
1206     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1207     uint32_t root_drc_type_mask = 0;
1208     int ret;
1209     void *fdt;
1210     SpaprPhbState *phb;
1211     char *buf;
1212 
1213     fdt = g_malloc0(space);
1214     _FDT((fdt_create_empty_tree(fdt, space)));
1215 
1216     /* Root node */
1217     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1218     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1219     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1220 
1221     /* Guest UUID & Name*/
1222     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1223     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1224     if (qemu_uuid_set) {
1225         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1226     }
1227     g_free(buf);
1228 
1229     if (qemu_get_vm_name()) {
1230         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1231                                 qemu_get_vm_name()));
1232     }
1233 
1234     /* Host Model & Serial Number */
1235     if (spapr->host_model) {
1236         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1237     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1238         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1239         g_free(buf);
1240     }
1241 
1242     if (spapr->host_serial) {
1243         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1244     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1245         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1246         g_free(buf);
1247     }
1248 
1249     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1250     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1251 
1252     /* /interrupt controller */
1253     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1254 
1255     ret = spapr_dt_memory(spapr, fdt);
1256     if (ret < 0) {
1257         error_report("couldn't setup memory nodes in fdt");
1258         exit(1);
1259     }
1260 
1261     /* /vdevice */
1262     spapr_dt_vdevice(spapr->vio_bus, fdt);
1263 
1264     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1265         ret = spapr_dt_rng(fdt);
1266         if (ret < 0) {
1267             error_report("could not set up rng device in the fdt");
1268             exit(1);
1269         }
1270     }
1271 
1272     QLIST_FOREACH(phb, &spapr->phbs, list) {
1273         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1274         if (ret < 0) {
1275             error_report("couldn't setup PCI devices in fdt");
1276             exit(1);
1277         }
1278     }
1279 
1280     spapr_dt_cpus(fdt, spapr);
1281 
1282     /* ibm,drc-indexes and friends */
1283     if (smc->dr_lmb_enabled) {
1284         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1285     }
1286     if (smc->dr_phb_enabled) {
1287         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1288     }
1289     if (mc->nvdimm_supported) {
1290         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1291     }
1292     if (root_drc_type_mask) {
1293         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1294     }
1295 
1296     if (mc->has_hotpluggable_cpus) {
1297         int offset = fdt_path_offset(fdt, "/cpus");
1298         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1299         if (ret < 0) {
1300             error_report("Couldn't set up CPU DR device tree properties");
1301             exit(1);
1302         }
1303     }
1304 
1305     /* /event-sources */
1306     spapr_dt_events(spapr, fdt);
1307 
1308     /* /rtas */
1309     spapr_dt_rtas(spapr, fdt);
1310 
1311     /* /chosen */
1312     spapr_dt_chosen(spapr, fdt, reset);
1313 
1314     /* /hypervisor */
1315     if (kvm_enabled()) {
1316         spapr_dt_hypervisor(spapr, fdt);
1317     }
1318 
1319     /* Build memory reserve map */
1320     if (reset) {
1321         if (spapr->kernel_size) {
1322             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1323                                   spapr->kernel_size)));
1324         }
1325         if (spapr->initrd_size) {
1326             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1327                                   spapr->initrd_size)));
1328         }
1329     }
1330 
1331     /* NVDIMM devices */
1332     if (mc->nvdimm_supported) {
1333         spapr_dt_persistent_memory(spapr, fdt);
1334     }
1335 
1336     return fdt;
1337 }
1338 
1339 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1340 {
1341     SpaprMachineState *spapr = opaque;
1342 
1343     return (addr & 0x0fffffff) + spapr->kernel_addr;
1344 }
1345 
1346 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1347                                     PowerPCCPU *cpu)
1348 {
1349     CPUPPCState *env = &cpu->env;
1350 
1351     /* The TCG path should also be holding the BQL at this point */
1352     g_assert(bql_locked());
1353 
1354     g_assert(!vhyp_cpu_in_nested(cpu));
1355 
1356     if (FIELD_EX64(env->msr, MSR, PR)) {
1357         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1358         env->gpr[3] = H_PRIVILEGE;
1359     } else {
1360         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1361     }
1362 }
1363 
1364 struct LPCRSyncState {
1365     target_ulong value;
1366     target_ulong mask;
1367 };
1368 
1369 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1370 {
1371     struct LPCRSyncState *s = arg.host_ptr;
1372     PowerPCCPU *cpu = POWERPC_CPU(cs);
1373     CPUPPCState *env = &cpu->env;
1374     target_ulong lpcr;
1375 
1376     cpu_synchronize_state(cs);
1377     lpcr = env->spr[SPR_LPCR];
1378     lpcr &= ~s->mask;
1379     lpcr |= s->value;
1380     ppc_store_lpcr(cpu, lpcr);
1381 }
1382 
1383 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1384 {
1385     CPUState *cs;
1386     struct LPCRSyncState s = {
1387         .value = value,
1388         .mask = mask
1389     };
1390     CPU_FOREACH(cs) {
1391         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1392     }
1393 }
1394 
1395 /* May be used when the machine is not running */
1396 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1397 {
1398     CPUState *cs;
1399     CPU_FOREACH(cs) {
1400         PowerPCCPU *cpu = POWERPC_CPU(cs);
1401         CPUPPCState *env = &cpu->env;
1402         target_ulong lpcr;
1403 
1404         lpcr = env->spr[SPR_LPCR];
1405         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1406         ppc_store_lpcr(cpu, lpcr);
1407     }
1408 }
1409 
1410 
1411 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1412                            target_ulong lpid, ppc_v3_pate_t *entry)
1413 {
1414     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1415     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1416 
1417     if (!spapr_cpu->in_nested) {
1418         assert(lpid == 0);
1419 
1420         /* Copy PATE1:GR into PATE0:HR */
1421         entry->dw0 = spapr->patb_entry & PATE0_HR;
1422         entry->dw1 = spapr->patb_entry;
1423 
1424     } else {
1425         uint64_t patb, pats;
1426 
1427         assert(lpid != 0);
1428 
1429         patb = spapr->nested_ptcr & PTCR_PATB;
1430         pats = spapr->nested_ptcr & PTCR_PATS;
1431 
1432         /* Check if partition table is properly aligned */
1433         if (patb & MAKE_64BIT_MASK(0, pats + 12)) {
1434             return false;
1435         }
1436 
1437         /* Calculate number of entries */
1438         pats = 1ull << (pats + 12 - 4);
1439         if (pats <= lpid) {
1440             return false;
1441         }
1442 
1443         /* Grab entry */
1444         patb += 16 * lpid;
1445         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1446         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1447     }
1448 
1449     return true;
1450 }
1451 
1452 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1453 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1454 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1455 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1456 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1457 
1458 /*
1459  * Get the fd to access the kernel htab, re-opening it if necessary
1460  */
1461 static int get_htab_fd(SpaprMachineState *spapr)
1462 {
1463     Error *local_err = NULL;
1464 
1465     if (spapr->htab_fd >= 0) {
1466         return spapr->htab_fd;
1467     }
1468 
1469     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1470     if (spapr->htab_fd < 0) {
1471         error_report_err(local_err);
1472     }
1473 
1474     return spapr->htab_fd;
1475 }
1476 
1477 void close_htab_fd(SpaprMachineState *spapr)
1478 {
1479     if (spapr->htab_fd >= 0) {
1480         close(spapr->htab_fd);
1481     }
1482     spapr->htab_fd = -1;
1483 }
1484 
1485 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1486 {
1487     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1488 
1489     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1490 }
1491 
1492 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1493 {
1494     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1495 
1496     assert(kvm_enabled());
1497 
1498     if (!spapr->htab) {
1499         return 0;
1500     }
1501 
1502     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1503 }
1504 
1505 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1506                                                 hwaddr ptex, int n)
1507 {
1508     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1509     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1510 
1511     if (!spapr->htab) {
1512         /*
1513          * HTAB is controlled by KVM. Fetch into temporary buffer
1514          */
1515         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1516         kvmppc_read_hptes(hptes, ptex, n);
1517         return hptes;
1518     }
1519 
1520     /*
1521      * HTAB is controlled by QEMU. Just point to the internally
1522      * accessible PTEG.
1523      */
1524     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1525 }
1526 
1527 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1528                               const ppc_hash_pte64_t *hptes,
1529                               hwaddr ptex, int n)
1530 {
1531     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1532 
1533     if (!spapr->htab) {
1534         g_free((void *)hptes);
1535     }
1536 
1537     /* Nothing to do for qemu managed HPT */
1538 }
1539 
1540 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1541                       uint64_t pte0, uint64_t pte1)
1542 {
1543     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1544     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1545 
1546     if (!spapr->htab) {
1547         kvmppc_write_hpte(ptex, pte0, pte1);
1548     } else {
1549         if (pte0 & HPTE64_V_VALID) {
1550             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1551             /*
1552              * When setting valid, we write PTE1 first. This ensures
1553              * proper synchronization with the reading code in
1554              * ppc_hash64_pteg_search()
1555              */
1556             smp_wmb();
1557             stq_p(spapr->htab + offset, pte0);
1558         } else {
1559             stq_p(spapr->htab + offset, pte0);
1560             /*
1561              * When clearing it we set PTE0 first. This ensures proper
1562              * synchronization with the reading code in
1563              * ppc_hash64_pteg_search()
1564              */
1565             smp_wmb();
1566             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1567         }
1568     }
1569 }
1570 
1571 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1572                              uint64_t pte1)
1573 {
1574     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1575     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1576 
1577     if (!spapr->htab) {
1578         /* There should always be a hash table when this is called */
1579         error_report("spapr_hpte_set_c called with no hash table !");
1580         return;
1581     }
1582 
1583     /* The HW performs a non-atomic byte update */
1584     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1585 }
1586 
1587 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1588                              uint64_t pte1)
1589 {
1590     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1591     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1592 
1593     if (!spapr->htab) {
1594         /* There should always be a hash table when this is called */
1595         error_report("spapr_hpte_set_r called with no hash table !");
1596         return;
1597     }
1598 
1599     /* The HW performs a non-atomic byte update */
1600     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1601 }
1602 
1603 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1604 {
1605     int shift;
1606 
1607     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1608      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1609      * that's much more than is needed for Linux guests */
1610     shift = ctz64(pow2ceil(ramsize)) - 7;
1611     shift = MAX(shift, 18); /* Minimum architected size */
1612     shift = MIN(shift, 46); /* Maximum architected size */
1613     return shift;
1614 }
1615 
1616 void spapr_free_hpt(SpaprMachineState *spapr)
1617 {
1618     qemu_vfree(spapr->htab);
1619     spapr->htab = NULL;
1620     spapr->htab_shift = 0;
1621     close_htab_fd(spapr);
1622 }
1623 
1624 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1625 {
1626     ERRP_GUARD();
1627     long rc;
1628 
1629     /* Clean up any HPT info from a previous boot */
1630     spapr_free_hpt(spapr);
1631 
1632     rc = kvmppc_reset_htab(shift);
1633 
1634     if (rc == -EOPNOTSUPP) {
1635         error_setg(errp, "HPT not supported in nested guests");
1636         return -EOPNOTSUPP;
1637     }
1638 
1639     if (rc < 0) {
1640         /* kernel-side HPT needed, but couldn't allocate one */
1641         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1642                          shift);
1643         error_append_hint(errp, "Try smaller maxmem?\n");
1644         return -errno;
1645     } else if (rc > 0) {
1646         /* kernel-side HPT allocated */
1647         if (rc != shift) {
1648             error_setg(errp,
1649                        "Requested order %d HPT, but kernel allocated order %ld",
1650                        shift, rc);
1651             error_append_hint(errp, "Try smaller maxmem?\n");
1652             return -ENOSPC;
1653         }
1654 
1655         spapr->htab_shift = shift;
1656         spapr->htab = NULL;
1657     } else {
1658         /* kernel-side HPT not needed, allocate in userspace instead */
1659         size_t size = 1ULL << shift;
1660         int i;
1661 
1662         spapr->htab = qemu_memalign(size, size);
1663         memset(spapr->htab, 0, size);
1664         spapr->htab_shift = shift;
1665 
1666         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1667             DIRTY_HPTE(HPTE(spapr->htab, i));
1668         }
1669     }
1670     /* We're setting up a hash table, so that means we're not radix */
1671     spapr->patb_entry = 0;
1672     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1673     return 0;
1674 }
1675 
1676 void spapr_setup_hpt(SpaprMachineState *spapr)
1677 {
1678     int hpt_shift;
1679 
1680     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1681         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1682     } else {
1683         uint64_t current_ram_size;
1684 
1685         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1686         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1687     }
1688     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1689 
1690     if (kvm_enabled()) {
1691         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1692 
1693         /* Check our RMA fits in the possible VRMA */
1694         if (vrma_limit < spapr->rma_size) {
1695             error_report("Unable to create %" HWADDR_PRIu
1696                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1697                          spapr->rma_size / MiB, vrma_limit / MiB);
1698             exit(EXIT_FAILURE);
1699         }
1700     }
1701 }
1702 
1703 void spapr_check_mmu_mode(bool guest_radix)
1704 {
1705     if (guest_radix) {
1706         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1707             error_report("Guest requested unavailable MMU mode (radix).");
1708             exit(EXIT_FAILURE);
1709         }
1710     } else {
1711         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1712             && !kvmppc_has_cap_mmu_hash_v3()) {
1713             error_report("Guest requested unavailable MMU mode (hash).");
1714             exit(EXIT_FAILURE);
1715         }
1716     }
1717 }
1718 
1719 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1720 {
1721     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1722     PowerPCCPU *first_ppc_cpu;
1723     hwaddr fdt_addr;
1724     void *fdt;
1725     int rc;
1726 
1727     if (reason != SHUTDOWN_CAUSE_SNAPSHOT_LOAD) {
1728         /*
1729          * Record-replay snapshot load must not consume random, this was
1730          * already replayed from initial machine reset.
1731          */
1732         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1733     }
1734 
1735     pef_kvm_reset(machine->cgs, &error_fatal);
1736     spapr_caps_apply(spapr);
1737     spapr_nested_reset(spapr);
1738 
1739     first_ppc_cpu = POWERPC_CPU(first_cpu);
1740     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1741         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1742                               spapr->max_compat_pvr)) {
1743         /*
1744          * If using KVM with radix mode available, VCPUs can be started
1745          * without a HPT because KVM will start them in radix mode.
1746          * Set the GR bit in PATE so that we know there is no HPT.
1747          */
1748         spapr->patb_entry = PATE1_GR;
1749         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1750     } else {
1751         spapr_setup_hpt(spapr);
1752     }
1753 
1754     qemu_devices_reset(reason);
1755 
1756     spapr_ovec_cleanup(spapr->ov5_cas);
1757     spapr->ov5_cas = spapr_ovec_new();
1758 
1759     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1760 
1761     /*
1762      * This is fixing some of the default configuration of the XIVE
1763      * devices. To be called after the reset of the machine devices.
1764      */
1765     spapr_irq_reset(spapr, &error_fatal);
1766 
1767     /*
1768      * There is no CAS under qtest. Simulate one to please the code that
1769      * depends on spapr->ov5_cas. This is especially needed to test device
1770      * unplug, so we do that before resetting the DRCs.
1771      */
1772     if (qtest_enabled()) {
1773         spapr_ovec_cleanup(spapr->ov5_cas);
1774         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1775     }
1776 
1777     spapr_nvdimm_finish_flushes();
1778 
1779     /* DRC reset may cause a device to be unplugged. This will cause troubles
1780      * if this device is used by another device (eg, a running vhost backend
1781      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1782      * situations, we reset DRCs after all devices have been reset.
1783      */
1784     spapr_drc_reset_all(spapr);
1785 
1786     spapr_clear_pending_events(spapr);
1787 
1788     /*
1789      * We place the device tree just below either the top of the RMA,
1790      * or just below 2GB, whichever is lower, so that it can be
1791      * processed with 32-bit real mode code if necessary
1792      */
1793     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1794 
1795     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1796     if (spapr->vof) {
1797         spapr_vof_reset(spapr, fdt, &error_fatal);
1798         /*
1799          * Do not pack the FDT as the client may change properties.
1800          * VOF client does not expect the FDT so we do not load it to the VM.
1801          */
1802     } else {
1803         rc = fdt_pack(fdt);
1804         /* Should only fail if we've built a corrupted tree */
1805         assert(rc == 0);
1806 
1807         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1808                                   0, fdt_addr, 0);
1809         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1810     }
1811     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1812 
1813     g_free(spapr->fdt_blob);
1814     spapr->fdt_size = fdt_totalsize(fdt);
1815     spapr->fdt_initial_size = spapr->fdt_size;
1816     spapr->fdt_blob = fdt;
1817 
1818     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1819     machine->fdt = fdt;
1820 
1821     /* Set up the entry state */
1822     first_ppc_cpu->env.gpr[5] = 0;
1823 
1824     spapr->fwnmi_system_reset_addr = -1;
1825     spapr->fwnmi_machine_check_addr = -1;
1826     spapr->fwnmi_machine_check_interlock = -1;
1827 
1828     /* Signal all vCPUs waiting on this condition */
1829     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1830 
1831     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1832 }
1833 
1834 static void spapr_create_nvram(SpaprMachineState *spapr)
1835 {
1836     DeviceState *dev = qdev_new("spapr-nvram");
1837     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1838 
1839     if (dinfo) {
1840         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1841                                 &error_fatal);
1842     }
1843 
1844     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1845 
1846     spapr->nvram = (struct SpaprNvram *)dev;
1847 }
1848 
1849 static void spapr_rtc_create(SpaprMachineState *spapr)
1850 {
1851     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1852                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1853                                        &error_fatal, NULL);
1854     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1855     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1856                               "date");
1857 }
1858 
1859 /* Returns whether we want to use VGA or not */
1860 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1861 {
1862     vga_interface_created = true;
1863     switch (vga_interface_type) {
1864     case VGA_NONE:
1865         return false;
1866     case VGA_DEVICE:
1867         return true;
1868     case VGA_STD:
1869     case VGA_VIRTIO:
1870     case VGA_CIRRUS:
1871         return pci_vga_init(pci_bus) != NULL;
1872     default:
1873         error_setg(errp,
1874                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1875         return false;
1876     }
1877 }
1878 
1879 static int spapr_pre_load(void *opaque)
1880 {
1881     int rc;
1882 
1883     rc = spapr_caps_pre_load(opaque);
1884     if (rc) {
1885         return rc;
1886     }
1887 
1888     return 0;
1889 }
1890 
1891 static int spapr_post_load(void *opaque, int version_id)
1892 {
1893     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1894     int err = 0;
1895 
1896     err = spapr_caps_post_migration(spapr);
1897     if (err) {
1898         return err;
1899     }
1900 
1901     /*
1902      * In earlier versions, there was no separate qdev for the PAPR
1903      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1904      * So when migrating from those versions, poke the incoming offset
1905      * value into the RTC device
1906      */
1907     if (version_id < 3) {
1908         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1909         if (err) {
1910             return err;
1911         }
1912     }
1913 
1914     if (kvm_enabled() && spapr->patb_entry) {
1915         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1916         bool radix = !!(spapr->patb_entry & PATE1_GR);
1917         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1918 
1919         /*
1920          * Update LPCR:HR and UPRT as they may not be set properly in
1921          * the stream
1922          */
1923         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1924                             LPCR_HR | LPCR_UPRT);
1925 
1926         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1927         if (err) {
1928             error_report("Process table config unsupported by the host");
1929             return -EINVAL;
1930         }
1931     }
1932 
1933     err = spapr_irq_post_load(spapr, version_id);
1934     if (err) {
1935         return err;
1936     }
1937 
1938     return err;
1939 }
1940 
1941 static int spapr_pre_save(void *opaque)
1942 {
1943     int rc;
1944 
1945     rc = spapr_caps_pre_save(opaque);
1946     if (rc) {
1947         return rc;
1948     }
1949 
1950     return 0;
1951 }
1952 
1953 static bool version_before_3(void *opaque, int version_id)
1954 {
1955     return version_id < 3;
1956 }
1957 
1958 static bool spapr_pending_events_needed(void *opaque)
1959 {
1960     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1961     return !QTAILQ_EMPTY(&spapr->pending_events);
1962 }
1963 
1964 static const VMStateDescription vmstate_spapr_event_entry = {
1965     .name = "spapr_event_log_entry",
1966     .version_id = 1,
1967     .minimum_version_id = 1,
1968     .fields = (const VMStateField[]) {
1969         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1970         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1971         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1972                                      NULL, extended_length),
1973         VMSTATE_END_OF_LIST()
1974     },
1975 };
1976 
1977 static const VMStateDescription vmstate_spapr_pending_events = {
1978     .name = "spapr_pending_events",
1979     .version_id = 1,
1980     .minimum_version_id = 1,
1981     .needed = spapr_pending_events_needed,
1982     .fields = (const VMStateField[]) {
1983         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1984                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1985         VMSTATE_END_OF_LIST()
1986     },
1987 };
1988 
1989 static bool spapr_ov5_cas_needed(void *opaque)
1990 {
1991     SpaprMachineState *spapr = opaque;
1992     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1993     bool cas_needed;
1994 
1995     /* Prior to the introduction of SpaprOptionVector, we had two option
1996      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1997      * Both of these options encode machine topology into the device-tree
1998      * in such a way that the now-booted OS should still be able to interact
1999      * appropriately with QEMU regardless of what options were actually
2000      * negotiatied on the source side.
2001      *
2002      * As such, we can avoid migrating the CAS-negotiated options if these
2003      * are the only options available on the current machine/platform.
2004      * Since these are the only options available for pseries-2.7 and
2005      * earlier, this allows us to maintain old->new/new->old migration
2006      * compatibility.
2007      *
2008      * For QEMU 2.8+, there are additional CAS-negotiatable options available
2009      * via default pseries-2.8 machines and explicit command-line parameters.
2010      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
2011      * of the actual CAS-negotiated values to continue working properly. For
2012      * example, availability of memory unplug depends on knowing whether
2013      * OV5_HP_EVT was negotiated via CAS.
2014      *
2015      * Thus, for any cases where the set of available CAS-negotiatable
2016      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2017      * include the CAS-negotiated options in the migration stream, unless
2018      * if they affect boot time behaviour only.
2019      */
2020     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2021     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2022     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2023 
2024     /* We need extra information if we have any bits outside the mask
2025      * defined above */
2026     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
2027 
2028     spapr_ovec_cleanup(ov5_mask);
2029 
2030     return cas_needed;
2031 }
2032 
2033 static const VMStateDescription vmstate_spapr_ov5_cas = {
2034     .name = "spapr_option_vector_ov5_cas",
2035     .version_id = 1,
2036     .minimum_version_id = 1,
2037     .needed = spapr_ov5_cas_needed,
2038     .fields = (const VMStateField[]) {
2039         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2040                                  vmstate_spapr_ovec, SpaprOptionVector),
2041         VMSTATE_END_OF_LIST()
2042     },
2043 };
2044 
2045 static bool spapr_patb_entry_needed(void *opaque)
2046 {
2047     SpaprMachineState *spapr = opaque;
2048 
2049     return !!spapr->patb_entry;
2050 }
2051 
2052 static const VMStateDescription vmstate_spapr_patb_entry = {
2053     .name = "spapr_patb_entry",
2054     .version_id = 1,
2055     .minimum_version_id = 1,
2056     .needed = spapr_patb_entry_needed,
2057     .fields = (const VMStateField[]) {
2058         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2059         VMSTATE_END_OF_LIST()
2060     },
2061 };
2062 
2063 static bool spapr_irq_map_needed(void *opaque)
2064 {
2065     SpaprMachineState *spapr = opaque;
2066 
2067     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2068 }
2069 
2070 static const VMStateDescription vmstate_spapr_irq_map = {
2071     .name = "spapr_irq_map",
2072     .version_id = 1,
2073     .minimum_version_id = 1,
2074     .needed = spapr_irq_map_needed,
2075     .fields = (const VMStateField[]) {
2076         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2077         VMSTATE_END_OF_LIST()
2078     },
2079 };
2080 
2081 static bool spapr_dtb_needed(void *opaque)
2082 {
2083     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2084 
2085     return smc->update_dt_enabled;
2086 }
2087 
2088 static int spapr_dtb_pre_load(void *opaque)
2089 {
2090     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2091 
2092     g_free(spapr->fdt_blob);
2093     spapr->fdt_blob = NULL;
2094     spapr->fdt_size = 0;
2095 
2096     return 0;
2097 }
2098 
2099 static const VMStateDescription vmstate_spapr_dtb = {
2100     .name = "spapr_dtb",
2101     .version_id = 1,
2102     .minimum_version_id = 1,
2103     .needed = spapr_dtb_needed,
2104     .pre_load = spapr_dtb_pre_load,
2105     .fields = (const VMStateField[]) {
2106         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2107         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2108         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2109                                      fdt_size),
2110         VMSTATE_END_OF_LIST()
2111     },
2112 };
2113 
2114 static bool spapr_fwnmi_needed(void *opaque)
2115 {
2116     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2117 
2118     return spapr->fwnmi_machine_check_addr != -1;
2119 }
2120 
2121 static int spapr_fwnmi_pre_save(void *opaque)
2122 {
2123     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2124 
2125     /*
2126      * Check if machine check handling is in progress and print a
2127      * warning message.
2128      */
2129     if (spapr->fwnmi_machine_check_interlock != -1) {
2130         warn_report("A machine check is being handled during migration. The"
2131                 "handler may run and log hardware error on the destination");
2132     }
2133 
2134     return 0;
2135 }
2136 
2137 static const VMStateDescription vmstate_spapr_fwnmi = {
2138     .name = "spapr_fwnmi",
2139     .version_id = 1,
2140     .minimum_version_id = 1,
2141     .needed = spapr_fwnmi_needed,
2142     .pre_save = spapr_fwnmi_pre_save,
2143     .fields = (const VMStateField[]) {
2144         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2145         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2146         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2147         VMSTATE_END_OF_LIST()
2148     },
2149 };
2150 
2151 static const VMStateDescription vmstate_spapr = {
2152     .name = "spapr",
2153     .version_id = 3,
2154     .minimum_version_id = 1,
2155     .pre_load = spapr_pre_load,
2156     .post_load = spapr_post_load,
2157     .pre_save = spapr_pre_save,
2158     .fields = (const VMStateField[]) {
2159         /* used to be @next_irq */
2160         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2161 
2162         /* RTC offset */
2163         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2164 
2165         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2166         VMSTATE_END_OF_LIST()
2167     },
2168     .subsections = (const VMStateDescription * const []) {
2169         &vmstate_spapr_ov5_cas,
2170         &vmstate_spapr_patb_entry,
2171         &vmstate_spapr_pending_events,
2172         &vmstate_spapr_cap_htm,
2173         &vmstate_spapr_cap_vsx,
2174         &vmstate_spapr_cap_dfp,
2175         &vmstate_spapr_cap_cfpc,
2176         &vmstate_spapr_cap_sbbc,
2177         &vmstate_spapr_cap_ibs,
2178         &vmstate_spapr_cap_hpt_maxpagesize,
2179         &vmstate_spapr_irq_map,
2180         &vmstate_spapr_cap_nested_kvm_hv,
2181         &vmstate_spapr_dtb,
2182         &vmstate_spapr_cap_large_decr,
2183         &vmstate_spapr_cap_ccf_assist,
2184         &vmstate_spapr_cap_fwnmi,
2185         &vmstate_spapr_fwnmi,
2186         &vmstate_spapr_cap_rpt_invalidate,
2187         NULL
2188     }
2189 };
2190 
2191 static int htab_save_setup(QEMUFile *f, void *opaque)
2192 {
2193     SpaprMachineState *spapr = opaque;
2194 
2195     /* "Iteration" header */
2196     if (!spapr->htab_shift) {
2197         qemu_put_be32(f, -1);
2198     } else {
2199         qemu_put_be32(f, spapr->htab_shift);
2200     }
2201 
2202     if (spapr->htab) {
2203         spapr->htab_save_index = 0;
2204         spapr->htab_first_pass = true;
2205     } else {
2206         if (spapr->htab_shift) {
2207             assert(kvm_enabled());
2208         }
2209     }
2210 
2211 
2212     return 0;
2213 }
2214 
2215 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2216                             int chunkstart, int n_valid, int n_invalid)
2217 {
2218     qemu_put_be32(f, chunkstart);
2219     qemu_put_be16(f, n_valid);
2220     qemu_put_be16(f, n_invalid);
2221     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2222                     HASH_PTE_SIZE_64 * n_valid);
2223 }
2224 
2225 static void htab_save_end_marker(QEMUFile *f)
2226 {
2227     qemu_put_be32(f, 0);
2228     qemu_put_be16(f, 0);
2229     qemu_put_be16(f, 0);
2230 }
2231 
2232 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2233                                  int64_t max_ns)
2234 {
2235     bool has_timeout = max_ns != -1;
2236     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2237     int index = spapr->htab_save_index;
2238     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2239 
2240     assert(spapr->htab_first_pass);
2241 
2242     do {
2243         int chunkstart;
2244 
2245         /* Consume invalid HPTEs */
2246         while ((index < htabslots)
2247                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2248             CLEAN_HPTE(HPTE(spapr->htab, index));
2249             index++;
2250         }
2251 
2252         /* Consume valid HPTEs */
2253         chunkstart = index;
2254         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2255                && HPTE_VALID(HPTE(spapr->htab, index))) {
2256             CLEAN_HPTE(HPTE(spapr->htab, index));
2257             index++;
2258         }
2259 
2260         if (index > chunkstart) {
2261             int n_valid = index - chunkstart;
2262 
2263             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2264 
2265             if (has_timeout &&
2266                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2267                 break;
2268             }
2269         }
2270     } while ((index < htabslots) && !migration_rate_exceeded(f));
2271 
2272     if (index >= htabslots) {
2273         assert(index == htabslots);
2274         index = 0;
2275         spapr->htab_first_pass = false;
2276     }
2277     spapr->htab_save_index = index;
2278 }
2279 
2280 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2281                                 int64_t max_ns)
2282 {
2283     bool final = max_ns < 0;
2284     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2285     int examined = 0, sent = 0;
2286     int index = spapr->htab_save_index;
2287     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2288 
2289     assert(!spapr->htab_first_pass);
2290 
2291     do {
2292         int chunkstart, invalidstart;
2293 
2294         /* Consume non-dirty HPTEs */
2295         while ((index < htabslots)
2296                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2297             index++;
2298             examined++;
2299         }
2300 
2301         chunkstart = index;
2302         /* Consume valid dirty HPTEs */
2303         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2304                && HPTE_DIRTY(HPTE(spapr->htab, index))
2305                && HPTE_VALID(HPTE(spapr->htab, index))) {
2306             CLEAN_HPTE(HPTE(spapr->htab, index));
2307             index++;
2308             examined++;
2309         }
2310 
2311         invalidstart = index;
2312         /* Consume invalid dirty HPTEs */
2313         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2314                && HPTE_DIRTY(HPTE(spapr->htab, index))
2315                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2316             CLEAN_HPTE(HPTE(spapr->htab, index));
2317             index++;
2318             examined++;
2319         }
2320 
2321         if (index > chunkstart) {
2322             int n_valid = invalidstart - chunkstart;
2323             int n_invalid = index - invalidstart;
2324 
2325             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2326             sent += index - chunkstart;
2327 
2328             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2329                 break;
2330             }
2331         }
2332 
2333         if (examined >= htabslots) {
2334             break;
2335         }
2336 
2337         if (index >= htabslots) {
2338             assert(index == htabslots);
2339             index = 0;
2340         }
2341     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2342 
2343     if (index >= htabslots) {
2344         assert(index == htabslots);
2345         index = 0;
2346     }
2347 
2348     spapr->htab_save_index = index;
2349 
2350     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2351 }
2352 
2353 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2354 #define MAX_KVM_BUF_SIZE    2048
2355 
2356 static int htab_save_iterate(QEMUFile *f, void *opaque)
2357 {
2358     SpaprMachineState *spapr = opaque;
2359     int fd;
2360     int rc = 0;
2361 
2362     /* Iteration header */
2363     if (!spapr->htab_shift) {
2364         qemu_put_be32(f, -1);
2365         return 1;
2366     } else {
2367         qemu_put_be32(f, 0);
2368     }
2369 
2370     if (!spapr->htab) {
2371         assert(kvm_enabled());
2372 
2373         fd = get_htab_fd(spapr);
2374         if (fd < 0) {
2375             return fd;
2376         }
2377 
2378         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2379         if (rc < 0) {
2380             return rc;
2381         }
2382     } else  if (spapr->htab_first_pass) {
2383         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2384     } else {
2385         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2386     }
2387 
2388     htab_save_end_marker(f);
2389 
2390     return rc;
2391 }
2392 
2393 static int htab_save_complete(QEMUFile *f, void *opaque)
2394 {
2395     SpaprMachineState *spapr = opaque;
2396     int fd;
2397 
2398     /* Iteration header */
2399     if (!spapr->htab_shift) {
2400         qemu_put_be32(f, -1);
2401         return 0;
2402     } else {
2403         qemu_put_be32(f, 0);
2404     }
2405 
2406     if (!spapr->htab) {
2407         int rc;
2408 
2409         assert(kvm_enabled());
2410 
2411         fd = get_htab_fd(spapr);
2412         if (fd < 0) {
2413             return fd;
2414         }
2415 
2416         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2417         if (rc < 0) {
2418             return rc;
2419         }
2420     } else {
2421         if (spapr->htab_first_pass) {
2422             htab_save_first_pass(f, spapr, -1);
2423         }
2424         htab_save_later_pass(f, spapr, -1);
2425     }
2426 
2427     /* End marker */
2428     htab_save_end_marker(f);
2429 
2430     return 0;
2431 }
2432 
2433 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2434 {
2435     SpaprMachineState *spapr = opaque;
2436     uint32_t section_hdr;
2437     int fd = -1;
2438     Error *local_err = NULL;
2439 
2440     if (version_id < 1 || version_id > 1) {
2441         error_report("htab_load() bad version");
2442         return -EINVAL;
2443     }
2444 
2445     section_hdr = qemu_get_be32(f);
2446 
2447     if (section_hdr == -1) {
2448         spapr_free_hpt(spapr);
2449         return 0;
2450     }
2451 
2452     if (section_hdr) {
2453         int ret;
2454 
2455         /* First section gives the htab size */
2456         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2457         if (ret < 0) {
2458             error_report_err(local_err);
2459             return ret;
2460         }
2461         return 0;
2462     }
2463 
2464     if (!spapr->htab) {
2465         assert(kvm_enabled());
2466 
2467         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2468         if (fd < 0) {
2469             error_report_err(local_err);
2470             return fd;
2471         }
2472     }
2473 
2474     while (true) {
2475         uint32_t index;
2476         uint16_t n_valid, n_invalid;
2477 
2478         index = qemu_get_be32(f);
2479         n_valid = qemu_get_be16(f);
2480         n_invalid = qemu_get_be16(f);
2481 
2482         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2483             /* End of Stream */
2484             break;
2485         }
2486 
2487         if ((index + n_valid + n_invalid) >
2488             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2489             /* Bad index in stream */
2490             error_report(
2491                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2492                 index, n_valid, n_invalid, spapr->htab_shift);
2493             return -EINVAL;
2494         }
2495 
2496         if (spapr->htab) {
2497             if (n_valid) {
2498                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2499                                 HASH_PTE_SIZE_64 * n_valid);
2500             }
2501             if (n_invalid) {
2502                 memset(HPTE(spapr->htab, index + n_valid), 0,
2503                        HASH_PTE_SIZE_64 * n_invalid);
2504             }
2505         } else {
2506             int rc;
2507 
2508             assert(fd >= 0);
2509 
2510             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2511                                         &local_err);
2512             if (rc < 0) {
2513                 error_report_err(local_err);
2514                 return rc;
2515             }
2516         }
2517     }
2518 
2519     if (!spapr->htab) {
2520         assert(fd >= 0);
2521         close(fd);
2522     }
2523 
2524     return 0;
2525 }
2526 
2527 static void htab_save_cleanup(void *opaque)
2528 {
2529     SpaprMachineState *spapr = opaque;
2530 
2531     close_htab_fd(spapr);
2532 }
2533 
2534 static SaveVMHandlers savevm_htab_handlers = {
2535     .save_setup = htab_save_setup,
2536     .save_live_iterate = htab_save_iterate,
2537     .save_live_complete_precopy = htab_save_complete,
2538     .save_cleanup = htab_save_cleanup,
2539     .load_state = htab_load,
2540 };
2541 
2542 static void spapr_boot_set(void *opaque, const char *boot_device,
2543                            Error **errp)
2544 {
2545     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2546 
2547     g_free(spapr->boot_device);
2548     spapr->boot_device = g_strdup(boot_device);
2549 }
2550 
2551 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2552 {
2553     MachineState *machine = MACHINE(spapr);
2554     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2555     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2556     int i;
2557 
2558     g_assert(!nr_lmbs || machine->device_memory);
2559     for (i = 0; i < nr_lmbs; i++) {
2560         uint64_t addr;
2561 
2562         addr = i * lmb_size + machine->device_memory->base;
2563         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2564                                addr / lmb_size);
2565     }
2566 }
2567 
2568 /*
2569  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2570  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2571  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2572  */
2573 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2574 {
2575     int i;
2576 
2577     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2578         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2579                    " is not aligned to %" PRIu64 " MiB",
2580                    machine->ram_size,
2581                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2582         return;
2583     }
2584 
2585     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2586         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2587                    " is not aligned to %" PRIu64 " MiB",
2588                    machine->ram_size,
2589                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2590         return;
2591     }
2592 
2593     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2594         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2595             error_setg(errp,
2596                        "Node %d memory size 0x%" PRIx64
2597                        " is not aligned to %" PRIu64 " MiB",
2598                        i, machine->numa_state->nodes[i].node_mem,
2599                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2600             return;
2601         }
2602     }
2603 }
2604 
2605 /* find cpu slot in machine->possible_cpus by core_id */
2606 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2607 {
2608     int index = id / ms->smp.threads;
2609 
2610     if (index >= ms->possible_cpus->len) {
2611         return NULL;
2612     }
2613     if (idx) {
2614         *idx = index;
2615     }
2616     return &ms->possible_cpus->cpus[index];
2617 }
2618 
2619 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2620 {
2621     MachineState *ms = MACHINE(spapr);
2622     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2623     Error *local_err = NULL;
2624     bool vsmt_user = !!spapr->vsmt;
2625     int kvm_smt = kvmppc_smt_threads();
2626     int ret;
2627     unsigned int smp_threads = ms->smp.threads;
2628 
2629     if (tcg_enabled()) {
2630         if (smp_threads > 1 &&
2631             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2632                                    spapr->max_compat_pvr)) {
2633             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2634             return;
2635         }
2636 
2637         if (smp_threads > 8) {
2638             error_setg(errp, "TCG cannot support more than 8 threads/core "
2639                        "on a pseries machine");
2640             return;
2641         }
2642     }
2643     if (!is_power_of_2(smp_threads)) {
2644         error_setg(errp, "Cannot support %d threads/core on a pseries "
2645                    "machine because it must be a power of 2", smp_threads);
2646         return;
2647     }
2648 
2649     /* Determine the VSMT mode to use: */
2650     if (vsmt_user) {
2651         if (spapr->vsmt < smp_threads) {
2652             error_setg(errp, "Cannot support VSMT mode %d"
2653                        " because it must be >= threads/core (%d)",
2654                        spapr->vsmt, smp_threads);
2655             return;
2656         }
2657         /* In this case, spapr->vsmt has been set by the command line */
2658     } else if (!smc->smp_threads_vsmt) {
2659         /*
2660          * Default VSMT value is tricky, because we need it to be as
2661          * consistent as possible (for migration), but this requires
2662          * changing it for at least some existing cases.  We pick 8 as
2663          * the value that we'd get with KVM on POWER8, the
2664          * overwhelmingly common case in production systems.
2665          */
2666         spapr->vsmt = MAX(8, smp_threads);
2667     } else {
2668         spapr->vsmt = smp_threads;
2669     }
2670 
2671     /* KVM: If necessary, set the SMT mode: */
2672     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2673         ret = kvmppc_set_smt_threads(spapr->vsmt);
2674         if (ret) {
2675             /* Looks like KVM isn't able to change VSMT mode */
2676             error_setg(&local_err,
2677                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2678                        spapr->vsmt, ret);
2679             /* We can live with that if the default one is big enough
2680              * for the number of threads, and a submultiple of the one
2681              * we want.  In this case we'll waste some vcpu ids, but
2682              * behaviour will be correct */
2683             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2684                 warn_report_err(local_err);
2685             } else {
2686                 if (!vsmt_user) {
2687                     error_append_hint(&local_err,
2688                                       "On PPC, a VM with %d threads/core"
2689                                       " on a host with %d threads/core"
2690                                       " requires the use of VSMT mode %d.\n",
2691                                       smp_threads, kvm_smt, spapr->vsmt);
2692                 }
2693                 kvmppc_error_append_smt_possible_hint(&local_err);
2694                 error_propagate(errp, local_err);
2695             }
2696         }
2697     }
2698     /* else TCG: nothing to do currently */
2699 }
2700 
2701 static void spapr_init_cpus(SpaprMachineState *spapr)
2702 {
2703     MachineState *machine = MACHINE(spapr);
2704     MachineClass *mc = MACHINE_GET_CLASS(machine);
2705     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2706     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2707     const CPUArchIdList *possible_cpus;
2708     unsigned int smp_cpus = machine->smp.cpus;
2709     unsigned int smp_threads = machine->smp.threads;
2710     unsigned int max_cpus = machine->smp.max_cpus;
2711     int boot_cores_nr = smp_cpus / smp_threads;
2712     int i;
2713 
2714     possible_cpus = mc->possible_cpu_arch_ids(machine);
2715     if (mc->has_hotpluggable_cpus) {
2716         if (smp_cpus % smp_threads) {
2717             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2718                          smp_cpus, smp_threads);
2719             exit(1);
2720         }
2721         if (max_cpus % smp_threads) {
2722             error_report("max_cpus (%u) must be multiple of threads (%u)",
2723                          max_cpus, smp_threads);
2724             exit(1);
2725         }
2726     } else {
2727         if (max_cpus != smp_cpus) {
2728             error_report("This machine version does not support CPU hotplug");
2729             exit(1);
2730         }
2731         boot_cores_nr = possible_cpus->len;
2732     }
2733 
2734     if (smc->pre_2_10_has_unused_icps) {
2735         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2736             /* Dummy entries get deregistered when real ICPState objects
2737              * are registered during CPU core hotplug.
2738              */
2739             pre_2_10_vmstate_register_dummy_icp(i);
2740         }
2741     }
2742 
2743     for (i = 0; i < possible_cpus->len; i++) {
2744         int core_id = i * smp_threads;
2745 
2746         if (mc->has_hotpluggable_cpus) {
2747             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2748                                    spapr_vcpu_id(spapr, core_id));
2749         }
2750 
2751         if (i < boot_cores_nr) {
2752             Object *core  = object_new(type);
2753             int nr_threads = smp_threads;
2754 
2755             /* Handle the partially filled core for older machine types */
2756             if ((i + 1) * smp_threads >= smp_cpus) {
2757                 nr_threads = smp_cpus - i * smp_threads;
2758             }
2759 
2760             object_property_set_int(core, "nr-threads", nr_threads,
2761                                     &error_fatal);
2762             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2763                                     &error_fatal);
2764             qdev_realize(DEVICE(core), NULL, &error_fatal);
2765 
2766             object_unref(core);
2767         }
2768     }
2769 }
2770 
2771 static PCIHostState *spapr_create_default_phb(void)
2772 {
2773     DeviceState *dev;
2774 
2775     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2776     qdev_prop_set_uint32(dev, "index", 0);
2777     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2778 
2779     return PCI_HOST_BRIDGE(dev);
2780 }
2781 
2782 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2783 {
2784     MachineState *machine = MACHINE(spapr);
2785     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2786     hwaddr rma_size = machine->ram_size;
2787     hwaddr node0_size = spapr_node0_size(machine);
2788 
2789     /* RMA has to fit in the first NUMA node */
2790     rma_size = MIN(rma_size, node0_size);
2791 
2792     /*
2793      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2794      * never exceed that
2795      */
2796     rma_size = MIN(rma_size, 1 * TiB);
2797 
2798     /*
2799      * Clamp the RMA size based on machine type.  This is for
2800      * migration compatibility with older qemu versions, which limited
2801      * the RMA size for complicated and mostly bad reasons.
2802      */
2803     if (smc->rma_limit) {
2804         rma_size = MIN(rma_size, smc->rma_limit);
2805     }
2806 
2807     if (rma_size < MIN_RMA_SLOF) {
2808         error_setg(errp,
2809                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2810                    "ldMiB guest RMA (Real Mode Area memory)",
2811                    MIN_RMA_SLOF / MiB);
2812         return 0;
2813     }
2814 
2815     return rma_size;
2816 }
2817 
2818 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2819 {
2820     MachineState *machine = MACHINE(spapr);
2821     int i;
2822 
2823     for (i = 0; i < machine->ram_slots; i++) {
2824         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2825     }
2826 }
2827 
2828 /* pSeries LPAR / sPAPR hardware init */
2829 static void spapr_machine_init(MachineState *machine)
2830 {
2831     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2832     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2833     MachineClass *mc = MACHINE_GET_CLASS(machine);
2834     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2835     const char *bios_name = machine->firmware ?: bios_default;
2836     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2837     const char *kernel_filename = machine->kernel_filename;
2838     const char *initrd_filename = machine->initrd_filename;
2839     PCIHostState *phb;
2840     bool has_vga;
2841     int i;
2842     MemoryRegion *sysmem = get_system_memory();
2843     long load_limit, fw_size;
2844     Error *resize_hpt_err = NULL;
2845     NICInfo *nd;
2846 
2847     if (!filename) {
2848         error_report("Could not find LPAR firmware '%s'", bios_name);
2849         exit(1);
2850     }
2851     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2852     if (fw_size <= 0) {
2853         error_report("Could not load LPAR firmware '%s'", filename);
2854         exit(1);
2855     }
2856 
2857     /*
2858      * if Secure VM (PEF) support is configured, then initialize it
2859      */
2860     pef_kvm_init(machine->cgs, &error_fatal);
2861 
2862     msi_nonbroken = true;
2863 
2864     QLIST_INIT(&spapr->phbs);
2865     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2866 
2867     /* Determine capabilities to run with */
2868     spapr_caps_init(spapr);
2869 
2870     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2871     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2872         /*
2873          * If the user explicitly requested a mode we should either
2874          * supply it, or fail completely (which we do below).  But if
2875          * it's not set explicitly, we reset our mode to something
2876          * that works
2877          */
2878         if (resize_hpt_err) {
2879             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2880             error_free(resize_hpt_err);
2881             resize_hpt_err = NULL;
2882         } else {
2883             spapr->resize_hpt = smc->resize_hpt_default;
2884         }
2885     }
2886 
2887     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2888 
2889     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2890         /*
2891          * User requested HPT resize, but this host can't supply it.  Bail out
2892          */
2893         error_report_err(resize_hpt_err);
2894         exit(1);
2895     }
2896     error_free(resize_hpt_err);
2897 
2898     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2899 
2900     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2901     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2902 
2903     /*
2904      * VSMT must be set in order to be able to compute VCPU ids, ie to
2905      * call spapr_max_server_number() or spapr_vcpu_id().
2906      */
2907     spapr_set_vsmt_mode(spapr, &error_fatal);
2908 
2909     /* Set up Interrupt Controller before we create the VCPUs */
2910     spapr_irq_init(spapr, &error_fatal);
2911 
2912     /* Set up containers for ibm,client-architecture-support negotiated options
2913      */
2914     spapr->ov5 = spapr_ovec_new();
2915     spapr->ov5_cas = spapr_ovec_new();
2916 
2917     if (smc->dr_lmb_enabled) {
2918         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2919         spapr_validate_node_memory(machine, &error_fatal);
2920     }
2921 
2922     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2923 
2924     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2925     if (!smc->pre_6_2_numa_affinity) {
2926         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2927     }
2928 
2929     /* advertise support for dedicated HP event source to guests */
2930     if (spapr->use_hotplug_event_source) {
2931         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2932     }
2933 
2934     /* advertise support for HPT resizing */
2935     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2936         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2937     }
2938 
2939     /* advertise support for ibm,dyamic-memory-v2 */
2940     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2941 
2942     /* advertise XIVE on POWER9 machines */
2943     if (spapr->irq->xive) {
2944         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2945     }
2946 
2947     /* init CPUs */
2948     spapr_init_cpus(spapr);
2949 
2950     /* Init numa_assoc_array */
2951     spapr_numa_associativity_init(spapr, machine);
2952 
2953     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2954         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2955                               spapr->max_compat_pvr)) {
2956         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2957         /* KVM and TCG always allow GTSE with radix... */
2958         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2959     }
2960     /* ... but not with hash (currently). */
2961 
2962     if (kvm_enabled()) {
2963         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2964         kvmppc_enable_logical_ci_hcalls();
2965         kvmppc_enable_set_mode_hcall();
2966 
2967         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2968         kvmppc_enable_clear_ref_mod_hcalls();
2969 
2970         /* Enable H_PAGE_INIT */
2971         kvmppc_enable_h_page_init();
2972     }
2973 
2974     /* map RAM */
2975     memory_region_add_subregion(sysmem, 0, machine->ram);
2976 
2977     /* initialize hotplug memory address space */
2978     if (machine->ram_size < machine->maxram_size) {
2979         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2980         hwaddr device_mem_base;
2981 
2982         /*
2983          * Limit the number of hotpluggable memory slots to half the number
2984          * slots that KVM supports, leaving the other half for PCI and other
2985          * devices. However ensure that number of slots doesn't drop below 32.
2986          */
2987         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2988                            SPAPR_MAX_RAM_SLOTS;
2989 
2990         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2991             max_memslots = SPAPR_MAX_RAM_SLOTS;
2992         }
2993         if (machine->ram_slots > max_memslots) {
2994             error_report("Specified number of memory slots %"
2995                          PRIu64" exceeds max supported %d",
2996                          machine->ram_slots, max_memslots);
2997             exit(1);
2998         }
2999 
3000         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
3001         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
3002     }
3003 
3004     if (smc->dr_lmb_enabled) {
3005         spapr_create_lmb_dr_connectors(spapr);
3006     }
3007 
3008     if (mc->nvdimm_supported) {
3009         spapr_create_nvdimm_dr_connectors(spapr);
3010     }
3011 
3012     /* Set up RTAS event infrastructure */
3013     spapr_events_init(spapr);
3014 
3015     /* Set up the RTC RTAS interfaces */
3016     spapr_rtc_create(spapr);
3017 
3018     /* Set up VIO bus */
3019     spapr->vio_bus = spapr_vio_bus_init();
3020 
3021     for (i = 0; serial_hd(i); i++) {
3022         spapr_vty_create(spapr->vio_bus, serial_hd(i));
3023     }
3024 
3025     /* We always have at least the nvram device on VIO */
3026     spapr_create_nvram(spapr);
3027 
3028     /*
3029      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3030      * connectors (described in root DT node's "ibm,drc-types" property)
3031      * are pre-initialized here. additional child connectors (such as
3032      * connectors for a PHBs PCI slots) are added as needed during their
3033      * parent's realization.
3034      */
3035     if (smc->dr_phb_enabled) {
3036         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3037             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3038         }
3039     }
3040 
3041     /* Set up PCI */
3042     spapr_pci_rtas_init();
3043 
3044     phb = spapr_create_default_phb();
3045 
3046     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3047         spapr_vlan_create(spapr->vio_bus, nd);
3048     }
3049 
3050     pci_init_nic_devices(phb->bus, NULL);
3051 
3052     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3053         spapr_vscsi_create(spapr->vio_bus);
3054     }
3055 
3056     /* Graphics */
3057     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3058     if (has_vga) {
3059         spapr->want_stdout_path = !machine->enable_graphics;
3060         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3061     } else {
3062         spapr->want_stdout_path = true;
3063     }
3064 
3065     if (machine->usb) {
3066         if (smc->use_ohci_by_default) {
3067             pci_create_simple(phb->bus, -1, "pci-ohci");
3068         } else {
3069             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3070         }
3071 
3072         if (has_vga) {
3073             USBBus *usb_bus;
3074 
3075             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3076                                                               &error_abort));
3077             usb_create_simple(usb_bus, "usb-kbd");
3078             usb_create_simple(usb_bus, "usb-mouse");
3079         }
3080     }
3081 
3082     if (kernel_filename) {
3083         uint64_t loaded_addr = 0;
3084 
3085         spapr->kernel_size = load_elf(kernel_filename, NULL,
3086                                       translate_kernel_address, spapr,
3087                                       NULL, &loaded_addr, NULL, NULL, 1,
3088                                       PPC_ELF_MACHINE, 0, 0);
3089         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3090             spapr->kernel_size = load_elf(kernel_filename, NULL,
3091                                           translate_kernel_address, spapr,
3092                                           NULL, &loaded_addr, NULL, NULL, 0,
3093                                           PPC_ELF_MACHINE, 0, 0);
3094             spapr->kernel_le = spapr->kernel_size > 0;
3095         }
3096         if (spapr->kernel_size < 0) {
3097             error_report("error loading %s: %s", kernel_filename,
3098                          load_elf_strerror(spapr->kernel_size));
3099             exit(1);
3100         }
3101 
3102         if (spapr->kernel_addr != loaded_addr) {
3103             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3104                         " to 0x%"PRIx64,
3105                         spapr->kernel_addr, loaded_addr);
3106             spapr->kernel_addr = loaded_addr;
3107         }
3108 
3109         /* load initrd */
3110         if (initrd_filename) {
3111             /* Try to locate the initrd in the gap between the kernel
3112              * and the firmware. Add a bit of space just in case
3113              */
3114             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3115                                   + 0x1ffff) & ~0xffff;
3116             spapr->initrd_size = load_image_targphys(initrd_filename,
3117                                                      spapr->initrd_base,
3118                                                      load_limit
3119                                                      - spapr->initrd_base);
3120             if (spapr->initrd_size < 0) {
3121                 error_report("could not load initial ram disk '%s'",
3122                              initrd_filename);
3123                 exit(1);
3124             }
3125         }
3126     }
3127 
3128     /* FIXME: Should register things through the MachineState's qdev
3129      * interface, this is a legacy from the sPAPREnvironment structure
3130      * which predated MachineState but had a similar function */
3131     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3132     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3133                          &savevm_htab_handlers, spapr);
3134 
3135     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3136 
3137     qemu_register_boot_set(spapr_boot_set, spapr);
3138 
3139     /*
3140      * Nothing needs to be done to resume a suspended guest because
3141      * suspending does not change the machine state, so no need for
3142      * a ->wakeup method.
3143      */
3144     qemu_register_wakeup_support();
3145 
3146     if (kvm_enabled()) {
3147         /* to stop and start vmclock */
3148         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3149                                          &spapr->tb);
3150 
3151         kvmppc_spapr_enable_inkernel_multitce();
3152     }
3153 
3154     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3155     if (spapr->vof) {
3156         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3157         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3158     }
3159 
3160     spapr_watchdog_init(spapr);
3161 }
3162 
3163 #define DEFAULT_KVM_TYPE "auto"
3164 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3165 {
3166     /*
3167      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3168      * accommodate the 'HV' and 'PV' formats that exists in the
3169      * wild. The 'auto' mode is being introduced already as
3170      * lower-case, thus we don't need to bother checking for
3171      * "AUTO".
3172      */
3173     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3174         return 0;
3175     }
3176 
3177     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3178         return 1;
3179     }
3180 
3181     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3182         return 2;
3183     }
3184 
3185     error_report("Unknown kvm-type specified '%s'", vm_type);
3186     return -1;
3187 }
3188 
3189 /*
3190  * Implementation of an interface to adjust firmware path
3191  * for the bootindex property handling.
3192  */
3193 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3194                                    DeviceState *dev)
3195 {
3196 #define CAST(type, obj, name) \
3197     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3198     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3199     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3200     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3201     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3202 
3203     if (d && bus) {
3204         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3205         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3206         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3207 
3208         if (spapr) {
3209             /*
3210              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3211              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3212              * 0x8000 | (target << 8) | (bus << 5) | lun
3213              * (see the "Logical unit addressing format" table in SAM5)
3214              */
3215             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3216             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3217                                    (uint64_t)id << 48);
3218         } else if (virtio) {
3219             /*
3220              * We use SRP luns of the form 01000000 | (target << 8) | lun
3221              * in the top 32 bits of the 64-bit LUN
3222              * Note: the quote above is from SLOF and it is wrong,
3223              * the actual binding is:
3224              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3225              */
3226             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3227             if (d->lun >= 256) {
3228                 /* Use the LUN "flat space addressing method" */
3229                 id |= 0x4000;
3230             }
3231             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3232                                    (uint64_t)id << 32);
3233         } else if (usb) {
3234             /*
3235              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3236              * in the top 32 bits of the 64-bit LUN
3237              */
3238             unsigned usb_port = atoi(usb->port->path);
3239             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3240             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3241                                    (uint64_t)id << 32);
3242         }
3243     }
3244 
3245     /*
3246      * SLOF probes the USB devices, and if it recognizes that the device is a
3247      * storage device, it changes its name to "storage" instead of "usb-host",
3248      * and additionally adds a child node for the SCSI LUN, so the correct
3249      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3250      */
3251     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3252         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3253         if (usb_device_is_scsi_storage(usbdev)) {
3254             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3255         }
3256     }
3257 
3258     if (phb) {
3259         /* Replace "pci" with "pci@800000020000000" */
3260         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3261     }
3262 
3263     if (vsc) {
3264         /* Same logic as virtio above */
3265         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3266         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3267     }
3268 
3269     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3270         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3271         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3272         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3273     }
3274 
3275     if (pcidev) {
3276         return spapr_pci_fw_dev_name(pcidev);
3277     }
3278 
3279     return NULL;
3280 }
3281 
3282 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3283 {
3284     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3285 
3286     return g_strdup(spapr->kvm_type);
3287 }
3288 
3289 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3290 {
3291     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3292 
3293     g_free(spapr->kvm_type);
3294     spapr->kvm_type = g_strdup(value);
3295 }
3296 
3297 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3298 {
3299     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3300 
3301     return spapr->use_hotplug_event_source;
3302 }
3303 
3304 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3305                                             Error **errp)
3306 {
3307     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3308 
3309     spapr->use_hotplug_event_source = value;
3310 }
3311 
3312 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3313 {
3314     return true;
3315 }
3316 
3317 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3318 {
3319     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3320 
3321     switch (spapr->resize_hpt) {
3322     case SPAPR_RESIZE_HPT_DEFAULT:
3323         return g_strdup("default");
3324     case SPAPR_RESIZE_HPT_DISABLED:
3325         return g_strdup("disabled");
3326     case SPAPR_RESIZE_HPT_ENABLED:
3327         return g_strdup("enabled");
3328     case SPAPR_RESIZE_HPT_REQUIRED:
3329         return g_strdup("required");
3330     }
3331     g_assert_not_reached();
3332 }
3333 
3334 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3335 {
3336     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3337 
3338     if (strcmp(value, "default") == 0) {
3339         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3340     } else if (strcmp(value, "disabled") == 0) {
3341         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3342     } else if (strcmp(value, "enabled") == 0) {
3343         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3344     } else if (strcmp(value, "required") == 0) {
3345         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3346     } else {
3347         error_setg(errp, "Bad value for \"resize-hpt\" property");
3348     }
3349 }
3350 
3351 static bool spapr_get_vof(Object *obj, Error **errp)
3352 {
3353     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3354 
3355     return spapr->vof != NULL;
3356 }
3357 
3358 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3359 {
3360     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3361 
3362     if (spapr->vof) {
3363         vof_cleanup(spapr->vof);
3364         g_free(spapr->vof);
3365         spapr->vof = NULL;
3366     }
3367     if (!value) {
3368         return;
3369     }
3370     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3371 }
3372 
3373 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3374 {
3375     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3376 
3377     if (spapr->irq == &spapr_irq_xics_legacy) {
3378         return g_strdup("legacy");
3379     } else if (spapr->irq == &spapr_irq_xics) {
3380         return g_strdup("xics");
3381     } else if (spapr->irq == &spapr_irq_xive) {
3382         return g_strdup("xive");
3383     } else if (spapr->irq == &spapr_irq_dual) {
3384         return g_strdup("dual");
3385     }
3386     g_assert_not_reached();
3387 }
3388 
3389 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3390 {
3391     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3392 
3393     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3394         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3395         return;
3396     }
3397 
3398     /* The legacy IRQ backend can not be set */
3399     if (strcmp(value, "xics") == 0) {
3400         spapr->irq = &spapr_irq_xics;
3401     } else if (strcmp(value, "xive") == 0) {
3402         spapr->irq = &spapr_irq_xive;
3403     } else if (strcmp(value, "dual") == 0) {
3404         spapr->irq = &spapr_irq_dual;
3405     } else {
3406         error_setg(errp, "Bad value for \"ic-mode\" property");
3407     }
3408 }
3409 
3410 static char *spapr_get_host_model(Object *obj, Error **errp)
3411 {
3412     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3413 
3414     return g_strdup(spapr->host_model);
3415 }
3416 
3417 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3418 {
3419     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3420 
3421     g_free(spapr->host_model);
3422     spapr->host_model = g_strdup(value);
3423 }
3424 
3425 static char *spapr_get_host_serial(Object *obj, Error **errp)
3426 {
3427     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3428 
3429     return g_strdup(spapr->host_serial);
3430 }
3431 
3432 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3433 {
3434     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3435 
3436     g_free(spapr->host_serial);
3437     spapr->host_serial = g_strdup(value);
3438 }
3439 
3440 static void spapr_instance_init(Object *obj)
3441 {
3442     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3443     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3444     MachineState *ms = MACHINE(spapr);
3445     MachineClass *mc = MACHINE_GET_CLASS(ms);
3446 
3447     /*
3448      * NVDIMM support went live in 5.1 without considering that, in
3449      * other archs, the user needs to enable NVDIMM support with the
3450      * 'nvdimm' machine option and the default behavior is NVDIMM
3451      * support disabled. It is too late to roll back to the standard
3452      * behavior without breaking 5.1 guests.
3453      */
3454     if (mc->nvdimm_supported) {
3455         ms->nvdimms_state->is_enabled = true;
3456     }
3457 
3458     spapr->htab_fd = -1;
3459     spapr->use_hotplug_event_source = true;
3460     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3461     object_property_add_str(obj, "kvm-type",
3462                             spapr_get_kvm_type, spapr_set_kvm_type);
3463     object_property_set_description(obj, "kvm-type",
3464                                     "Specifies the KVM virtualization mode (auto,"
3465                                     " hv, pr). Defaults to 'auto'. This mode will use"
3466                                     " any available KVM module loaded in the host,"
3467                                     " where kvm_hv takes precedence if both kvm_hv and"
3468                                     " kvm_pr are loaded.");
3469     object_property_add_bool(obj, "modern-hotplug-events",
3470                             spapr_get_modern_hotplug_events,
3471                             spapr_set_modern_hotplug_events);
3472     object_property_set_description(obj, "modern-hotplug-events",
3473                                     "Use dedicated hotplug event mechanism in"
3474                                     " place of standard EPOW events when possible"
3475                                     " (required for memory hot-unplug support)");
3476     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3477                             "Maximum permitted CPU compatibility mode");
3478 
3479     object_property_add_str(obj, "resize-hpt",
3480                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3481     object_property_set_description(obj, "resize-hpt",
3482                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3483     object_property_add_uint32_ptr(obj, "vsmt",
3484                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3485     object_property_set_description(obj, "vsmt",
3486                                     "Virtual SMT: KVM behaves as if this were"
3487                                     " the host's SMT mode");
3488 
3489     object_property_add_bool(obj, "vfio-no-msix-emulation",
3490                              spapr_get_msix_emulation, NULL);
3491 
3492     object_property_add_uint64_ptr(obj, "kernel-addr",
3493                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3494     object_property_set_description(obj, "kernel-addr",
3495                                     stringify(KERNEL_LOAD_ADDR)
3496                                     " for -kernel is the default");
3497     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3498 
3499     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3500     object_property_set_description(obj, "x-vof",
3501                                     "Enable Virtual Open Firmware (experimental)");
3502 
3503     /* The machine class defines the default interrupt controller mode */
3504     spapr->irq = smc->irq;
3505     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3506                             spapr_set_ic_mode);
3507     object_property_set_description(obj, "ic-mode",
3508                  "Specifies the interrupt controller mode (xics, xive, dual)");
3509 
3510     object_property_add_str(obj, "host-model",
3511         spapr_get_host_model, spapr_set_host_model);
3512     object_property_set_description(obj, "host-model",
3513         "Host model to advertise in guest device tree");
3514     object_property_add_str(obj, "host-serial",
3515         spapr_get_host_serial, spapr_set_host_serial);
3516     object_property_set_description(obj, "host-serial",
3517         "Host serial number to advertise in guest device tree");
3518 }
3519 
3520 static void spapr_machine_finalizefn(Object *obj)
3521 {
3522     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3523 
3524     g_free(spapr->kvm_type);
3525 }
3526 
3527 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3528 {
3529     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3530     PowerPCCPU *cpu = POWERPC_CPU(cs);
3531     CPUPPCState *env = &cpu->env;
3532 
3533     cpu_synchronize_state(cs);
3534     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3535     if (spapr->fwnmi_system_reset_addr != -1) {
3536         uint64_t rtas_addr, addr;
3537 
3538         /* get rtas addr from fdt */
3539         rtas_addr = spapr_get_rtas_addr();
3540         if (!rtas_addr) {
3541             qemu_system_guest_panicked(NULL);
3542             return;
3543         }
3544 
3545         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3546         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3547         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3548         env->gpr[3] = addr;
3549     }
3550     ppc_cpu_do_system_reset(cs);
3551     if (spapr->fwnmi_system_reset_addr != -1) {
3552         env->nip = spapr->fwnmi_system_reset_addr;
3553     }
3554 }
3555 
3556 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3557 {
3558     CPUState *cs;
3559 
3560     CPU_FOREACH(cs) {
3561         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3562     }
3563 }
3564 
3565 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3566                           void *fdt, int *fdt_start_offset, Error **errp)
3567 {
3568     uint64_t addr;
3569     uint32_t node;
3570 
3571     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3572     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3573                                     &error_abort);
3574     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3575                                              SPAPR_MEMORY_BLOCK_SIZE);
3576     return 0;
3577 }
3578 
3579 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3580                            bool dedicated_hp_event_source)
3581 {
3582     SpaprDrc *drc;
3583     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3584     int i;
3585     uint64_t addr = addr_start;
3586     bool hotplugged = spapr_drc_hotplugged(dev);
3587 
3588     for (i = 0; i < nr_lmbs; i++) {
3589         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3590                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3591         g_assert(drc);
3592 
3593         /*
3594          * memory_device_get_free_addr() provided a range of free addresses
3595          * that doesn't overlap with any existing mapping at pre-plug. The
3596          * corresponding LMB DRCs are thus assumed to be all attachable.
3597          */
3598         spapr_drc_attach(drc, dev);
3599         if (!hotplugged) {
3600             spapr_drc_reset(drc);
3601         }
3602         addr += SPAPR_MEMORY_BLOCK_SIZE;
3603     }
3604     /* send hotplug notification to the
3605      * guest only in case of hotplugged memory
3606      */
3607     if (hotplugged) {
3608         if (dedicated_hp_event_source) {
3609             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3610                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3611             g_assert(drc);
3612             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3613                                                    nr_lmbs,
3614                                                    spapr_drc_index(drc));
3615         } else {
3616             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3617                                            nr_lmbs);
3618         }
3619     }
3620 }
3621 
3622 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3623 {
3624     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3625     PCDIMMDevice *dimm = PC_DIMM(dev);
3626     uint64_t size, addr;
3627     int64_t slot;
3628     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3629 
3630     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3631 
3632     pc_dimm_plug(dimm, MACHINE(ms));
3633 
3634     if (!is_nvdimm) {
3635         addr = object_property_get_uint(OBJECT(dimm),
3636                                         PC_DIMM_ADDR_PROP, &error_abort);
3637         spapr_add_lmbs(dev, addr, size,
3638                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3639     } else {
3640         slot = object_property_get_int(OBJECT(dimm),
3641                                        PC_DIMM_SLOT_PROP, &error_abort);
3642         /* We should have valid slot number at this point */
3643         g_assert(slot >= 0);
3644         spapr_add_nvdimm(dev, slot);
3645     }
3646 }
3647 
3648 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3649                                   Error **errp)
3650 {
3651     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3652     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3653     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3654     PCDIMMDevice *dimm = PC_DIMM(dev);
3655     Error *local_err = NULL;
3656     uint64_t size;
3657     Object *memdev;
3658     hwaddr pagesize;
3659 
3660     if (!smc->dr_lmb_enabled) {
3661         error_setg(errp, "Memory hotplug not supported for this machine");
3662         return;
3663     }
3664 
3665     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3666     if (local_err) {
3667         error_propagate(errp, local_err);
3668         return;
3669     }
3670 
3671     if (is_nvdimm) {
3672         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3673             return;
3674         }
3675     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3676         error_setg(errp, "Hotplugged memory size must be a multiple of "
3677                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3678         return;
3679     }
3680 
3681     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3682                                       &error_abort);
3683     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3684     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3685         return;
3686     }
3687 
3688     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3689 }
3690 
3691 struct SpaprDimmState {
3692     PCDIMMDevice *dimm;
3693     uint32_t nr_lmbs;
3694     QTAILQ_ENTRY(SpaprDimmState) next;
3695 };
3696 
3697 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3698                                                        PCDIMMDevice *dimm)
3699 {
3700     SpaprDimmState *dimm_state = NULL;
3701 
3702     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3703         if (dimm_state->dimm == dimm) {
3704             break;
3705         }
3706     }
3707     return dimm_state;
3708 }
3709 
3710 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3711                                                       uint32_t nr_lmbs,
3712                                                       PCDIMMDevice *dimm)
3713 {
3714     SpaprDimmState *ds = NULL;
3715 
3716     /*
3717      * If this request is for a DIMM whose removal had failed earlier
3718      * (due to guest's refusal to remove the LMBs), we would have this
3719      * dimm already in the pending_dimm_unplugs list. In that
3720      * case don't add again.
3721      */
3722     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3723     if (!ds) {
3724         ds = g_new0(SpaprDimmState, 1);
3725         ds->nr_lmbs = nr_lmbs;
3726         ds->dimm = dimm;
3727         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3728     }
3729     return ds;
3730 }
3731 
3732 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3733                                               SpaprDimmState *dimm_state)
3734 {
3735     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3736     g_free(dimm_state);
3737 }
3738 
3739 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3740                                                         PCDIMMDevice *dimm)
3741 {
3742     SpaprDrc *drc;
3743     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3744                                                   &error_abort);
3745     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3746     uint32_t avail_lmbs = 0;
3747     uint64_t addr_start, addr;
3748     int i;
3749 
3750     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3751                                           &error_abort);
3752 
3753     addr = addr_start;
3754     for (i = 0; i < nr_lmbs; i++) {
3755         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3756                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3757         g_assert(drc);
3758         if (drc->dev) {
3759             avail_lmbs++;
3760         }
3761         addr += SPAPR_MEMORY_BLOCK_SIZE;
3762     }
3763 
3764     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3765 }
3766 
3767 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3768 {
3769     SpaprDimmState *ds;
3770     PCDIMMDevice *dimm;
3771     SpaprDrc *drc;
3772     uint32_t nr_lmbs;
3773     uint64_t size, addr_start, addr;
3774     g_autofree char *qapi_error = NULL;
3775     int i;
3776 
3777     if (!dev) {
3778         return;
3779     }
3780 
3781     dimm = PC_DIMM(dev);
3782     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3783 
3784     /*
3785      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3786      * unplug state, but one of its DRC is marked as unplug_requested.
3787      * This is bad and weird enough to g_assert() out.
3788      */
3789     g_assert(ds);
3790 
3791     spapr_pending_dimm_unplugs_remove(spapr, ds);
3792 
3793     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3794     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3795 
3796     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3797                                           &error_abort);
3798 
3799     addr = addr_start;
3800     for (i = 0; i < nr_lmbs; i++) {
3801         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3802                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3803         g_assert(drc);
3804 
3805         drc->unplug_requested = false;
3806         addr += SPAPR_MEMORY_BLOCK_SIZE;
3807     }
3808 
3809     /*
3810      * Tell QAPI that something happened and the memory
3811      * hotunplug wasn't successful. Keep sending
3812      * MEM_UNPLUG_ERROR even while sending
3813      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3814      * MEM_UNPLUG_ERROR is due.
3815      */
3816     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3817                                  "for device %s", dev->id);
3818 
3819     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3820 
3821     qapi_event_send_device_unplug_guest_error(dev->id,
3822                                               dev->canonical_path);
3823 }
3824 
3825 /* Callback to be called during DRC release. */
3826 void spapr_lmb_release(DeviceState *dev)
3827 {
3828     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3829     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3830     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3831 
3832     /* This information will get lost if a migration occurs
3833      * during the unplug process. In this case recover it. */
3834     if (ds == NULL) {
3835         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3836         g_assert(ds);
3837         /* The DRC being examined by the caller at least must be counted */
3838         g_assert(ds->nr_lmbs);
3839     }
3840 
3841     if (--ds->nr_lmbs) {
3842         return;
3843     }
3844 
3845     /*
3846      * Now that all the LMBs have been removed by the guest, call the
3847      * unplug handler chain. This can never fail.
3848      */
3849     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3850     object_unparent(OBJECT(dev));
3851 }
3852 
3853 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3854 {
3855     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3856     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3857 
3858     /* We really shouldn't get this far without anything to unplug */
3859     g_assert(ds);
3860 
3861     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3862     qdev_unrealize(dev);
3863     spapr_pending_dimm_unplugs_remove(spapr, ds);
3864 }
3865 
3866 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3867                                         DeviceState *dev, Error **errp)
3868 {
3869     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3870     PCDIMMDevice *dimm = PC_DIMM(dev);
3871     uint32_t nr_lmbs;
3872     uint64_t size, addr_start, addr;
3873     int i;
3874     SpaprDrc *drc;
3875 
3876     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3877         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3878         return;
3879     }
3880 
3881     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3882     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3883 
3884     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3885                                           &error_abort);
3886 
3887     /*
3888      * An existing pending dimm state for this DIMM means that there is an
3889      * unplug operation in progress, waiting for the spapr_lmb_release
3890      * callback to complete the job (BQL can't cover that far). In this case,
3891      * bail out to avoid detaching DRCs that were already released.
3892      */
3893     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3894         error_setg(errp, "Memory unplug already in progress for device %s",
3895                    dev->id);
3896         return;
3897     }
3898 
3899     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3900 
3901     addr = addr_start;
3902     for (i = 0; i < nr_lmbs; i++) {
3903         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3904                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3905         g_assert(drc);
3906 
3907         spapr_drc_unplug_request(drc);
3908         addr += SPAPR_MEMORY_BLOCK_SIZE;
3909     }
3910 
3911     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3912                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3913     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3914                                               nr_lmbs, spapr_drc_index(drc));
3915 }
3916 
3917 /* Callback to be called during DRC release. */
3918 void spapr_core_release(DeviceState *dev)
3919 {
3920     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3921 
3922     /* Call the unplug handler chain. This can never fail. */
3923     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3924     object_unparent(OBJECT(dev));
3925 }
3926 
3927 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3928 {
3929     MachineState *ms = MACHINE(hotplug_dev);
3930     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3931     CPUCore *cc = CPU_CORE(dev);
3932     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3933 
3934     if (smc->pre_2_10_has_unused_icps) {
3935         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3936         int i;
3937 
3938         for (i = 0; i < cc->nr_threads; i++) {
3939             CPUState *cs = CPU(sc->threads[i]);
3940 
3941             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3942         }
3943     }
3944 
3945     assert(core_slot);
3946     core_slot->cpu = NULL;
3947     qdev_unrealize(dev);
3948 }
3949 
3950 static
3951 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3952                                Error **errp)
3953 {
3954     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3955     int index;
3956     SpaprDrc *drc;
3957     CPUCore *cc = CPU_CORE(dev);
3958 
3959     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3960         error_setg(errp, "Unable to find CPU core with core-id: %d",
3961                    cc->core_id);
3962         return;
3963     }
3964     if (index == 0) {
3965         error_setg(errp, "Boot CPU core may not be unplugged");
3966         return;
3967     }
3968 
3969     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3970                           spapr_vcpu_id(spapr, cc->core_id));
3971     g_assert(drc);
3972 
3973     if (!spapr_drc_unplug_requested(drc)) {
3974         spapr_drc_unplug_request(drc);
3975     }
3976 
3977     /*
3978      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3979      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3980      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3981      * attempt (e.g. the kernel will refuse to remove the last online
3982      * CPU), we will never attempt it again because unplug_requested
3983      * will still be 'true' in that case.
3984      */
3985     spapr_hotplug_req_remove_by_index(drc);
3986 }
3987 
3988 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3989                            void *fdt, int *fdt_start_offset, Error **errp)
3990 {
3991     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3992     CPUState *cs = CPU(core->threads[0]);
3993     PowerPCCPU *cpu = POWERPC_CPU(cs);
3994     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3995     int id = spapr_get_vcpu_id(cpu);
3996     g_autofree char *nodename = NULL;
3997     int offset;
3998 
3999     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
4000     offset = fdt_add_subnode(fdt, 0, nodename);
4001 
4002     spapr_dt_cpu(cs, fdt, offset, spapr);
4003 
4004     /*
4005      * spapr_dt_cpu() does not fill the 'name' property in the
4006      * CPU node. The function is called during boot process, before
4007      * and after CAS, and overwriting the 'name' property written
4008      * by SLOF is not allowed.
4009      *
4010      * Write it manually after spapr_dt_cpu(). This makes the hotplug
4011      * CPUs more compatible with the coldplugged ones, which have
4012      * the 'name' property. Linux Kernel also relies on this
4013      * property to identify CPU nodes.
4014      */
4015     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
4016 
4017     *fdt_start_offset = offset;
4018     return 0;
4019 }
4020 
4021 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4022 {
4023     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4024     MachineClass *mc = MACHINE_GET_CLASS(spapr);
4025     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4026     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
4027     CPUCore *cc = CPU_CORE(dev);
4028     CPUState *cs;
4029     SpaprDrc *drc;
4030     CPUArchId *core_slot;
4031     int index;
4032     bool hotplugged = spapr_drc_hotplugged(dev);
4033     int i;
4034 
4035     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4036     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4037 
4038     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4039                           spapr_vcpu_id(spapr, cc->core_id));
4040 
4041     g_assert(drc || !mc->has_hotpluggable_cpus);
4042 
4043     if (drc) {
4044         /*
4045          * spapr_core_pre_plug() already buys us this is a brand new
4046          * core being plugged into a free slot. Nothing should already
4047          * be attached to the corresponding DRC.
4048          */
4049         spapr_drc_attach(drc, dev);
4050 
4051         if (hotplugged) {
4052             /*
4053              * Send hotplug notification interrupt to the guest only
4054              * in case of hotplugged CPUs.
4055              */
4056             spapr_hotplug_req_add_by_index(drc);
4057         } else {
4058             spapr_drc_reset(drc);
4059         }
4060     }
4061 
4062     core_slot->cpu = OBJECT(dev);
4063 
4064     /*
4065      * Set compatibility mode to match the boot CPU, which was either set
4066      * by the machine reset code or by CAS. This really shouldn't fail at
4067      * this point.
4068      */
4069     if (hotplugged) {
4070         for (i = 0; i < cc->nr_threads; i++) {
4071             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4072                            &error_abort);
4073         }
4074     }
4075 
4076     if (smc->pre_2_10_has_unused_icps) {
4077         for (i = 0; i < cc->nr_threads; i++) {
4078             cs = CPU(core->threads[i]);
4079             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4080         }
4081     }
4082 }
4083 
4084 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4085                                 Error **errp)
4086 {
4087     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4088     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4089     CPUCore *cc = CPU_CORE(dev);
4090     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4091     const char *type = object_get_typename(OBJECT(dev));
4092     CPUArchId *core_slot;
4093     int index;
4094     unsigned int smp_threads = machine->smp.threads;
4095 
4096     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4097         error_setg(errp, "CPU hotplug not supported for this machine");
4098         return;
4099     }
4100 
4101     if (strcmp(base_core_type, type)) {
4102         error_setg(errp, "CPU core type should be %s", base_core_type);
4103         return;
4104     }
4105 
4106     if (cc->core_id % smp_threads) {
4107         error_setg(errp, "invalid core id %d", cc->core_id);
4108         return;
4109     }
4110 
4111     /*
4112      * In general we should have homogeneous threads-per-core, but old
4113      * (pre hotplug support) machine types allow the last core to have
4114      * reduced threads as a compatibility hack for when we allowed
4115      * total vcpus not a multiple of threads-per-core.
4116      */
4117     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4118         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4119                    smp_threads);
4120         return;
4121     }
4122 
4123     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4124     if (!core_slot) {
4125         error_setg(errp, "core id %d out of range", cc->core_id);
4126         return;
4127     }
4128 
4129     if (core_slot->cpu) {
4130         error_setg(errp, "core %d already populated", cc->core_id);
4131         return;
4132     }
4133 
4134     numa_cpu_pre_plug(core_slot, dev, errp);
4135 }
4136 
4137 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4138                           void *fdt, int *fdt_start_offset, Error **errp)
4139 {
4140     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4141     int intc_phandle;
4142 
4143     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4144     if (intc_phandle <= 0) {
4145         return -1;
4146     }
4147 
4148     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4149         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4150         return -1;
4151     }
4152 
4153     /* generally SLOF creates these, for hotplug it's up to QEMU */
4154     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4155 
4156     return 0;
4157 }
4158 
4159 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4160                                Error **errp)
4161 {
4162     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4163     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4164     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4165     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4166     SpaprDrc *drc;
4167 
4168     if (dev->hotplugged && !smc->dr_phb_enabled) {
4169         error_setg(errp, "PHB hotplug not supported for this machine");
4170         return false;
4171     }
4172 
4173     if (sphb->index == (uint32_t)-1) {
4174         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4175         return false;
4176     }
4177 
4178     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4179     if (drc && drc->dev) {
4180         error_setg(errp, "PHB %d already attached", sphb->index);
4181         return false;
4182     }
4183 
4184     /*
4185      * This will check that sphb->index doesn't exceed the maximum number of
4186      * PHBs for the current machine type.
4187      */
4188     return
4189         smc->phb_placement(spapr, sphb->index,
4190                            &sphb->buid, &sphb->io_win_addr,
4191                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4192                            windows_supported, sphb->dma_liobn,
4193                            errp);
4194 }
4195 
4196 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4197 {
4198     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4199     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4200     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4201     SpaprDrc *drc;
4202     bool hotplugged = spapr_drc_hotplugged(dev);
4203 
4204     if (!smc->dr_phb_enabled) {
4205         return;
4206     }
4207 
4208     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4209     /* hotplug hooks should check it's enabled before getting this far */
4210     assert(drc);
4211 
4212     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4213     spapr_drc_attach(drc, dev);
4214 
4215     if (hotplugged) {
4216         spapr_hotplug_req_add_by_index(drc);
4217     } else {
4218         spapr_drc_reset(drc);
4219     }
4220 }
4221 
4222 void spapr_phb_release(DeviceState *dev)
4223 {
4224     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4225 
4226     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4227     object_unparent(OBJECT(dev));
4228 }
4229 
4230 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4231 {
4232     qdev_unrealize(dev);
4233 }
4234 
4235 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4236                                      DeviceState *dev, Error **errp)
4237 {
4238     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4239     SpaprDrc *drc;
4240 
4241     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4242     assert(drc);
4243 
4244     if (!spapr_drc_unplug_requested(drc)) {
4245         spapr_drc_unplug_request(drc);
4246         spapr_hotplug_req_remove_by_index(drc);
4247     } else {
4248         error_setg(errp,
4249                    "PCI Host Bridge unplug already in progress for device %s",
4250                    dev->id);
4251     }
4252 }
4253 
4254 static
4255 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4256                               Error **errp)
4257 {
4258     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4259 
4260     if (spapr->tpm_proxy != NULL) {
4261         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4262         return false;
4263     }
4264 
4265     return true;
4266 }
4267 
4268 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4269 {
4270     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4271     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4272 
4273     /* Already checked in spapr_tpm_proxy_pre_plug() */
4274     g_assert(spapr->tpm_proxy == NULL);
4275 
4276     spapr->tpm_proxy = tpm_proxy;
4277 }
4278 
4279 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4280 {
4281     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4282 
4283     qdev_unrealize(dev);
4284     object_unparent(OBJECT(dev));
4285     spapr->tpm_proxy = NULL;
4286 }
4287 
4288 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4289                                       DeviceState *dev, Error **errp)
4290 {
4291     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4292         spapr_memory_plug(hotplug_dev, dev);
4293     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4294         spapr_core_plug(hotplug_dev, dev);
4295     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4296         spapr_phb_plug(hotplug_dev, dev);
4297     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4298         spapr_tpm_proxy_plug(hotplug_dev, dev);
4299     }
4300 }
4301 
4302 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4303                                         DeviceState *dev, Error **errp)
4304 {
4305     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4306         spapr_memory_unplug(hotplug_dev, dev);
4307     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4308         spapr_core_unplug(hotplug_dev, dev);
4309     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4310         spapr_phb_unplug(hotplug_dev, dev);
4311     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4312         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4313     }
4314 }
4315 
4316 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4317 {
4318     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4319         /*
4320          * CAS will process all pending unplug requests.
4321          *
4322          * HACK: a guest could theoretically have cleared all bits in OV5,
4323          * but none of the guests we care for do.
4324          */
4325         spapr_ovec_empty(spapr->ov5_cas);
4326 }
4327 
4328 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4329                                                 DeviceState *dev, Error **errp)
4330 {
4331     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4332     MachineClass *mc = MACHINE_GET_CLASS(sms);
4333     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4334 
4335     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4336         if (spapr_memory_hot_unplug_supported(sms)) {
4337             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4338         } else {
4339             error_setg(errp, "Memory hot unplug not supported for this guest");
4340         }
4341     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4342         if (!mc->has_hotpluggable_cpus) {
4343             error_setg(errp, "CPU hot unplug not supported on this machine");
4344             return;
4345         }
4346         spapr_core_unplug_request(hotplug_dev, dev, errp);
4347     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4348         if (!smc->dr_phb_enabled) {
4349             error_setg(errp, "PHB hot unplug not supported on this machine");
4350             return;
4351         }
4352         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4353     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4354         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4355     }
4356 }
4357 
4358 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4359                                           DeviceState *dev, Error **errp)
4360 {
4361     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4362         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4363     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4364         spapr_core_pre_plug(hotplug_dev, dev, errp);
4365     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4366         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4367     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4368         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4369     }
4370 }
4371 
4372 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4373                                                  DeviceState *dev)
4374 {
4375     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4376         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4377         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4378         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4379         return HOTPLUG_HANDLER(machine);
4380     }
4381     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4382         PCIDevice *pcidev = PCI_DEVICE(dev);
4383         PCIBus *root = pci_device_root_bus(pcidev);
4384         SpaprPhbState *phb =
4385             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4386                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4387 
4388         if (phb) {
4389             return HOTPLUG_HANDLER(phb);
4390         }
4391     }
4392     return NULL;
4393 }
4394 
4395 static CpuInstanceProperties
4396 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4397 {
4398     CPUArchId *core_slot;
4399     MachineClass *mc = MACHINE_GET_CLASS(machine);
4400 
4401     /* make sure possible_cpu are initialized */
4402     mc->possible_cpu_arch_ids(machine);
4403     /* get CPU core slot containing thread that matches cpu_index */
4404     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4405     assert(core_slot);
4406     return core_slot->props;
4407 }
4408 
4409 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4410 {
4411     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4412 }
4413 
4414 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4415 {
4416     int i;
4417     unsigned int smp_threads = machine->smp.threads;
4418     unsigned int smp_cpus = machine->smp.cpus;
4419     const char *core_type;
4420     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4421     MachineClass *mc = MACHINE_GET_CLASS(machine);
4422 
4423     if (!mc->has_hotpluggable_cpus) {
4424         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4425     }
4426     if (machine->possible_cpus) {
4427         assert(machine->possible_cpus->len == spapr_max_cores);
4428         return machine->possible_cpus;
4429     }
4430 
4431     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4432     if (!core_type) {
4433         error_report("Unable to find sPAPR CPU Core definition");
4434         exit(1);
4435     }
4436 
4437     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4438                              sizeof(CPUArchId) * spapr_max_cores);
4439     machine->possible_cpus->len = spapr_max_cores;
4440     for (i = 0; i < machine->possible_cpus->len; i++) {
4441         int core_id = i * smp_threads;
4442 
4443         machine->possible_cpus->cpus[i].type = core_type;
4444         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4445         machine->possible_cpus->cpus[i].arch_id = core_id;
4446         machine->possible_cpus->cpus[i].props.has_core_id = true;
4447         machine->possible_cpus->cpus[i].props.core_id = core_id;
4448     }
4449     return machine->possible_cpus;
4450 }
4451 
4452 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4453                                 uint64_t *buid, hwaddr *pio,
4454                                 hwaddr *mmio32, hwaddr *mmio64,
4455                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4456 {
4457     /*
4458      * New-style PHB window placement.
4459      *
4460      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4461      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4462      * windows.
4463      *
4464      * Some guest kernels can't work with MMIO windows above 1<<46
4465      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4466      *
4467      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4468      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4469      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4470      * 1TiB 64-bit MMIO windows for each PHB.
4471      */
4472     const uint64_t base_buid = 0x800000020000000ULL;
4473     int i;
4474 
4475     /* Sanity check natural alignments */
4476     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4477     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4478     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4479     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4480     /* Sanity check bounds */
4481     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4482                       SPAPR_PCI_MEM32_WIN_SIZE);
4483     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4484                       SPAPR_PCI_MEM64_WIN_SIZE);
4485 
4486     if (index >= SPAPR_MAX_PHBS) {
4487         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4488                    SPAPR_MAX_PHBS - 1);
4489         return false;
4490     }
4491 
4492     *buid = base_buid + index;
4493     for (i = 0; i < n_dma; ++i) {
4494         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4495     }
4496 
4497     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4498     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4499     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4500     return true;
4501 }
4502 
4503 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4504 {
4505     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4506 
4507     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4508 }
4509 
4510 static void spapr_ics_resend(XICSFabric *dev)
4511 {
4512     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4513 
4514     ics_resend(spapr->ics);
4515 }
4516 
4517 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4518 {
4519     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4520 
4521     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4522 }
4523 
4524 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4525                                  Monitor *mon)
4526 {
4527     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4528 
4529     spapr_irq_print_info(spapr, mon);
4530     monitor_printf(mon, "irqchip: %s\n",
4531                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4532 }
4533 
4534 /*
4535  * This is a XIVE only operation
4536  */
4537 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4538                            uint8_t nvt_blk, uint32_t nvt_idx,
4539                            bool cam_ignore, uint8_t priority,
4540                            uint32_t logic_serv, XiveTCTXMatch *match)
4541 {
4542     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4543     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4544     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4545     int count;
4546 
4547     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4548                            priority, logic_serv, match);
4549     if (count < 0) {
4550         return count;
4551     }
4552 
4553     /*
4554      * When we implement the save and restore of the thread interrupt
4555      * contexts in the enter/exit CPU handlers of the machine and the
4556      * escalations in QEMU, we should be able to handle non dispatched
4557      * vCPUs.
4558      *
4559      * Until this is done, the sPAPR machine should find at least one
4560      * matching context always.
4561      */
4562     if (count == 0) {
4563         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4564                       nvt_blk, nvt_idx);
4565     }
4566 
4567     return count;
4568 }
4569 
4570 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4571 {
4572     return cpu->vcpu_id;
4573 }
4574 
4575 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4576 {
4577     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4578     MachineState *ms = MACHINE(spapr);
4579     int vcpu_id;
4580 
4581     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4582 
4583     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4584         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4585         error_append_hint(errp, "Adjust the number of cpus to %d "
4586                           "or try to raise the number of threads per core\n",
4587                           vcpu_id * ms->smp.threads / spapr->vsmt);
4588         return false;
4589     }
4590 
4591     cpu->vcpu_id = vcpu_id;
4592     return true;
4593 }
4594 
4595 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4596 {
4597     CPUState *cs;
4598 
4599     CPU_FOREACH(cs) {
4600         PowerPCCPU *cpu = POWERPC_CPU(cs);
4601 
4602         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4603             return cpu;
4604         }
4605     }
4606 
4607     return NULL;
4608 }
4609 
4610 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4611 {
4612     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4613 
4614     return spapr_cpu->in_nested;
4615 }
4616 
4617 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4618 {
4619     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4620 
4621     /* These are only called by TCG, KVM maintains dispatch state */
4622 
4623     spapr_cpu->prod = false;
4624     if (spapr_cpu->vpa_addr) {
4625         CPUState *cs = CPU(cpu);
4626         uint32_t dispatch;
4627 
4628         dispatch = ldl_be_phys(cs->as,
4629                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4630         dispatch++;
4631         if ((dispatch & 1) != 0) {
4632             qemu_log_mask(LOG_GUEST_ERROR,
4633                           "VPA: incorrect dispatch counter value for "
4634                           "dispatched partition %u, correcting.\n", dispatch);
4635             dispatch++;
4636         }
4637         stl_be_phys(cs->as,
4638                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4639     }
4640 }
4641 
4642 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4643 {
4644     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4645 
4646     if (spapr_cpu->vpa_addr) {
4647         CPUState *cs = CPU(cpu);
4648         uint32_t dispatch;
4649 
4650         dispatch = ldl_be_phys(cs->as,
4651                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4652         dispatch++;
4653         if ((dispatch & 1) != 1) {
4654             qemu_log_mask(LOG_GUEST_ERROR,
4655                           "VPA: incorrect dispatch counter value for "
4656                           "preempted partition %u, correcting.\n", dispatch);
4657             dispatch++;
4658         }
4659         stl_be_phys(cs->as,
4660                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4661     }
4662 }
4663 
4664 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4665 {
4666     MachineClass *mc = MACHINE_CLASS(oc);
4667     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4668     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4669     NMIClass *nc = NMI_CLASS(oc);
4670     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4671     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4672     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4673     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4674     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4675     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4676 
4677     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4678     mc->ignore_boot_device_suffixes = true;
4679 
4680     /*
4681      * We set up the default / latest behaviour here.  The class_init
4682      * functions for the specific versioned machine types can override
4683      * these details for backwards compatibility
4684      */
4685     mc->init = spapr_machine_init;
4686     mc->reset = spapr_machine_reset;
4687     mc->block_default_type = IF_SCSI;
4688 
4689     /*
4690      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4691      * In TCG the limit is restricted by the range of CPU IPIs available.
4692      */
4693     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4694 
4695     mc->no_parallel = 1;
4696     mc->default_boot_order = "";
4697     mc->default_ram_size = 512 * MiB;
4698     mc->default_ram_id = "ppc_spapr.ram";
4699     mc->default_display = "std";
4700     mc->kvm_type = spapr_kvm_type;
4701     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4702     mc->pci_allow_0_address = true;
4703     assert(!mc->get_hotplug_handler);
4704     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4705     hc->pre_plug = spapr_machine_device_pre_plug;
4706     hc->plug = spapr_machine_device_plug;
4707     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4708     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4709     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4710     hc->unplug_request = spapr_machine_device_unplug_request;
4711     hc->unplug = spapr_machine_device_unplug;
4712 
4713     smc->dr_lmb_enabled = true;
4714     smc->update_dt_enabled = true;
4715     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4716     mc->has_hotpluggable_cpus = true;
4717     mc->nvdimm_supported = true;
4718     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4719     fwc->get_dev_path = spapr_get_fw_dev_path;
4720     nc->nmi_monitor_handler = spapr_nmi;
4721     smc->phb_placement = spapr_phb_placement;
4722     vhc->cpu_in_nested = spapr_cpu_in_nested;
4723     vhc->deliver_hv_excp = spapr_exit_nested;
4724     vhc->hypercall = emulate_spapr_hypercall;
4725     vhc->hpt_mask = spapr_hpt_mask;
4726     vhc->map_hptes = spapr_map_hptes;
4727     vhc->unmap_hptes = spapr_unmap_hptes;
4728     vhc->hpte_set_c = spapr_hpte_set_c;
4729     vhc->hpte_set_r = spapr_hpte_set_r;
4730     vhc->get_pate = spapr_get_pate;
4731     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4732     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4733     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4734     xic->ics_get = spapr_ics_get;
4735     xic->ics_resend = spapr_ics_resend;
4736     xic->icp_get = spapr_icp_get;
4737     ispc->print_info = spapr_pic_print_info;
4738     /* Force NUMA node memory size to be a multiple of
4739      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4740      * in which LMBs are represented and hot-added
4741      */
4742     mc->numa_mem_align_shift = 28;
4743     mc->auto_enable_numa = true;
4744 
4745     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4746     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4747     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4748     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4749     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4750     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4751     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4752     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4753     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4754     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4755     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4756     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4757 
4758     /*
4759      * This cap specifies whether the AIL 3 mode for
4760      * H_SET_RESOURCE is supported. The default is modified
4761      * by default_caps_with_cpu().
4762      */
4763     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4764     spapr_caps_add_properties(smc);
4765     smc->irq = &spapr_irq_dual;
4766     smc->dr_phb_enabled = true;
4767     smc->linux_pci_probe = true;
4768     smc->smp_threads_vsmt = true;
4769     smc->nr_xirqs = SPAPR_NR_XIRQS;
4770     xfc->match_nvt = spapr_match_nvt;
4771     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4772     vmc->quiesce = spapr_vof_quiesce;
4773     vmc->setprop = spapr_vof_setprop;
4774 }
4775 
4776 static const TypeInfo spapr_machine_info = {
4777     .name          = TYPE_SPAPR_MACHINE,
4778     .parent        = TYPE_MACHINE,
4779     .abstract      = true,
4780     .instance_size = sizeof(SpaprMachineState),
4781     .instance_init = spapr_instance_init,
4782     .instance_finalize = spapr_machine_finalizefn,
4783     .class_size    = sizeof(SpaprMachineClass),
4784     .class_init    = spapr_machine_class_init,
4785     .interfaces = (InterfaceInfo[]) {
4786         { TYPE_FW_PATH_PROVIDER },
4787         { TYPE_NMI },
4788         { TYPE_HOTPLUG_HANDLER },
4789         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4790         { TYPE_XICS_FABRIC },
4791         { TYPE_INTERRUPT_STATS_PROVIDER },
4792         { TYPE_XIVE_FABRIC },
4793         { TYPE_VOF_MACHINE_IF },
4794         { }
4795     },
4796 };
4797 
4798 static void spapr_machine_latest_class_options(MachineClass *mc)
4799 {
4800     mc->alias = "pseries";
4801     mc->is_default = true;
4802 }
4803 
4804 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4805     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4806                                                     void *data)      \
4807     {                                                                \
4808         MachineClass *mc = MACHINE_CLASS(oc);                        \
4809         spapr_machine_##suffix##_class_options(mc);                  \
4810         if (latest) {                                                \
4811             spapr_machine_latest_class_options(mc);                  \
4812         }                                                            \
4813     }                                                                \
4814     static const TypeInfo spapr_machine_##suffix##_info = {          \
4815         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4816         .parent = TYPE_SPAPR_MACHINE,                                \
4817         .class_init = spapr_machine_##suffix##_class_init,           \
4818     };                                                               \
4819     static void spapr_machine_register_##suffix(void)                \
4820     {                                                                \
4821         type_register(&spapr_machine_##suffix##_info);               \
4822     }                                                                \
4823     type_init(spapr_machine_register_##suffix)
4824 
4825 /*
4826  * pseries-9.0
4827  */
4828 static void spapr_machine_9_0_class_options(MachineClass *mc)
4829 {
4830     /* Defaults for the latest behaviour inherited from the base class */
4831 }
4832 
4833 DEFINE_SPAPR_MACHINE(9_0, "9.0", true);
4834 
4835 /*
4836  * pseries-8.2
4837  */
4838 static void spapr_machine_8_2_class_options(MachineClass *mc)
4839 {
4840     spapr_machine_9_0_class_options(mc);
4841     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4842 }
4843 
4844 DEFINE_SPAPR_MACHINE(8_2, "8.2", false);
4845 
4846 /*
4847  * pseries-8.1
4848  */
4849 static void spapr_machine_8_1_class_options(MachineClass *mc)
4850 {
4851     spapr_machine_8_2_class_options(mc);
4852     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4853 }
4854 
4855 DEFINE_SPAPR_MACHINE(8_1, "8.1", false);
4856 
4857 /*
4858  * pseries-8.0
4859  */
4860 static void spapr_machine_8_0_class_options(MachineClass *mc)
4861 {
4862     spapr_machine_8_1_class_options(mc);
4863     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4864 }
4865 
4866 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4867 
4868 /*
4869  * pseries-7.2
4870  */
4871 static void spapr_machine_7_2_class_options(MachineClass *mc)
4872 {
4873     spapr_machine_8_0_class_options(mc);
4874     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4875 }
4876 
4877 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4878 
4879 /*
4880  * pseries-7.1
4881  */
4882 static void spapr_machine_7_1_class_options(MachineClass *mc)
4883 {
4884     spapr_machine_7_2_class_options(mc);
4885     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4886 }
4887 
4888 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4889 
4890 /*
4891  * pseries-7.0
4892  */
4893 static void spapr_machine_7_0_class_options(MachineClass *mc)
4894 {
4895     spapr_machine_7_1_class_options(mc);
4896     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4897 }
4898 
4899 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4900 
4901 /*
4902  * pseries-6.2
4903  */
4904 static void spapr_machine_6_2_class_options(MachineClass *mc)
4905 {
4906     spapr_machine_7_0_class_options(mc);
4907     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4908 }
4909 
4910 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4911 
4912 /*
4913  * pseries-6.1
4914  */
4915 static void spapr_machine_6_1_class_options(MachineClass *mc)
4916 {
4917     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4918 
4919     spapr_machine_6_2_class_options(mc);
4920     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4921     smc->pre_6_2_numa_affinity = true;
4922     mc->smp_props.prefer_sockets = true;
4923 }
4924 
4925 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4926 
4927 /*
4928  * pseries-6.0
4929  */
4930 static void spapr_machine_6_0_class_options(MachineClass *mc)
4931 {
4932     spapr_machine_6_1_class_options(mc);
4933     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4934 }
4935 
4936 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4937 
4938 /*
4939  * pseries-5.2
4940  */
4941 static void spapr_machine_5_2_class_options(MachineClass *mc)
4942 {
4943     spapr_machine_6_0_class_options(mc);
4944     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4945 }
4946 
4947 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4948 
4949 /*
4950  * pseries-5.1
4951  */
4952 static void spapr_machine_5_1_class_options(MachineClass *mc)
4953 {
4954     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4955 
4956     spapr_machine_5_2_class_options(mc);
4957     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4958     smc->pre_5_2_numa_associativity = true;
4959 }
4960 
4961 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4962 
4963 /*
4964  * pseries-5.0
4965  */
4966 static void spapr_machine_5_0_class_options(MachineClass *mc)
4967 {
4968     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4969     static GlobalProperty compat[] = {
4970         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4971     };
4972 
4973     spapr_machine_5_1_class_options(mc);
4974     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4975     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4976     mc->numa_mem_supported = true;
4977     smc->pre_5_1_assoc_refpoints = true;
4978 }
4979 
4980 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4981 
4982 /*
4983  * pseries-4.2
4984  */
4985 static void spapr_machine_4_2_class_options(MachineClass *mc)
4986 {
4987     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4988 
4989     spapr_machine_5_0_class_options(mc);
4990     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4991     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4992     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4993     smc->rma_limit = 16 * GiB;
4994     mc->nvdimm_supported = false;
4995 }
4996 
4997 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4998 
4999 /*
5000  * pseries-4.1
5001  */
5002 static void spapr_machine_4_1_class_options(MachineClass *mc)
5003 {
5004     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5005     static GlobalProperty compat[] = {
5006         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
5007         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
5008     };
5009 
5010     spapr_machine_4_2_class_options(mc);
5011     smc->linux_pci_probe = false;
5012     smc->smp_threads_vsmt = false;
5013     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
5014     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5015 }
5016 
5017 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
5018 
5019 /*
5020  * pseries-4.0
5021  */
5022 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
5023                               uint64_t *buid, hwaddr *pio,
5024                               hwaddr *mmio32, hwaddr *mmio64,
5025                               unsigned n_dma, uint32_t *liobns, Error **errp)
5026 {
5027     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
5028                              liobns, errp)) {
5029         return false;
5030     }
5031     return true;
5032 }
5033 static void spapr_machine_4_0_class_options(MachineClass *mc)
5034 {
5035     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5036 
5037     spapr_machine_4_1_class_options(mc);
5038     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5039     smc->phb_placement = phb_placement_4_0;
5040     smc->irq = &spapr_irq_xics;
5041     smc->pre_4_1_migration = true;
5042 }
5043 
5044 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
5045 
5046 /*
5047  * pseries-3.1
5048  */
5049 static void spapr_machine_3_1_class_options(MachineClass *mc)
5050 {
5051     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5052 
5053     spapr_machine_4_0_class_options(mc);
5054     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5055 
5056     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5057     smc->update_dt_enabled = false;
5058     smc->dr_phb_enabled = false;
5059     smc->broken_host_serial_model = true;
5060     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5061     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5062     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5063     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5064 }
5065 
5066 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
5067 
5068 /*
5069  * pseries-3.0
5070  */
5071 
5072 static void spapr_machine_3_0_class_options(MachineClass *mc)
5073 {
5074     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5075 
5076     spapr_machine_3_1_class_options(mc);
5077     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5078 
5079     smc->legacy_irq_allocation = true;
5080     smc->nr_xirqs = 0x400;
5081     smc->irq = &spapr_irq_xics_legacy;
5082 }
5083 
5084 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
5085 
5086 /*
5087  * pseries-2.12
5088  */
5089 static void spapr_machine_2_12_class_options(MachineClass *mc)
5090 {
5091     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5092     static GlobalProperty compat[] = {
5093         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5094         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5095     };
5096 
5097     spapr_machine_3_0_class_options(mc);
5098     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5099     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5100 
5101     /* We depend on kvm_enabled() to choose a default value for the
5102      * hpt-max-page-size capability. Of course we can't do it here
5103      * because this is too early and the HW accelerator isn't initialized
5104      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5105      */
5106     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5107 }
5108 
5109 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5110 
5111 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5112 {
5113     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5114 
5115     spapr_machine_2_12_class_options(mc);
5116     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5117     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5118     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5119 }
5120 
5121 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5122 
5123 /*
5124  * pseries-2.11
5125  */
5126 
5127 static void spapr_machine_2_11_class_options(MachineClass *mc)
5128 {
5129     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5130 
5131     spapr_machine_2_12_class_options(mc);
5132     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5133     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5134     mc->deprecation_reason = "old and not maintained - use a 2.12+ version";
5135 }
5136 
5137 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5138 
5139 /*
5140  * pseries-2.10
5141  */
5142 
5143 static void spapr_machine_2_10_class_options(MachineClass *mc)
5144 {
5145     spapr_machine_2_11_class_options(mc);
5146     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5147 }
5148 
5149 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5150 
5151 /*
5152  * pseries-2.9
5153  */
5154 
5155 static void spapr_machine_2_9_class_options(MachineClass *mc)
5156 {
5157     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5158     static GlobalProperty compat[] = {
5159         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5160     };
5161 
5162     spapr_machine_2_10_class_options(mc);
5163     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5164     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5165     smc->pre_2_10_has_unused_icps = true;
5166     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5167 }
5168 
5169 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5170 
5171 /*
5172  * pseries-2.8
5173  */
5174 
5175 static void spapr_machine_2_8_class_options(MachineClass *mc)
5176 {
5177     static GlobalProperty compat[] = {
5178         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5179     };
5180 
5181     spapr_machine_2_9_class_options(mc);
5182     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5183     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5184     mc->numa_mem_align_shift = 23;
5185 }
5186 
5187 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5188 
5189 /*
5190  * pseries-2.7
5191  */
5192 
5193 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5194                               uint64_t *buid, hwaddr *pio,
5195                               hwaddr *mmio32, hwaddr *mmio64,
5196                               unsigned n_dma, uint32_t *liobns, Error **errp)
5197 {
5198     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5199     const uint64_t base_buid = 0x800000020000000ULL;
5200     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5201     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5202     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5203     const uint32_t max_index = 255;
5204     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5205 
5206     uint64_t ram_top = MACHINE(spapr)->ram_size;
5207     hwaddr phb0_base, phb_base;
5208     int i;
5209 
5210     /* Do we have device memory? */
5211     if (MACHINE(spapr)->device_memory) {
5212         /* Can't just use maxram_size, because there may be an
5213          * alignment gap between normal and device memory regions
5214          */
5215         ram_top = MACHINE(spapr)->device_memory->base +
5216             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5217     }
5218 
5219     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5220 
5221     if (index > max_index) {
5222         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5223                    max_index);
5224         return false;
5225     }
5226 
5227     *buid = base_buid + index;
5228     for (i = 0; i < n_dma; ++i) {
5229         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5230     }
5231 
5232     phb_base = phb0_base + index * phb_spacing;
5233     *pio = phb_base + pio_offset;
5234     *mmio32 = phb_base + mmio_offset;
5235     /*
5236      * We don't set the 64-bit MMIO window, relying on the PHB's
5237      * fallback behaviour of automatically splitting a large "32-bit"
5238      * window into contiguous 32-bit and 64-bit windows
5239      */
5240 
5241     return true;
5242 }
5243 
5244 static void spapr_machine_2_7_class_options(MachineClass *mc)
5245 {
5246     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5247     static GlobalProperty compat[] = {
5248         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5249         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5250         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5251         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5252     };
5253 
5254     spapr_machine_2_8_class_options(mc);
5255     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5256     mc->default_machine_opts = "modern-hotplug-events=off";
5257     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5258     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5259     smc->phb_placement = phb_placement_2_7;
5260 }
5261 
5262 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5263 
5264 /*
5265  * pseries-2.6
5266  */
5267 
5268 static void spapr_machine_2_6_class_options(MachineClass *mc)
5269 {
5270     static GlobalProperty compat[] = {
5271         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5272     };
5273 
5274     spapr_machine_2_7_class_options(mc);
5275     mc->has_hotpluggable_cpus = false;
5276     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5277     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5278 }
5279 
5280 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5281 
5282 /*
5283  * pseries-2.5
5284  */
5285 
5286 static void spapr_machine_2_5_class_options(MachineClass *mc)
5287 {
5288     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5289     static GlobalProperty compat[] = {
5290         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5291     };
5292 
5293     spapr_machine_2_6_class_options(mc);
5294     smc->use_ohci_by_default = true;
5295     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5296     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5297 }
5298 
5299 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5300 
5301 /*
5302  * pseries-2.4
5303  */
5304 
5305 static void spapr_machine_2_4_class_options(MachineClass *mc)
5306 {
5307     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5308 
5309     spapr_machine_2_5_class_options(mc);
5310     smc->dr_lmb_enabled = false;
5311     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5312 }
5313 
5314 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5315 
5316 /*
5317  * pseries-2.3
5318  */
5319 
5320 static void spapr_machine_2_3_class_options(MachineClass *mc)
5321 {
5322     static GlobalProperty compat[] = {
5323         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5324     };
5325     spapr_machine_2_4_class_options(mc);
5326     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5327     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5328 }
5329 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5330 
5331 /*
5332  * pseries-2.2
5333  */
5334 
5335 static void spapr_machine_2_2_class_options(MachineClass *mc)
5336 {
5337     static GlobalProperty compat[] = {
5338         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5339     };
5340 
5341     spapr_machine_2_3_class_options(mc);
5342     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5343     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5344     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5345 }
5346 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5347 
5348 /*
5349  * pseries-2.1
5350  */
5351 
5352 static void spapr_machine_2_1_class_options(MachineClass *mc)
5353 {
5354     spapr_machine_2_2_class_options(mc);
5355     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5356 }
5357 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5358 
5359 static void spapr_machine_register_types(void)
5360 {
5361     type_register_static(&spapr_machine_info);
5362 }
5363 
5364 type_init(spapr_machine_register_types)
5365