xref: /qemu/hw/ppc/spapr.c (revision d83d350cb2f9edbdb4eccc6454c82202a09483c8)
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     uint8_t pa_features_206[] = { 6, 0,
247         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
248     uint8_t pa_features_207[] = { 24, 0,
249         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
250         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
251         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
252         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
253     uint8_t pa_features_300[] = { 66, 0,
254         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
255         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
256         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
257         /* 6: DS207 */
258         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
259         /* 16: Vector */
260         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
261         /* 18: Vec. Scalar, 20: Vec. XOR */
262         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
263         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
264         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
265         /* 32: LE atomic, 34: EBB + ext EBB */
266         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
267         /* 40: Radix MMU */
268         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
269         /* 42: PM, 44: PC RA, 46: SC vec'd */
270         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
271         /* 48: SIMD, 50: QP BFP, 52: String */
272         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
273         /* 54: DecFP, 56: DecI, 58: SHA */
274         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
275         /* 60: NM atomic, 62: RNG */
276         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
277     };
278     /* 3.1 removes SAO, HTM support */
279     uint8_t pa_features_31[] = { 74, 0,
280         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
281         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
282         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
283         /* 6: DS207 */
284         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
285         /* 16: Vector */
286         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
287         /* 18: Vec. Scalar, 20: Vec. XOR */
288         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
289         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
290         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
291         /* 32: LE atomic, 34: EBB + ext EBB */
292         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
293         /* 40: Radix MMU */
294         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
295         /* 42: PM, 44: PC RA, 46: SC vec'd */
296         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
297         /* 48: SIMD, 50: QP BFP, 52: String */
298         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
299         /* 54: DecFP, 56: DecI, 58: SHA */
300         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
301         /* 60: NM atomic, 62: RNG */
302         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
303         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
304         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
305         /* 72: [P]HASHST/[P]HASHCHK */
306         0x80, 0x00,                         /* 72 - 73 */
307     };
308     uint8_t *pa_features = NULL;
309     size_t pa_size;
310 
311     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
312         pa_features = pa_features_206;
313         pa_size = sizeof(pa_features_206);
314     }
315     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
316         pa_features = pa_features_207;
317         pa_size = sizeof(pa_features_207);
318     }
319     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
320         pa_features = pa_features_300;
321         pa_size = sizeof(pa_features_300);
322     }
323     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
324         pa_features = pa_features_31;
325         pa_size = sizeof(pa_features_31);
326     }
327     if (!pa_features) {
328         return;
329     }
330 
331     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
332         /*
333          * Note: we keep CI large pages off by default because a 64K capable
334          * guest provisioned with large pages might otherwise try to map a qemu
335          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
336          * even if that qemu runs on a 4k host.
337          * We dd this bit back here if we are confident this is not an issue
338          */
339         pa_features[3] |= 0x20;
340     }
341     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
342         pa_features[24] |= 0x80;    /* Transactional memory support */
343     }
344     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
345         /* Workaround for broken kernels that attempt (guest) radix
346          * mode when they can't handle it, if they see the radix bit set
347          * in pa-features. So hide it from them. */
348         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
349     }
350 
351     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
352 }
353 
354 static hwaddr spapr_node0_size(MachineState *machine)
355 {
356     if (machine->numa_state->num_nodes) {
357         int i;
358         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
359             if (machine->numa_state->nodes[i].node_mem) {
360                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
361                            machine->ram_size);
362             }
363         }
364     }
365     return machine->ram_size;
366 }
367 
368 static void add_str(GString *s, const gchar *s1)
369 {
370     g_string_append_len(s, s1, strlen(s1) + 1);
371 }
372 
373 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
374                                 hwaddr start, hwaddr size)
375 {
376     char mem_name[32];
377     uint64_t mem_reg_property[2];
378     int off;
379 
380     mem_reg_property[0] = cpu_to_be64(start);
381     mem_reg_property[1] = cpu_to_be64(size);
382 
383     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
384     off = fdt_add_subnode(fdt, 0, mem_name);
385     _FDT(off);
386     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
387     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
388                       sizeof(mem_reg_property))));
389     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
390     return off;
391 }
392 
393 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
394 {
395     MemoryDeviceInfoList *info;
396 
397     for (info = list; info; info = info->next) {
398         MemoryDeviceInfo *value = info->value;
399 
400         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
401             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
402 
403             if (addr >= pcdimm_info->addr &&
404                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
405                 return pcdimm_info->node;
406             }
407         }
408     }
409 
410     return -1;
411 }
412 
413 struct sPAPRDrconfCellV2 {
414      uint32_t seq_lmbs;
415      uint64_t base_addr;
416      uint32_t drc_index;
417      uint32_t aa_index;
418      uint32_t flags;
419 } QEMU_PACKED;
420 
421 typedef struct DrconfCellQueue {
422     struct sPAPRDrconfCellV2 cell;
423     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
424 } DrconfCellQueue;
425 
426 static DrconfCellQueue *
427 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
428                       uint32_t drc_index, uint32_t aa_index,
429                       uint32_t flags)
430 {
431     DrconfCellQueue *elem;
432 
433     elem = g_malloc0(sizeof(*elem));
434     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
435     elem->cell.base_addr = cpu_to_be64(base_addr);
436     elem->cell.drc_index = cpu_to_be32(drc_index);
437     elem->cell.aa_index = cpu_to_be32(aa_index);
438     elem->cell.flags = cpu_to_be32(flags);
439 
440     return elem;
441 }
442 
443 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
444                                       int offset, MemoryDeviceInfoList *dimms)
445 {
446     MachineState *machine = MACHINE(spapr);
447     uint8_t *int_buf, *cur_index;
448     int ret;
449     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
450     uint64_t addr, cur_addr, size;
451     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
452     uint64_t mem_end = machine->device_memory->base +
453                        memory_region_size(&machine->device_memory->mr);
454     uint32_t node, buf_len, nr_entries = 0;
455     SpaprDrc *drc;
456     DrconfCellQueue *elem, *next;
457     MemoryDeviceInfoList *info;
458     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
459         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
460 
461     /* Entry to cover RAM and the gap area */
462     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
463                                  SPAPR_LMB_FLAGS_RESERVED |
464                                  SPAPR_LMB_FLAGS_DRC_INVALID);
465     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
466     nr_entries++;
467 
468     cur_addr = machine->device_memory->base;
469     for (info = dimms; info; info = info->next) {
470         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
471 
472         addr = di->addr;
473         size = di->size;
474         node = di->node;
475 
476         /*
477          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
478          * area is marked hotpluggable in the next iteration for the bigger
479          * chunk including the NVDIMM occupied area.
480          */
481         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
482             continue;
483 
484         /* Entry for hot-pluggable area */
485         if (cur_addr < addr) {
486             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
487             g_assert(drc);
488             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
489                                          cur_addr, spapr_drc_index(drc), -1, 0);
490             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
491             nr_entries++;
492         }
493 
494         /* Entry for DIMM */
495         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
496         g_assert(drc);
497         elem = spapr_get_drconf_cell(size / lmb_size, addr,
498                                      spapr_drc_index(drc), node,
499                                      (SPAPR_LMB_FLAGS_ASSIGNED |
500                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
501         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
502         nr_entries++;
503         cur_addr = addr + size;
504     }
505 
506     /* Entry for remaining hotpluggable area */
507     if (cur_addr < mem_end) {
508         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
509         g_assert(drc);
510         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
511                                      cur_addr, spapr_drc_index(drc), -1, 0);
512         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
513         nr_entries++;
514     }
515 
516     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
517     int_buf = cur_index = g_malloc0(buf_len);
518     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
519     cur_index += sizeof(nr_entries);
520 
521     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
522         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
523         cur_index += sizeof(elem->cell);
524         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
525         g_free(elem);
526     }
527 
528     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
529     g_free(int_buf);
530     if (ret < 0) {
531         return -1;
532     }
533     return 0;
534 }
535 
536 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
537                                    int offset, MemoryDeviceInfoList *dimms)
538 {
539     MachineState *machine = MACHINE(spapr);
540     int i, ret;
541     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
542     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
543     uint32_t nr_lmbs = (machine->device_memory->base +
544                        memory_region_size(&machine->device_memory->mr)) /
545                        lmb_size;
546     uint32_t *int_buf, *cur_index, buf_len;
547 
548     /*
549      * Allocate enough buffer size to fit in ibm,dynamic-memory
550      */
551     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
552     cur_index = int_buf = g_malloc0(buf_len);
553     int_buf[0] = cpu_to_be32(nr_lmbs);
554     cur_index++;
555     for (i = 0; i < nr_lmbs; i++) {
556         uint64_t addr = i * lmb_size;
557         uint32_t *dynamic_memory = cur_index;
558 
559         if (i >= device_lmb_start) {
560             SpaprDrc *drc;
561 
562             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
563             g_assert(drc);
564 
565             dynamic_memory[0] = cpu_to_be32(addr >> 32);
566             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
567             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
568             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
569             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
570             if (memory_region_present(get_system_memory(), addr)) {
571                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
572             } else {
573                 dynamic_memory[5] = cpu_to_be32(0);
574             }
575         } else {
576             /*
577              * LMB information for RMA, boot time RAM and gap b/n RAM and
578              * device memory region -- all these are marked as reserved
579              * and as having no valid DRC.
580              */
581             dynamic_memory[0] = cpu_to_be32(addr >> 32);
582             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
583             dynamic_memory[2] = cpu_to_be32(0);
584             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
585             dynamic_memory[4] = cpu_to_be32(-1);
586             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
587                                             SPAPR_LMB_FLAGS_DRC_INVALID);
588         }
589 
590         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
591     }
592     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
593     g_free(int_buf);
594     if (ret < 0) {
595         return -1;
596     }
597     return 0;
598 }
599 
600 /*
601  * Adds ibm,dynamic-reconfiguration-memory node.
602  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
603  * of this device tree node.
604  */
605 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
606                                                    void *fdt)
607 {
608     MachineState *machine = MACHINE(spapr);
609     int ret, offset;
610     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
611     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
612                                 cpu_to_be32(lmb_size & 0xffffffff)};
613     MemoryDeviceInfoList *dimms = NULL;
614 
615     /* Don't create the node if there is no device memory. */
616     if (!machine->device_memory) {
617         return 0;
618     }
619 
620     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
621 
622     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
623                     sizeof(prop_lmb_size));
624     if (ret < 0) {
625         return ret;
626     }
627 
628     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
629     if (ret < 0) {
630         return ret;
631     }
632 
633     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
634     if (ret < 0) {
635         return ret;
636     }
637 
638     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
639     dimms = qmp_memory_device_list();
640     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
641         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
642     } else {
643         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
644     }
645     qapi_free_MemoryDeviceInfoList(dimms);
646 
647     if (ret < 0) {
648         return ret;
649     }
650 
651     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
652 
653     return ret;
654 }
655 
656 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
657 {
658     MachineState *machine = MACHINE(spapr);
659     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
660     hwaddr mem_start, node_size;
661     int i, nb_nodes = machine->numa_state->num_nodes;
662     NodeInfo *nodes = machine->numa_state->nodes;
663 
664     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
665         if (!nodes[i].node_mem) {
666             continue;
667         }
668         if (mem_start >= machine->ram_size) {
669             node_size = 0;
670         } else {
671             node_size = nodes[i].node_mem;
672             if (node_size > machine->ram_size - mem_start) {
673                 node_size = machine->ram_size - mem_start;
674             }
675         }
676         if (!mem_start) {
677             /* spapr_machine_init() checks for rma_size <= node0_size
678              * already */
679             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
680             mem_start += spapr->rma_size;
681             node_size -= spapr->rma_size;
682         }
683         for ( ; node_size; ) {
684             hwaddr sizetmp = pow2floor(node_size);
685 
686             /* mem_start != 0 here */
687             if (ctzl(mem_start) < ctzl(sizetmp)) {
688                 sizetmp = 1ULL << ctzl(mem_start);
689             }
690 
691             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
692             node_size -= sizetmp;
693             mem_start += sizetmp;
694         }
695     }
696 
697     /* Generate ibm,dynamic-reconfiguration-memory node if required */
698     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
699         int ret;
700 
701         g_assert(smc->dr_lmb_enabled);
702         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
703         if (ret) {
704             return ret;
705         }
706     }
707 
708     return 0;
709 }
710 
711 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
712                          SpaprMachineState *spapr)
713 {
714     MachineState *ms = MACHINE(spapr);
715     PowerPCCPU *cpu = POWERPC_CPU(cs);
716     CPUPPCState *env = &cpu->env;
717     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
718     int index = spapr_get_vcpu_id(cpu);
719     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
720                        0xffffffff, 0xffffffff};
721     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
722         : SPAPR_TIMEBASE_FREQ;
723     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
724     uint32_t page_sizes_prop[64];
725     size_t page_sizes_prop_size;
726     unsigned int smp_threads = ms->smp.threads;
727     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
728     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
729     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
730     SpaprDrc *drc;
731     int drc_index;
732     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
733     int i;
734 
735     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
736     if (drc) {
737         drc_index = spapr_drc_index(drc);
738         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
739     }
740 
741     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
742     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
743 
744     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
745     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
746                            env->dcache_line_size)));
747     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
748                            env->dcache_line_size)));
749     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
750                            env->icache_line_size)));
751     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
752                            env->icache_line_size)));
753 
754     if (pcc->l1_dcache_size) {
755         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
756                                pcc->l1_dcache_size)));
757     } else {
758         warn_report("Unknown L1 dcache size for cpu");
759     }
760     if (pcc->l1_icache_size) {
761         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
762                                pcc->l1_icache_size)));
763     } else {
764         warn_report("Unknown L1 icache size for cpu");
765     }
766 
767     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
768     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
769     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
770     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
771     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
772     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
773 
774     if (ppc_has_spr(cpu, SPR_PURR)) {
775         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
776     }
777     if (ppc_has_spr(cpu, SPR_PURR)) {
778         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
779     }
780 
781     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
782         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
783                           segs, sizeof(segs))));
784     }
785 
786     /* Advertise VSX (vector extensions) if available
787      *   1               == VMX / Altivec available
788      *   2               == VSX available
789      *
790      * Only CPUs for which we create core types in spapr_cpu_core.c
791      * are possible, and all of those have VMX */
792     if (env->insns_flags & PPC_ALTIVEC) {
793         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
794             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
795         } else {
796             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
797         }
798     }
799 
800     /* Advertise DFP (Decimal Floating Point) if available
801      *   0 / no property == no DFP
802      *   1               == DFP available */
803     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
804         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
805     }
806 
807     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
808                                                       sizeof(page_sizes_prop));
809     if (page_sizes_prop_size) {
810         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
811                           page_sizes_prop, page_sizes_prop_size)));
812     }
813 
814     spapr_dt_pa_features(spapr, cpu, fdt, offset);
815 
816     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
817                            cs->cpu_index / vcpus_per_socket)));
818 
819     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
820                       pft_size_prop, sizeof(pft_size_prop))));
821 
822     if (ms->numa_state->num_nodes > 1) {
823         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
824     }
825 
826     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
827 
828     if (pcc->radix_page_info) {
829         for (i = 0; i < pcc->radix_page_info->count; i++) {
830             radix_AP_encodings[i] =
831                 cpu_to_be32(pcc->radix_page_info->entries[i]);
832         }
833         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
834                           radix_AP_encodings,
835                           pcc->radix_page_info->count *
836                           sizeof(radix_AP_encodings[0]))));
837     }
838 
839     /*
840      * We set this property to let the guest know that it can use the large
841      * decrementer and its width in bits.
842      */
843     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
844         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
845                               pcc->lrg_decr_bits)));
846 }
847 
848 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
849                              int cpus_offset)
850 {
851     PowerPCCPU *cpu = POWERPC_CPU(cs);
852     int index = spapr_get_vcpu_id(cpu);
853     DeviceClass *dc = DEVICE_GET_CLASS(cs);
854     g_autofree char *nodename = NULL;
855     int offset;
856 
857     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
858         return;
859     }
860 
861     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
862     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
863     _FDT(offset);
864     spapr_dt_cpu(cs, fdt, offset, spapr);
865 }
866 
867 
868 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
869 {
870     CPUState **rev;
871     CPUState *cs;
872     int n_cpus;
873     int cpus_offset;
874     int i;
875 
876     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
877     _FDT(cpus_offset);
878     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
879     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
880 
881     /*
882      * We walk the CPUs in reverse order to ensure that CPU DT nodes
883      * created by fdt_add_subnode() end up in the right order in FDT
884      * for the guest kernel the enumerate the CPUs correctly.
885      *
886      * The CPU list cannot be traversed in reverse order, so we need
887      * to do extra work.
888      */
889     n_cpus = 0;
890     rev = NULL;
891     CPU_FOREACH(cs) {
892         rev = g_renew(CPUState *, rev, n_cpus + 1);
893         rev[n_cpus++] = cs;
894     }
895 
896     for (i = n_cpus - 1; i >= 0; i--) {
897         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
898     }
899 
900     g_free(rev);
901 }
902 
903 static int spapr_dt_rng(void *fdt)
904 {
905     int node;
906     int ret;
907 
908     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
909     if (node <= 0) {
910         return -1;
911     }
912     ret = fdt_setprop_string(fdt, node, "device_type",
913                              "ibm,platform-facilities");
914     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
915     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
916 
917     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
918     if (node <= 0) {
919         return -1;
920     }
921     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
922 
923     return ret ? -1 : 0;
924 }
925 
926 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
927 {
928     MachineState *ms = MACHINE(spapr);
929     int rtas;
930     GString *hypertas = g_string_sized_new(256);
931     GString *qemu_hypertas = g_string_sized_new(256);
932     uint32_t lrdr_capacity[] = {
933         0,
934         0,
935         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
936         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
937         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
938     };
939 
940     /* Do we have device memory? */
941     if (MACHINE(spapr)->device_memory) {
942         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
943             memory_region_size(&MACHINE(spapr)->device_memory->mr);
944 
945         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
946         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
947     }
948 
949     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
950 
951     /* hypertas */
952     add_str(hypertas, "hcall-pft");
953     add_str(hypertas, "hcall-term");
954     add_str(hypertas, "hcall-dabr");
955     add_str(hypertas, "hcall-interrupt");
956     add_str(hypertas, "hcall-tce");
957     add_str(hypertas, "hcall-vio");
958     add_str(hypertas, "hcall-splpar");
959     add_str(hypertas, "hcall-join");
960     add_str(hypertas, "hcall-bulk");
961     add_str(hypertas, "hcall-set-mode");
962     add_str(hypertas, "hcall-sprg0");
963     add_str(hypertas, "hcall-copy");
964     add_str(hypertas, "hcall-debug");
965     add_str(hypertas, "hcall-vphn");
966     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
967         add_str(hypertas, "hcall-rpt-invalidate");
968     }
969 
970     add_str(qemu_hypertas, "hcall-memop1");
971 
972     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
973         add_str(hypertas, "hcall-multi-tce");
974     }
975 
976     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
977         add_str(hypertas, "hcall-hpt-resize");
978     }
979 
980     add_str(hypertas, "hcall-watchdog");
981 
982     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
983                      hypertas->str, hypertas->len));
984     g_string_free(hypertas, TRUE);
985     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
986                      qemu_hypertas->str, qemu_hypertas->len));
987     g_string_free(qemu_hypertas, TRUE);
988 
989     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
990 
991     /*
992      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
993      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
994      *
995      * The system reset requirements are driven by existing Linux and PowerVM
996      * implementation which (contrary to PAPR) saves r3 in the error log
997      * structure like machine check, so Linux expects to find the saved r3
998      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
999      * does not look at the error value).
1000      *
1001      * System reset interrupts are not subject to interlock like machine
1002      * check, so this memory area could be corrupted if the sreset is
1003      * interrupted by a machine check (or vice versa) if it was shared. To
1004      * prevent this, system reset uses per-CPU areas for the sreset save
1005      * area. A system reset that interrupts a system reset handler could
1006      * still overwrite this area, but Linux doesn't try to recover in that
1007      * case anyway.
1008      *
1009      * The extra 8 bytes is required because Linux's FWNMI error log check
1010      * is off-by-one.
1011      *
1012      * RTAS_MIN_SIZE is required for the RTAS blob itself.
1013      */
1014     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
1015                           RTAS_ERROR_LOG_MAX +
1016                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
1017                           sizeof(uint64_t)));
1018     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1019                           RTAS_ERROR_LOG_MAX));
1020     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1021                           RTAS_EVENT_SCAN_RATE));
1022 
1023     g_assert(msi_nonbroken);
1024     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1025 
1026     /*
1027      * According to PAPR, rtas ibm,os-term does not guarantee a return
1028      * back to the guest cpu.
1029      *
1030      * While an additional ibm,extended-os-term property indicates
1031      * that rtas call return will always occur. Set this property.
1032      */
1033     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1034 
1035     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1036                      lrdr_capacity, sizeof(lrdr_capacity)));
1037 
1038     spapr_dt_rtas_tokens(fdt, rtas);
1039 }
1040 
1041 /*
1042  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1043  * and the XIVE features that the guest may request and thus the valid
1044  * values for bytes 23..26 of option vector 5:
1045  */
1046 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1047                                           int chosen)
1048 {
1049     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1050 
1051     char val[2 * 4] = {
1052         23, 0x00, /* XICS / XIVE mode */
1053         24, 0x00, /* Hash/Radix, filled in below. */
1054         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1055         26, 0x40, /* Radix options: GTSE == yes. */
1056     };
1057 
1058     if (spapr->irq->xics && spapr->irq->xive) {
1059         val[1] = SPAPR_OV5_XIVE_BOTH;
1060     } else if (spapr->irq->xive) {
1061         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1062     } else {
1063         assert(spapr->irq->xics);
1064         val[1] = SPAPR_OV5_XIVE_LEGACY;
1065     }
1066 
1067     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1068                           first_ppc_cpu->compat_pvr)) {
1069         /*
1070          * If we're in a pre POWER9 compat mode then the guest should
1071          * do hash and use the legacy interrupt mode
1072          */
1073         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1074         val[3] = 0x00; /* Hash */
1075         spapr_check_mmu_mode(false);
1076     } else if (kvm_enabled()) {
1077         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1078             val[3] = 0x80; /* OV5_MMU_BOTH */
1079         } else if (kvmppc_has_cap_mmu_radix()) {
1080             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1081         } else {
1082             val[3] = 0x00; /* Hash */
1083         }
1084     } else {
1085         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1086         val[3] = 0xC0;
1087     }
1088     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1089                      val, sizeof(val)));
1090 }
1091 
1092 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1093 {
1094     MachineState *machine = MACHINE(spapr);
1095     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1096     int chosen;
1097 
1098     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1099 
1100     if (reset) {
1101         const char *boot_device = spapr->boot_device;
1102         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1103         size_t cb = 0;
1104         g_autofree char *bootlist = get_boot_devices_list(&cb);
1105 
1106         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1107             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1108                                     machine->kernel_cmdline));
1109         }
1110 
1111         if (spapr->initrd_size) {
1112             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1113                                   spapr->initrd_base));
1114             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1115                                   spapr->initrd_base + spapr->initrd_size));
1116         }
1117 
1118         if (spapr->kernel_size) {
1119             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1120                                   cpu_to_be64(spapr->kernel_size) };
1121 
1122             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1123                          &kprop, sizeof(kprop)));
1124             if (spapr->kernel_le) {
1125                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1126             }
1127         }
1128         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1129             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1130         }
1131         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1132         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1133         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1134 
1135         if (cb && bootlist) {
1136             int i;
1137 
1138             for (i = 0; i < cb; i++) {
1139                 if (bootlist[i] == '\n') {
1140                     bootlist[i] = ' ';
1141                 }
1142             }
1143             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1144         }
1145 
1146         if (boot_device && strlen(boot_device)) {
1147             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1148         }
1149 
1150         if (spapr->want_stdout_path && stdout_path) {
1151             /*
1152              * "linux,stdout-path" and "stdout" properties are
1153              * deprecated by linux kernel. New platforms should only
1154              * use the "stdout-path" property. Set the new property
1155              * and continue using older property to remain compatible
1156              * with the existing firmware.
1157              */
1158             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1159             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1160         }
1161 
1162         /*
1163          * We can deal with BAR reallocation just fine, advertise it
1164          * to the guest
1165          */
1166         if (smc->linux_pci_probe) {
1167             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1168         }
1169 
1170         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1171     }
1172 
1173     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1174 
1175     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1176 }
1177 
1178 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1179 {
1180     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1181      * KVM to work under pHyp with some guest co-operation */
1182     int hypervisor;
1183     uint8_t hypercall[16];
1184 
1185     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1186     /* indicate KVM hypercall interface */
1187     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1188     if (kvmppc_has_cap_fixup_hcalls()) {
1189         /*
1190          * Older KVM versions with older guest kernels were broken
1191          * with the magic page, don't allow the guest to map it.
1192          */
1193         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1194                                   sizeof(hypercall))) {
1195             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1196                              hypercall, sizeof(hypercall)));
1197         }
1198     }
1199 }
1200 
1201 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1202 {
1203     MachineState *machine = MACHINE(spapr);
1204     MachineClass *mc = MACHINE_GET_CLASS(machine);
1205     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1206     uint32_t root_drc_type_mask = 0;
1207     int ret;
1208     void *fdt;
1209     SpaprPhbState *phb;
1210     char *buf;
1211 
1212     fdt = g_malloc0(space);
1213     _FDT((fdt_create_empty_tree(fdt, space)));
1214 
1215     /* Root node */
1216     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1217     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1218     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1219 
1220     /* Guest UUID & Name*/
1221     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1222     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1223     if (qemu_uuid_set) {
1224         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1225     }
1226     g_free(buf);
1227 
1228     if (qemu_get_vm_name()) {
1229         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1230                                 qemu_get_vm_name()));
1231     }
1232 
1233     /* Host Model & Serial Number */
1234     if (spapr->host_model) {
1235         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1236     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1237         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1238         g_free(buf);
1239     }
1240 
1241     if (spapr->host_serial) {
1242         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1243     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1244         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1245         g_free(buf);
1246     }
1247 
1248     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1249     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1250 
1251     /* /interrupt controller */
1252     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1253 
1254     ret = spapr_dt_memory(spapr, fdt);
1255     if (ret < 0) {
1256         error_report("couldn't setup memory nodes in fdt");
1257         exit(1);
1258     }
1259 
1260     /* /vdevice */
1261     spapr_dt_vdevice(spapr->vio_bus, fdt);
1262 
1263     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1264         ret = spapr_dt_rng(fdt);
1265         if (ret < 0) {
1266             error_report("could not set up rng device in the fdt");
1267             exit(1);
1268         }
1269     }
1270 
1271     QLIST_FOREACH(phb, &spapr->phbs, list) {
1272         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1273         if (ret < 0) {
1274             error_report("couldn't setup PCI devices in fdt");
1275             exit(1);
1276         }
1277     }
1278 
1279     spapr_dt_cpus(fdt, spapr);
1280 
1281     /* ibm,drc-indexes and friends */
1282     if (smc->dr_lmb_enabled) {
1283         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1284     }
1285     if (smc->dr_phb_enabled) {
1286         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1287     }
1288     if (mc->nvdimm_supported) {
1289         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1290     }
1291     if (root_drc_type_mask) {
1292         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1293     }
1294 
1295     if (mc->has_hotpluggable_cpus) {
1296         int offset = fdt_path_offset(fdt, "/cpus");
1297         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1298         if (ret < 0) {
1299             error_report("Couldn't set up CPU DR device tree properties");
1300             exit(1);
1301         }
1302     }
1303 
1304     /* /event-sources */
1305     spapr_dt_events(spapr, fdt);
1306 
1307     /* /rtas */
1308     spapr_dt_rtas(spapr, fdt);
1309 
1310     /* /chosen */
1311     spapr_dt_chosen(spapr, fdt, reset);
1312 
1313     /* /hypervisor */
1314     if (kvm_enabled()) {
1315         spapr_dt_hypervisor(spapr, fdt);
1316     }
1317 
1318     /* Build memory reserve map */
1319     if (reset) {
1320         if (spapr->kernel_size) {
1321             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1322                                   spapr->kernel_size)));
1323         }
1324         if (spapr->initrd_size) {
1325             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1326                                   spapr->initrd_size)));
1327         }
1328     }
1329 
1330     /* NVDIMM devices */
1331     if (mc->nvdimm_supported) {
1332         spapr_dt_persistent_memory(spapr, fdt);
1333     }
1334 
1335     return fdt;
1336 }
1337 
1338 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1339 {
1340     SpaprMachineState *spapr = opaque;
1341 
1342     return (addr & 0x0fffffff) + spapr->kernel_addr;
1343 }
1344 
1345 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1346                                     PowerPCCPU *cpu)
1347 {
1348     CPUPPCState *env = &cpu->env;
1349 
1350     /* The TCG path should also be holding the BQL at this point */
1351     g_assert(bql_locked());
1352 
1353     g_assert(!vhyp_cpu_in_nested(cpu));
1354 
1355     if (FIELD_EX64(env->msr, MSR, PR)) {
1356         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1357         env->gpr[3] = H_PRIVILEGE;
1358     } else {
1359         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1360     }
1361 }
1362 
1363 struct LPCRSyncState {
1364     target_ulong value;
1365     target_ulong mask;
1366 };
1367 
1368 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1369 {
1370     struct LPCRSyncState *s = arg.host_ptr;
1371     PowerPCCPU *cpu = POWERPC_CPU(cs);
1372     CPUPPCState *env = &cpu->env;
1373     target_ulong lpcr;
1374 
1375     cpu_synchronize_state(cs);
1376     lpcr = env->spr[SPR_LPCR];
1377     lpcr &= ~s->mask;
1378     lpcr |= s->value;
1379     ppc_store_lpcr(cpu, lpcr);
1380 }
1381 
1382 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1383 {
1384     CPUState *cs;
1385     struct LPCRSyncState s = {
1386         .value = value,
1387         .mask = mask
1388     };
1389     CPU_FOREACH(cs) {
1390         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1391     }
1392 }
1393 
1394 /* May be used when the machine is not running */
1395 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1396 {
1397     CPUState *cs;
1398     CPU_FOREACH(cs) {
1399         PowerPCCPU *cpu = POWERPC_CPU(cs);
1400         CPUPPCState *env = &cpu->env;
1401         target_ulong lpcr;
1402 
1403         lpcr = env->spr[SPR_LPCR];
1404         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1405         ppc_store_lpcr(cpu, lpcr);
1406     }
1407 }
1408 
1409 
1410 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1411                            target_ulong lpid, ppc_v3_pate_t *entry)
1412 {
1413     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1414     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1415 
1416     if (!spapr_cpu->in_nested) {
1417         assert(lpid == 0);
1418 
1419         /* Copy PATE1:GR into PATE0:HR */
1420         entry->dw0 = spapr->patb_entry & PATE0_HR;
1421         entry->dw1 = spapr->patb_entry;
1422 
1423     } else {
1424         uint64_t patb, pats;
1425 
1426         assert(lpid != 0);
1427 
1428         patb = spapr->nested_ptcr & PTCR_PATB;
1429         pats = spapr->nested_ptcr & PTCR_PATS;
1430 
1431         /* Check if partition table is properly aligned */
1432         if (patb & MAKE_64BIT_MASK(0, pats + 12)) {
1433             return false;
1434         }
1435 
1436         /* Calculate number of entries */
1437         pats = 1ull << (pats + 12 - 4);
1438         if (pats <= lpid) {
1439             return false;
1440         }
1441 
1442         /* Grab entry */
1443         patb += 16 * lpid;
1444         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1445         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1446     }
1447 
1448     return true;
1449 }
1450 
1451 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1452 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1453 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1454 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1455 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1456 
1457 /*
1458  * Get the fd to access the kernel htab, re-opening it if necessary
1459  */
1460 static int get_htab_fd(SpaprMachineState *spapr)
1461 {
1462     Error *local_err = NULL;
1463 
1464     if (spapr->htab_fd >= 0) {
1465         return spapr->htab_fd;
1466     }
1467 
1468     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1469     if (spapr->htab_fd < 0) {
1470         error_report_err(local_err);
1471     }
1472 
1473     return spapr->htab_fd;
1474 }
1475 
1476 void close_htab_fd(SpaprMachineState *spapr)
1477 {
1478     if (spapr->htab_fd >= 0) {
1479         close(spapr->htab_fd);
1480     }
1481     spapr->htab_fd = -1;
1482 }
1483 
1484 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1485 {
1486     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1487 
1488     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1489 }
1490 
1491 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1492 {
1493     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1494 
1495     assert(kvm_enabled());
1496 
1497     if (!spapr->htab) {
1498         return 0;
1499     }
1500 
1501     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1502 }
1503 
1504 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1505                                                 hwaddr ptex, int n)
1506 {
1507     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1508     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1509 
1510     if (!spapr->htab) {
1511         /*
1512          * HTAB is controlled by KVM. Fetch into temporary buffer
1513          */
1514         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1515         kvmppc_read_hptes(hptes, ptex, n);
1516         return hptes;
1517     }
1518 
1519     /*
1520      * HTAB is controlled by QEMU. Just point to the internally
1521      * accessible PTEG.
1522      */
1523     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1524 }
1525 
1526 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1527                               const ppc_hash_pte64_t *hptes,
1528                               hwaddr ptex, int n)
1529 {
1530     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1531 
1532     if (!spapr->htab) {
1533         g_free((void *)hptes);
1534     }
1535 
1536     /* Nothing to do for qemu managed HPT */
1537 }
1538 
1539 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1540                       uint64_t pte0, uint64_t pte1)
1541 {
1542     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1543     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1544 
1545     if (!spapr->htab) {
1546         kvmppc_write_hpte(ptex, pte0, pte1);
1547     } else {
1548         if (pte0 & HPTE64_V_VALID) {
1549             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1550             /*
1551              * When setting valid, we write PTE1 first. This ensures
1552              * proper synchronization with the reading code in
1553              * ppc_hash64_pteg_search()
1554              */
1555             smp_wmb();
1556             stq_p(spapr->htab + offset, pte0);
1557         } else {
1558             stq_p(spapr->htab + offset, pte0);
1559             /*
1560              * When clearing it we set PTE0 first. This ensures proper
1561              * synchronization with the reading code in
1562              * ppc_hash64_pteg_search()
1563              */
1564             smp_wmb();
1565             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1566         }
1567     }
1568 }
1569 
1570 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1571                              uint64_t pte1)
1572 {
1573     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1574     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1575 
1576     if (!spapr->htab) {
1577         /* There should always be a hash table when this is called */
1578         error_report("spapr_hpte_set_c called with no hash table !");
1579         return;
1580     }
1581 
1582     /* The HW performs a non-atomic byte update */
1583     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1584 }
1585 
1586 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1587                              uint64_t pte1)
1588 {
1589     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1590     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1591 
1592     if (!spapr->htab) {
1593         /* There should always be a hash table when this is called */
1594         error_report("spapr_hpte_set_r called with no hash table !");
1595         return;
1596     }
1597 
1598     /* The HW performs a non-atomic byte update */
1599     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1600 }
1601 
1602 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1603 {
1604     int shift;
1605 
1606     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1607      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1608      * that's much more than is needed for Linux guests */
1609     shift = ctz64(pow2ceil(ramsize)) - 7;
1610     shift = MAX(shift, 18); /* Minimum architected size */
1611     shift = MIN(shift, 46); /* Maximum architected size */
1612     return shift;
1613 }
1614 
1615 void spapr_free_hpt(SpaprMachineState *spapr)
1616 {
1617     qemu_vfree(spapr->htab);
1618     spapr->htab = NULL;
1619     spapr->htab_shift = 0;
1620     close_htab_fd(spapr);
1621 }
1622 
1623 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1624 {
1625     ERRP_GUARD();
1626     long rc;
1627 
1628     /* Clean up any HPT info from a previous boot */
1629     spapr_free_hpt(spapr);
1630 
1631     rc = kvmppc_reset_htab(shift);
1632 
1633     if (rc == -EOPNOTSUPP) {
1634         error_setg(errp, "HPT not supported in nested guests");
1635         return -EOPNOTSUPP;
1636     }
1637 
1638     if (rc < 0) {
1639         /* kernel-side HPT needed, but couldn't allocate one */
1640         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1641                          shift);
1642         error_append_hint(errp, "Try smaller maxmem?\n");
1643         return -errno;
1644     } else if (rc > 0) {
1645         /* kernel-side HPT allocated */
1646         if (rc != shift) {
1647             error_setg(errp,
1648                        "Requested order %d HPT, but kernel allocated order %ld",
1649                        shift, rc);
1650             error_append_hint(errp, "Try smaller maxmem?\n");
1651             return -ENOSPC;
1652         }
1653 
1654         spapr->htab_shift = shift;
1655         spapr->htab = NULL;
1656     } else {
1657         /* kernel-side HPT not needed, allocate in userspace instead */
1658         size_t size = 1ULL << shift;
1659         int i;
1660 
1661         spapr->htab = qemu_memalign(size, size);
1662         memset(spapr->htab, 0, size);
1663         spapr->htab_shift = shift;
1664 
1665         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1666             DIRTY_HPTE(HPTE(spapr->htab, i));
1667         }
1668     }
1669     /* We're setting up a hash table, so that means we're not radix */
1670     spapr->patb_entry = 0;
1671     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1672     return 0;
1673 }
1674 
1675 void spapr_setup_hpt(SpaprMachineState *spapr)
1676 {
1677     int hpt_shift;
1678 
1679     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1680         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1681     } else {
1682         uint64_t current_ram_size;
1683 
1684         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1685         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1686     }
1687     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1688 
1689     if (kvm_enabled()) {
1690         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1691 
1692         /* Check our RMA fits in the possible VRMA */
1693         if (vrma_limit < spapr->rma_size) {
1694             error_report("Unable to create %" HWADDR_PRIu
1695                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1696                          spapr->rma_size / MiB, vrma_limit / MiB);
1697             exit(EXIT_FAILURE);
1698         }
1699     }
1700 }
1701 
1702 void spapr_check_mmu_mode(bool guest_radix)
1703 {
1704     if (guest_radix) {
1705         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1706             error_report("Guest requested unavailable MMU mode (radix).");
1707             exit(EXIT_FAILURE);
1708         }
1709     } else {
1710         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1711             && !kvmppc_has_cap_mmu_hash_v3()) {
1712             error_report("Guest requested unavailable MMU mode (hash).");
1713             exit(EXIT_FAILURE);
1714         }
1715     }
1716 }
1717 
1718 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1719 {
1720     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1721     PowerPCCPU *first_ppc_cpu;
1722     hwaddr fdt_addr;
1723     void *fdt;
1724     int rc;
1725 
1726     if (reason != SHUTDOWN_CAUSE_SNAPSHOT_LOAD) {
1727         /*
1728          * Record-replay snapshot load must not consume random, this was
1729          * already replayed from initial machine reset.
1730          */
1731         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1732     }
1733 
1734     pef_kvm_reset(machine->cgs, &error_fatal);
1735     spapr_caps_apply(spapr);
1736 
1737     first_ppc_cpu = POWERPC_CPU(first_cpu);
1738     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1739         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1740                               spapr->max_compat_pvr)) {
1741         /*
1742          * If using KVM with radix mode available, VCPUs can be started
1743          * without a HPT because KVM will start them in radix mode.
1744          * Set the GR bit in PATE so that we know there is no HPT.
1745          */
1746         spapr->patb_entry = PATE1_GR;
1747         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1748     } else {
1749         spapr_setup_hpt(spapr);
1750     }
1751 
1752     qemu_devices_reset(reason);
1753 
1754     spapr_ovec_cleanup(spapr->ov5_cas);
1755     spapr->ov5_cas = spapr_ovec_new();
1756 
1757     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1758 
1759     /*
1760      * This is fixing some of the default configuration of the XIVE
1761      * devices. To be called after the reset of the machine devices.
1762      */
1763     spapr_irq_reset(spapr, &error_fatal);
1764 
1765     /*
1766      * There is no CAS under qtest. Simulate one to please the code that
1767      * depends on spapr->ov5_cas. This is especially needed to test device
1768      * unplug, so we do that before resetting the DRCs.
1769      */
1770     if (qtest_enabled()) {
1771         spapr_ovec_cleanup(spapr->ov5_cas);
1772         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1773     }
1774 
1775     spapr_nvdimm_finish_flushes();
1776 
1777     /* DRC reset may cause a device to be unplugged. This will cause troubles
1778      * if this device is used by another device (eg, a running vhost backend
1779      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1780      * situations, we reset DRCs after all devices have been reset.
1781      */
1782     spapr_drc_reset_all(spapr);
1783 
1784     spapr_clear_pending_events(spapr);
1785 
1786     /*
1787      * We place the device tree just below either the top of the RMA,
1788      * or just below 2GB, whichever is lower, so that it can be
1789      * processed with 32-bit real mode code if necessary
1790      */
1791     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1792 
1793     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1794     if (spapr->vof) {
1795         spapr_vof_reset(spapr, fdt, &error_fatal);
1796         /*
1797          * Do not pack the FDT as the client may change properties.
1798          * VOF client does not expect the FDT so we do not load it to the VM.
1799          */
1800     } else {
1801         rc = fdt_pack(fdt);
1802         /* Should only fail if we've built a corrupted tree */
1803         assert(rc == 0);
1804 
1805         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1806                                   0, fdt_addr, 0);
1807         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1808     }
1809     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1810 
1811     g_free(spapr->fdt_blob);
1812     spapr->fdt_size = fdt_totalsize(fdt);
1813     spapr->fdt_initial_size = spapr->fdt_size;
1814     spapr->fdt_blob = fdt;
1815 
1816     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1817     machine->fdt = fdt;
1818 
1819     /* Set up the entry state */
1820     first_ppc_cpu->env.gpr[5] = 0;
1821 
1822     spapr->fwnmi_system_reset_addr = -1;
1823     spapr->fwnmi_machine_check_addr = -1;
1824     spapr->fwnmi_machine_check_interlock = -1;
1825 
1826     /* Signal all vCPUs waiting on this condition */
1827     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1828 
1829     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1830 }
1831 
1832 static void spapr_create_nvram(SpaprMachineState *spapr)
1833 {
1834     DeviceState *dev = qdev_new("spapr-nvram");
1835     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1836 
1837     if (dinfo) {
1838         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1839                                 &error_fatal);
1840     }
1841 
1842     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1843 
1844     spapr->nvram = (struct SpaprNvram *)dev;
1845 }
1846 
1847 static void spapr_rtc_create(SpaprMachineState *spapr)
1848 {
1849     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1850                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1851                                        &error_fatal, NULL);
1852     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1853     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1854                               "date");
1855 }
1856 
1857 /* Returns whether we want to use VGA or not */
1858 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1859 {
1860     vga_interface_created = true;
1861     switch (vga_interface_type) {
1862     case VGA_NONE:
1863         return false;
1864     case VGA_DEVICE:
1865         return true;
1866     case VGA_STD:
1867     case VGA_VIRTIO:
1868     case VGA_CIRRUS:
1869         return pci_vga_init(pci_bus) != NULL;
1870     default:
1871         error_setg(errp,
1872                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1873         return false;
1874     }
1875 }
1876 
1877 static int spapr_pre_load(void *opaque)
1878 {
1879     int rc;
1880 
1881     rc = spapr_caps_pre_load(opaque);
1882     if (rc) {
1883         return rc;
1884     }
1885 
1886     return 0;
1887 }
1888 
1889 static int spapr_post_load(void *opaque, int version_id)
1890 {
1891     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1892     int err = 0;
1893 
1894     err = spapr_caps_post_migration(spapr);
1895     if (err) {
1896         return err;
1897     }
1898 
1899     /*
1900      * In earlier versions, there was no separate qdev for the PAPR
1901      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1902      * So when migrating from those versions, poke the incoming offset
1903      * value into the RTC device
1904      */
1905     if (version_id < 3) {
1906         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1907         if (err) {
1908             return err;
1909         }
1910     }
1911 
1912     if (kvm_enabled() && spapr->patb_entry) {
1913         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1914         bool radix = !!(spapr->patb_entry & PATE1_GR);
1915         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1916 
1917         /*
1918          * Update LPCR:HR and UPRT as they may not be set properly in
1919          * the stream
1920          */
1921         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1922                             LPCR_HR | LPCR_UPRT);
1923 
1924         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1925         if (err) {
1926             error_report("Process table config unsupported by the host");
1927             return -EINVAL;
1928         }
1929     }
1930 
1931     err = spapr_irq_post_load(spapr, version_id);
1932     if (err) {
1933         return err;
1934     }
1935 
1936     return err;
1937 }
1938 
1939 static int spapr_pre_save(void *opaque)
1940 {
1941     int rc;
1942 
1943     rc = spapr_caps_pre_save(opaque);
1944     if (rc) {
1945         return rc;
1946     }
1947 
1948     return 0;
1949 }
1950 
1951 static bool version_before_3(void *opaque, int version_id)
1952 {
1953     return version_id < 3;
1954 }
1955 
1956 static bool spapr_pending_events_needed(void *opaque)
1957 {
1958     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1959     return !QTAILQ_EMPTY(&spapr->pending_events);
1960 }
1961 
1962 static const VMStateDescription vmstate_spapr_event_entry = {
1963     .name = "spapr_event_log_entry",
1964     .version_id = 1,
1965     .minimum_version_id = 1,
1966     .fields = (const VMStateField[]) {
1967         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1968         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1969         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1970                                      NULL, extended_length),
1971         VMSTATE_END_OF_LIST()
1972     },
1973 };
1974 
1975 static const VMStateDescription vmstate_spapr_pending_events = {
1976     .name = "spapr_pending_events",
1977     .version_id = 1,
1978     .minimum_version_id = 1,
1979     .needed = spapr_pending_events_needed,
1980     .fields = (const VMStateField[]) {
1981         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1982                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1983         VMSTATE_END_OF_LIST()
1984     },
1985 };
1986 
1987 static bool spapr_ov5_cas_needed(void *opaque)
1988 {
1989     SpaprMachineState *spapr = opaque;
1990     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1991     bool cas_needed;
1992 
1993     /* Prior to the introduction of SpaprOptionVector, we had two option
1994      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1995      * Both of these options encode machine topology into the device-tree
1996      * in such a way that the now-booted OS should still be able to interact
1997      * appropriately with QEMU regardless of what options were actually
1998      * negotiatied on the source side.
1999      *
2000      * As such, we can avoid migrating the CAS-negotiated options if these
2001      * are the only options available on the current machine/platform.
2002      * Since these are the only options available for pseries-2.7 and
2003      * earlier, this allows us to maintain old->new/new->old migration
2004      * compatibility.
2005      *
2006      * For QEMU 2.8+, there are additional CAS-negotiatable options available
2007      * via default pseries-2.8 machines and explicit command-line parameters.
2008      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
2009      * of the actual CAS-negotiated values to continue working properly. For
2010      * example, availability of memory unplug depends on knowing whether
2011      * OV5_HP_EVT was negotiated via CAS.
2012      *
2013      * Thus, for any cases where the set of available CAS-negotiatable
2014      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2015      * include the CAS-negotiated options in the migration stream, unless
2016      * if they affect boot time behaviour only.
2017      */
2018     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2019     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2020     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2021 
2022     /* We need extra information if we have any bits outside the mask
2023      * defined above */
2024     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
2025 
2026     spapr_ovec_cleanup(ov5_mask);
2027 
2028     return cas_needed;
2029 }
2030 
2031 static const VMStateDescription vmstate_spapr_ov5_cas = {
2032     .name = "spapr_option_vector_ov5_cas",
2033     .version_id = 1,
2034     .minimum_version_id = 1,
2035     .needed = spapr_ov5_cas_needed,
2036     .fields = (const VMStateField[]) {
2037         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2038                                  vmstate_spapr_ovec, SpaprOptionVector),
2039         VMSTATE_END_OF_LIST()
2040     },
2041 };
2042 
2043 static bool spapr_patb_entry_needed(void *opaque)
2044 {
2045     SpaprMachineState *spapr = opaque;
2046 
2047     return !!spapr->patb_entry;
2048 }
2049 
2050 static const VMStateDescription vmstate_spapr_patb_entry = {
2051     .name = "spapr_patb_entry",
2052     .version_id = 1,
2053     .minimum_version_id = 1,
2054     .needed = spapr_patb_entry_needed,
2055     .fields = (const VMStateField[]) {
2056         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2057         VMSTATE_END_OF_LIST()
2058     },
2059 };
2060 
2061 static bool spapr_irq_map_needed(void *opaque)
2062 {
2063     SpaprMachineState *spapr = opaque;
2064 
2065     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2066 }
2067 
2068 static const VMStateDescription vmstate_spapr_irq_map = {
2069     .name = "spapr_irq_map",
2070     .version_id = 1,
2071     .minimum_version_id = 1,
2072     .needed = spapr_irq_map_needed,
2073     .fields = (const VMStateField[]) {
2074         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2075         VMSTATE_END_OF_LIST()
2076     },
2077 };
2078 
2079 static bool spapr_dtb_needed(void *opaque)
2080 {
2081     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2082 
2083     return smc->update_dt_enabled;
2084 }
2085 
2086 static int spapr_dtb_pre_load(void *opaque)
2087 {
2088     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2089 
2090     g_free(spapr->fdt_blob);
2091     spapr->fdt_blob = NULL;
2092     spapr->fdt_size = 0;
2093 
2094     return 0;
2095 }
2096 
2097 static const VMStateDescription vmstate_spapr_dtb = {
2098     .name = "spapr_dtb",
2099     .version_id = 1,
2100     .minimum_version_id = 1,
2101     .needed = spapr_dtb_needed,
2102     .pre_load = spapr_dtb_pre_load,
2103     .fields = (const VMStateField[]) {
2104         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2105         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2106         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2107                                      fdt_size),
2108         VMSTATE_END_OF_LIST()
2109     },
2110 };
2111 
2112 static bool spapr_fwnmi_needed(void *opaque)
2113 {
2114     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2115 
2116     return spapr->fwnmi_machine_check_addr != -1;
2117 }
2118 
2119 static int spapr_fwnmi_pre_save(void *opaque)
2120 {
2121     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2122 
2123     /*
2124      * Check if machine check handling is in progress and print a
2125      * warning message.
2126      */
2127     if (spapr->fwnmi_machine_check_interlock != -1) {
2128         warn_report("A machine check is being handled during migration. The"
2129                 "handler may run and log hardware error on the destination");
2130     }
2131 
2132     return 0;
2133 }
2134 
2135 static const VMStateDescription vmstate_spapr_fwnmi = {
2136     .name = "spapr_fwnmi",
2137     .version_id = 1,
2138     .minimum_version_id = 1,
2139     .needed = spapr_fwnmi_needed,
2140     .pre_save = spapr_fwnmi_pre_save,
2141     .fields = (const VMStateField[]) {
2142         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2143         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2144         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2145         VMSTATE_END_OF_LIST()
2146     },
2147 };
2148 
2149 static const VMStateDescription vmstate_spapr = {
2150     .name = "spapr",
2151     .version_id = 3,
2152     .minimum_version_id = 1,
2153     .pre_load = spapr_pre_load,
2154     .post_load = spapr_post_load,
2155     .pre_save = spapr_pre_save,
2156     .fields = (const VMStateField[]) {
2157         /* used to be @next_irq */
2158         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2159 
2160         /* RTC offset */
2161         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2162 
2163         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2164         VMSTATE_END_OF_LIST()
2165     },
2166     .subsections = (const VMStateDescription * const []) {
2167         &vmstate_spapr_ov5_cas,
2168         &vmstate_spapr_patb_entry,
2169         &vmstate_spapr_pending_events,
2170         &vmstate_spapr_cap_htm,
2171         &vmstate_spapr_cap_vsx,
2172         &vmstate_spapr_cap_dfp,
2173         &vmstate_spapr_cap_cfpc,
2174         &vmstate_spapr_cap_sbbc,
2175         &vmstate_spapr_cap_ibs,
2176         &vmstate_spapr_cap_hpt_maxpagesize,
2177         &vmstate_spapr_irq_map,
2178         &vmstate_spapr_cap_nested_kvm_hv,
2179         &vmstate_spapr_dtb,
2180         &vmstate_spapr_cap_large_decr,
2181         &vmstate_spapr_cap_ccf_assist,
2182         &vmstate_spapr_cap_fwnmi,
2183         &vmstate_spapr_fwnmi,
2184         &vmstate_spapr_cap_rpt_invalidate,
2185         NULL
2186     }
2187 };
2188 
2189 static int htab_save_setup(QEMUFile *f, void *opaque)
2190 {
2191     SpaprMachineState *spapr = opaque;
2192 
2193     /* "Iteration" header */
2194     if (!spapr->htab_shift) {
2195         qemu_put_be32(f, -1);
2196     } else {
2197         qemu_put_be32(f, spapr->htab_shift);
2198     }
2199 
2200     if (spapr->htab) {
2201         spapr->htab_save_index = 0;
2202         spapr->htab_first_pass = true;
2203     } else {
2204         if (spapr->htab_shift) {
2205             assert(kvm_enabled());
2206         }
2207     }
2208 
2209 
2210     return 0;
2211 }
2212 
2213 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2214                             int chunkstart, int n_valid, int n_invalid)
2215 {
2216     qemu_put_be32(f, chunkstart);
2217     qemu_put_be16(f, n_valid);
2218     qemu_put_be16(f, n_invalid);
2219     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2220                     HASH_PTE_SIZE_64 * n_valid);
2221 }
2222 
2223 static void htab_save_end_marker(QEMUFile *f)
2224 {
2225     qemu_put_be32(f, 0);
2226     qemu_put_be16(f, 0);
2227     qemu_put_be16(f, 0);
2228 }
2229 
2230 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2231                                  int64_t max_ns)
2232 {
2233     bool has_timeout = max_ns != -1;
2234     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2235     int index = spapr->htab_save_index;
2236     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2237 
2238     assert(spapr->htab_first_pass);
2239 
2240     do {
2241         int chunkstart;
2242 
2243         /* Consume invalid HPTEs */
2244         while ((index < htabslots)
2245                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2246             CLEAN_HPTE(HPTE(spapr->htab, index));
2247             index++;
2248         }
2249 
2250         /* Consume valid HPTEs */
2251         chunkstart = index;
2252         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2253                && HPTE_VALID(HPTE(spapr->htab, index))) {
2254             CLEAN_HPTE(HPTE(spapr->htab, index));
2255             index++;
2256         }
2257 
2258         if (index > chunkstart) {
2259             int n_valid = index - chunkstart;
2260 
2261             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2262 
2263             if (has_timeout &&
2264                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2265                 break;
2266             }
2267         }
2268     } while ((index < htabslots) && !migration_rate_exceeded(f));
2269 
2270     if (index >= htabslots) {
2271         assert(index == htabslots);
2272         index = 0;
2273         spapr->htab_first_pass = false;
2274     }
2275     spapr->htab_save_index = index;
2276 }
2277 
2278 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2279                                 int64_t max_ns)
2280 {
2281     bool final = max_ns < 0;
2282     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2283     int examined = 0, sent = 0;
2284     int index = spapr->htab_save_index;
2285     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2286 
2287     assert(!spapr->htab_first_pass);
2288 
2289     do {
2290         int chunkstart, invalidstart;
2291 
2292         /* Consume non-dirty HPTEs */
2293         while ((index < htabslots)
2294                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2295             index++;
2296             examined++;
2297         }
2298 
2299         chunkstart = index;
2300         /* Consume valid dirty HPTEs */
2301         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2302                && HPTE_DIRTY(HPTE(spapr->htab, index))
2303                && HPTE_VALID(HPTE(spapr->htab, index))) {
2304             CLEAN_HPTE(HPTE(spapr->htab, index));
2305             index++;
2306             examined++;
2307         }
2308 
2309         invalidstart = index;
2310         /* Consume invalid dirty HPTEs */
2311         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2312                && HPTE_DIRTY(HPTE(spapr->htab, index))
2313                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2314             CLEAN_HPTE(HPTE(spapr->htab, index));
2315             index++;
2316             examined++;
2317         }
2318 
2319         if (index > chunkstart) {
2320             int n_valid = invalidstart - chunkstart;
2321             int n_invalid = index - invalidstart;
2322 
2323             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2324             sent += index - chunkstart;
2325 
2326             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2327                 break;
2328             }
2329         }
2330 
2331         if (examined >= htabslots) {
2332             break;
2333         }
2334 
2335         if (index >= htabslots) {
2336             assert(index == htabslots);
2337             index = 0;
2338         }
2339     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2340 
2341     if (index >= htabslots) {
2342         assert(index == htabslots);
2343         index = 0;
2344     }
2345 
2346     spapr->htab_save_index = index;
2347 
2348     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2349 }
2350 
2351 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2352 #define MAX_KVM_BUF_SIZE    2048
2353 
2354 static int htab_save_iterate(QEMUFile *f, void *opaque)
2355 {
2356     SpaprMachineState *spapr = opaque;
2357     int fd;
2358     int rc = 0;
2359 
2360     /* Iteration header */
2361     if (!spapr->htab_shift) {
2362         qemu_put_be32(f, -1);
2363         return 1;
2364     } else {
2365         qemu_put_be32(f, 0);
2366     }
2367 
2368     if (!spapr->htab) {
2369         assert(kvm_enabled());
2370 
2371         fd = get_htab_fd(spapr);
2372         if (fd < 0) {
2373             return fd;
2374         }
2375 
2376         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2377         if (rc < 0) {
2378             return rc;
2379         }
2380     } else  if (spapr->htab_first_pass) {
2381         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2382     } else {
2383         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2384     }
2385 
2386     htab_save_end_marker(f);
2387 
2388     return rc;
2389 }
2390 
2391 static int htab_save_complete(QEMUFile *f, void *opaque)
2392 {
2393     SpaprMachineState *spapr = opaque;
2394     int fd;
2395 
2396     /* Iteration header */
2397     if (!spapr->htab_shift) {
2398         qemu_put_be32(f, -1);
2399         return 0;
2400     } else {
2401         qemu_put_be32(f, 0);
2402     }
2403 
2404     if (!spapr->htab) {
2405         int rc;
2406 
2407         assert(kvm_enabled());
2408 
2409         fd = get_htab_fd(spapr);
2410         if (fd < 0) {
2411             return fd;
2412         }
2413 
2414         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2415         if (rc < 0) {
2416             return rc;
2417         }
2418     } else {
2419         if (spapr->htab_first_pass) {
2420             htab_save_first_pass(f, spapr, -1);
2421         }
2422         htab_save_later_pass(f, spapr, -1);
2423     }
2424 
2425     /* End marker */
2426     htab_save_end_marker(f);
2427 
2428     return 0;
2429 }
2430 
2431 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2432 {
2433     SpaprMachineState *spapr = opaque;
2434     uint32_t section_hdr;
2435     int fd = -1;
2436     Error *local_err = NULL;
2437 
2438     if (version_id < 1 || version_id > 1) {
2439         error_report("htab_load() bad version");
2440         return -EINVAL;
2441     }
2442 
2443     section_hdr = qemu_get_be32(f);
2444 
2445     if (section_hdr == -1) {
2446         spapr_free_hpt(spapr);
2447         return 0;
2448     }
2449 
2450     if (section_hdr) {
2451         int ret;
2452 
2453         /* First section gives the htab size */
2454         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2455         if (ret < 0) {
2456             error_report_err(local_err);
2457             return ret;
2458         }
2459         return 0;
2460     }
2461 
2462     if (!spapr->htab) {
2463         assert(kvm_enabled());
2464 
2465         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2466         if (fd < 0) {
2467             error_report_err(local_err);
2468             return fd;
2469         }
2470     }
2471 
2472     while (true) {
2473         uint32_t index;
2474         uint16_t n_valid, n_invalid;
2475 
2476         index = qemu_get_be32(f);
2477         n_valid = qemu_get_be16(f);
2478         n_invalid = qemu_get_be16(f);
2479 
2480         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2481             /* End of Stream */
2482             break;
2483         }
2484 
2485         if ((index + n_valid + n_invalid) >
2486             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2487             /* Bad index in stream */
2488             error_report(
2489                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2490                 index, n_valid, n_invalid, spapr->htab_shift);
2491             return -EINVAL;
2492         }
2493 
2494         if (spapr->htab) {
2495             if (n_valid) {
2496                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2497                                 HASH_PTE_SIZE_64 * n_valid);
2498             }
2499             if (n_invalid) {
2500                 memset(HPTE(spapr->htab, index + n_valid), 0,
2501                        HASH_PTE_SIZE_64 * n_invalid);
2502             }
2503         } else {
2504             int rc;
2505 
2506             assert(fd >= 0);
2507 
2508             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2509                                         &local_err);
2510             if (rc < 0) {
2511                 error_report_err(local_err);
2512                 return rc;
2513             }
2514         }
2515     }
2516 
2517     if (!spapr->htab) {
2518         assert(fd >= 0);
2519         close(fd);
2520     }
2521 
2522     return 0;
2523 }
2524 
2525 static void htab_save_cleanup(void *opaque)
2526 {
2527     SpaprMachineState *spapr = opaque;
2528 
2529     close_htab_fd(spapr);
2530 }
2531 
2532 static SaveVMHandlers savevm_htab_handlers = {
2533     .save_setup = htab_save_setup,
2534     .save_live_iterate = htab_save_iterate,
2535     .save_live_complete_precopy = htab_save_complete,
2536     .save_cleanup = htab_save_cleanup,
2537     .load_state = htab_load,
2538 };
2539 
2540 static void spapr_boot_set(void *opaque, const char *boot_device,
2541                            Error **errp)
2542 {
2543     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2544 
2545     g_free(spapr->boot_device);
2546     spapr->boot_device = g_strdup(boot_device);
2547 }
2548 
2549 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2550 {
2551     MachineState *machine = MACHINE(spapr);
2552     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2553     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2554     int i;
2555 
2556     g_assert(!nr_lmbs || machine->device_memory);
2557     for (i = 0; i < nr_lmbs; i++) {
2558         uint64_t addr;
2559 
2560         addr = i * lmb_size + machine->device_memory->base;
2561         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2562                                addr / lmb_size);
2563     }
2564 }
2565 
2566 /*
2567  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2568  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2569  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2570  */
2571 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2572 {
2573     int i;
2574 
2575     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2576         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2577                    " is not aligned to %" PRIu64 " MiB",
2578                    machine->ram_size,
2579                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2580         return;
2581     }
2582 
2583     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2584         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2585                    " is not aligned to %" PRIu64 " MiB",
2586                    machine->ram_size,
2587                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2588         return;
2589     }
2590 
2591     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2592         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2593             error_setg(errp,
2594                        "Node %d memory size 0x%" PRIx64
2595                        " is not aligned to %" PRIu64 " MiB",
2596                        i, machine->numa_state->nodes[i].node_mem,
2597                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2598             return;
2599         }
2600     }
2601 }
2602 
2603 /* find cpu slot in machine->possible_cpus by core_id */
2604 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2605 {
2606     int index = id / ms->smp.threads;
2607 
2608     if (index >= ms->possible_cpus->len) {
2609         return NULL;
2610     }
2611     if (idx) {
2612         *idx = index;
2613     }
2614     return &ms->possible_cpus->cpus[index];
2615 }
2616 
2617 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2618 {
2619     MachineState *ms = MACHINE(spapr);
2620     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2621     Error *local_err = NULL;
2622     bool vsmt_user = !!spapr->vsmt;
2623     int kvm_smt = kvmppc_smt_threads();
2624     int ret;
2625     unsigned int smp_threads = ms->smp.threads;
2626 
2627     if (tcg_enabled()) {
2628         if (smp_threads > 1 &&
2629             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2630                                    spapr->max_compat_pvr)) {
2631             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2632             return;
2633         }
2634 
2635         if (smp_threads > 8) {
2636             error_setg(errp, "TCG cannot support more than 8 threads/core "
2637                        "on a pseries machine");
2638             return;
2639         }
2640     }
2641     if (!is_power_of_2(smp_threads)) {
2642         error_setg(errp, "Cannot support %d threads/core on a pseries "
2643                    "machine because it must be a power of 2", smp_threads);
2644         return;
2645     }
2646 
2647     /* Determine the VSMT mode to use: */
2648     if (vsmt_user) {
2649         if (spapr->vsmt < smp_threads) {
2650             error_setg(errp, "Cannot support VSMT mode %d"
2651                        " because it must be >= threads/core (%d)",
2652                        spapr->vsmt, smp_threads);
2653             return;
2654         }
2655         /* In this case, spapr->vsmt has been set by the command line */
2656     } else if (!smc->smp_threads_vsmt) {
2657         /*
2658          * Default VSMT value is tricky, because we need it to be as
2659          * consistent as possible (for migration), but this requires
2660          * changing it for at least some existing cases.  We pick 8 as
2661          * the value that we'd get with KVM on POWER8, the
2662          * overwhelmingly common case in production systems.
2663          */
2664         spapr->vsmt = MAX(8, smp_threads);
2665     } else {
2666         spapr->vsmt = smp_threads;
2667     }
2668 
2669     /* KVM: If necessary, set the SMT mode: */
2670     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2671         ret = kvmppc_set_smt_threads(spapr->vsmt);
2672         if (ret) {
2673             /* Looks like KVM isn't able to change VSMT mode */
2674             error_setg(&local_err,
2675                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2676                        spapr->vsmt, ret);
2677             /* We can live with that if the default one is big enough
2678              * for the number of threads, and a submultiple of the one
2679              * we want.  In this case we'll waste some vcpu ids, but
2680              * behaviour will be correct */
2681             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2682                 warn_report_err(local_err);
2683             } else {
2684                 if (!vsmt_user) {
2685                     error_append_hint(&local_err,
2686                                       "On PPC, a VM with %d threads/core"
2687                                       " on a host with %d threads/core"
2688                                       " requires the use of VSMT mode %d.\n",
2689                                       smp_threads, kvm_smt, spapr->vsmt);
2690                 }
2691                 kvmppc_error_append_smt_possible_hint(&local_err);
2692                 error_propagate(errp, local_err);
2693             }
2694         }
2695     }
2696     /* else TCG: nothing to do currently */
2697 }
2698 
2699 static void spapr_init_cpus(SpaprMachineState *spapr)
2700 {
2701     MachineState *machine = MACHINE(spapr);
2702     MachineClass *mc = MACHINE_GET_CLASS(machine);
2703     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2704     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2705     const CPUArchIdList *possible_cpus;
2706     unsigned int smp_cpus = machine->smp.cpus;
2707     unsigned int smp_threads = machine->smp.threads;
2708     unsigned int max_cpus = machine->smp.max_cpus;
2709     int boot_cores_nr = smp_cpus / smp_threads;
2710     int i;
2711 
2712     possible_cpus = mc->possible_cpu_arch_ids(machine);
2713     if (mc->has_hotpluggable_cpus) {
2714         if (smp_cpus % smp_threads) {
2715             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2716                          smp_cpus, smp_threads);
2717             exit(1);
2718         }
2719         if (max_cpus % smp_threads) {
2720             error_report("max_cpus (%u) must be multiple of threads (%u)",
2721                          max_cpus, smp_threads);
2722             exit(1);
2723         }
2724     } else {
2725         if (max_cpus != smp_cpus) {
2726             error_report("This machine version does not support CPU hotplug");
2727             exit(1);
2728         }
2729         boot_cores_nr = possible_cpus->len;
2730     }
2731 
2732     if (smc->pre_2_10_has_unused_icps) {
2733         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2734             /* Dummy entries get deregistered when real ICPState objects
2735              * are registered during CPU core hotplug.
2736              */
2737             pre_2_10_vmstate_register_dummy_icp(i);
2738         }
2739     }
2740 
2741     for (i = 0; i < possible_cpus->len; i++) {
2742         int core_id = i * smp_threads;
2743 
2744         if (mc->has_hotpluggable_cpus) {
2745             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2746                                    spapr_vcpu_id(spapr, core_id));
2747         }
2748 
2749         if (i < boot_cores_nr) {
2750             Object *core  = object_new(type);
2751             int nr_threads = smp_threads;
2752 
2753             /* Handle the partially filled core for older machine types */
2754             if ((i + 1) * smp_threads >= smp_cpus) {
2755                 nr_threads = smp_cpus - i * smp_threads;
2756             }
2757 
2758             object_property_set_int(core, "nr-threads", nr_threads,
2759                                     &error_fatal);
2760             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2761                                     &error_fatal);
2762             qdev_realize(DEVICE(core), NULL, &error_fatal);
2763 
2764             object_unref(core);
2765         }
2766     }
2767 }
2768 
2769 static PCIHostState *spapr_create_default_phb(void)
2770 {
2771     DeviceState *dev;
2772 
2773     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2774     qdev_prop_set_uint32(dev, "index", 0);
2775     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2776 
2777     return PCI_HOST_BRIDGE(dev);
2778 }
2779 
2780 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2781 {
2782     MachineState *machine = MACHINE(spapr);
2783     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2784     hwaddr rma_size = machine->ram_size;
2785     hwaddr node0_size = spapr_node0_size(machine);
2786 
2787     /* RMA has to fit in the first NUMA node */
2788     rma_size = MIN(rma_size, node0_size);
2789 
2790     /*
2791      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2792      * never exceed that
2793      */
2794     rma_size = MIN(rma_size, 1 * TiB);
2795 
2796     /*
2797      * Clamp the RMA size based on machine type.  This is for
2798      * migration compatibility with older qemu versions, which limited
2799      * the RMA size for complicated and mostly bad reasons.
2800      */
2801     if (smc->rma_limit) {
2802         rma_size = MIN(rma_size, smc->rma_limit);
2803     }
2804 
2805     if (rma_size < MIN_RMA_SLOF) {
2806         error_setg(errp,
2807                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2808                    "ldMiB guest RMA (Real Mode Area memory)",
2809                    MIN_RMA_SLOF / MiB);
2810         return 0;
2811     }
2812 
2813     return rma_size;
2814 }
2815 
2816 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2817 {
2818     MachineState *machine = MACHINE(spapr);
2819     int i;
2820 
2821     for (i = 0; i < machine->ram_slots; i++) {
2822         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2823     }
2824 }
2825 
2826 /* pSeries LPAR / sPAPR hardware init */
2827 static void spapr_machine_init(MachineState *machine)
2828 {
2829     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2830     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2831     MachineClass *mc = MACHINE_GET_CLASS(machine);
2832     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2833     const char *bios_name = machine->firmware ?: bios_default;
2834     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2835     const char *kernel_filename = machine->kernel_filename;
2836     const char *initrd_filename = machine->initrd_filename;
2837     PCIHostState *phb;
2838     bool has_vga;
2839     int i;
2840     MemoryRegion *sysmem = get_system_memory();
2841     long load_limit, fw_size;
2842     Error *resize_hpt_err = NULL;
2843     NICInfo *nd;
2844 
2845     if (!filename) {
2846         error_report("Could not find LPAR firmware '%s'", bios_name);
2847         exit(1);
2848     }
2849     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2850     if (fw_size <= 0) {
2851         error_report("Could not load LPAR firmware '%s'", filename);
2852         exit(1);
2853     }
2854 
2855     /*
2856      * if Secure VM (PEF) support is configured, then initialize it
2857      */
2858     pef_kvm_init(machine->cgs, &error_fatal);
2859 
2860     msi_nonbroken = true;
2861 
2862     QLIST_INIT(&spapr->phbs);
2863     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2864 
2865     /* Determine capabilities to run with */
2866     spapr_caps_init(spapr);
2867 
2868     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2869     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2870         /*
2871          * If the user explicitly requested a mode we should either
2872          * supply it, or fail completely (which we do below).  But if
2873          * it's not set explicitly, we reset our mode to something
2874          * that works
2875          */
2876         if (resize_hpt_err) {
2877             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2878             error_free(resize_hpt_err);
2879             resize_hpt_err = NULL;
2880         } else {
2881             spapr->resize_hpt = smc->resize_hpt_default;
2882         }
2883     }
2884 
2885     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2886 
2887     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2888         /*
2889          * User requested HPT resize, but this host can't supply it.  Bail out
2890          */
2891         error_report_err(resize_hpt_err);
2892         exit(1);
2893     }
2894     error_free(resize_hpt_err);
2895 
2896     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2897 
2898     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2899     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2900 
2901     /*
2902      * VSMT must be set in order to be able to compute VCPU ids, ie to
2903      * call spapr_max_server_number() or spapr_vcpu_id().
2904      */
2905     spapr_set_vsmt_mode(spapr, &error_fatal);
2906 
2907     /* Set up Interrupt Controller before we create the VCPUs */
2908     spapr_irq_init(spapr, &error_fatal);
2909 
2910     /* Set up containers for ibm,client-architecture-support negotiated options
2911      */
2912     spapr->ov5 = spapr_ovec_new();
2913     spapr->ov5_cas = spapr_ovec_new();
2914 
2915     if (smc->dr_lmb_enabled) {
2916         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2917         spapr_validate_node_memory(machine, &error_fatal);
2918     }
2919 
2920     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2921 
2922     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2923     if (!smc->pre_6_2_numa_affinity) {
2924         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2925     }
2926 
2927     /* advertise support for dedicated HP event source to guests */
2928     if (spapr->use_hotplug_event_source) {
2929         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2930     }
2931 
2932     /* advertise support for HPT resizing */
2933     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2934         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2935     }
2936 
2937     /* advertise support for ibm,dyamic-memory-v2 */
2938     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2939 
2940     /* advertise XIVE on POWER9 machines */
2941     if (spapr->irq->xive) {
2942         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2943     }
2944 
2945     /* init CPUs */
2946     spapr_init_cpus(spapr);
2947 
2948     /* Init numa_assoc_array */
2949     spapr_numa_associativity_init(spapr, machine);
2950 
2951     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2952         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2953                               spapr->max_compat_pvr)) {
2954         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2955         /* KVM and TCG always allow GTSE with radix... */
2956         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2957     }
2958     /* ... but not with hash (currently). */
2959 
2960     if (kvm_enabled()) {
2961         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2962         kvmppc_enable_logical_ci_hcalls();
2963         kvmppc_enable_set_mode_hcall();
2964 
2965         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2966         kvmppc_enable_clear_ref_mod_hcalls();
2967 
2968         /* Enable H_PAGE_INIT */
2969         kvmppc_enable_h_page_init();
2970     }
2971 
2972     /* map RAM */
2973     memory_region_add_subregion(sysmem, 0, machine->ram);
2974 
2975     /* initialize hotplug memory address space */
2976     if (machine->ram_size < machine->maxram_size) {
2977         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2978         hwaddr device_mem_base;
2979 
2980         /*
2981          * Limit the number of hotpluggable memory slots to half the number
2982          * slots that KVM supports, leaving the other half for PCI and other
2983          * devices. However ensure that number of slots doesn't drop below 32.
2984          */
2985         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2986                            SPAPR_MAX_RAM_SLOTS;
2987 
2988         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2989             max_memslots = SPAPR_MAX_RAM_SLOTS;
2990         }
2991         if (machine->ram_slots > max_memslots) {
2992             error_report("Specified number of memory slots %"
2993                          PRIu64" exceeds max supported %d",
2994                          machine->ram_slots, max_memslots);
2995             exit(1);
2996         }
2997 
2998         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
2999         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
3000     }
3001 
3002     if (smc->dr_lmb_enabled) {
3003         spapr_create_lmb_dr_connectors(spapr);
3004     }
3005 
3006     if (mc->nvdimm_supported) {
3007         spapr_create_nvdimm_dr_connectors(spapr);
3008     }
3009 
3010     /* Set up RTAS event infrastructure */
3011     spapr_events_init(spapr);
3012 
3013     /* Set up the RTC RTAS interfaces */
3014     spapr_rtc_create(spapr);
3015 
3016     /* Set up VIO bus */
3017     spapr->vio_bus = spapr_vio_bus_init();
3018 
3019     for (i = 0; serial_hd(i); i++) {
3020         spapr_vty_create(spapr->vio_bus, serial_hd(i));
3021     }
3022 
3023     /* We always have at least the nvram device on VIO */
3024     spapr_create_nvram(spapr);
3025 
3026     /*
3027      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3028      * connectors (described in root DT node's "ibm,drc-types" property)
3029      * are pre-initialized here. additional child connectors (such as
3030      * connectors for a PHBs PCI slots) are added as needed during their
3031      * parent's realization.
3032      */
3033     if (smc->dr_phb_enabled) {
3034         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3035             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3036         }
3037     }
3038 
3039     /* Set up PCI */
3040     spapr_pci_rtas_init();
3041 
3042     phb = spapr_create_default_phb();
3043 
3044     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3045         spapr_vlan_create(spapr->vio_bus, nd);
3046     }
3047 
3048     pci_init_nic_devices(phb->bus, NULL);
3049 
3050     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3051         spapr_vscsi_create(spapr->vio_bus);
3052     }
3053 
3054     /* Graphics */
3055     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3056     if (has_vga) {
3057         spapr->want_stdout_path = !machine->enable_graphics;
3058         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3059     } else {
3060         spapr->want_stdout_path = true;
3061     }
3062 
3063     if (machine->usb) {
3064         if (smc->use_ohci_by_default) {
3065             pci_create_simple(phb->bus, -1, "pci-ohci");
3066         } else {
3067             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3068         }
3069 
3070         if (has_vga) {
3071             USBBus *usb_bus;
3072 
3073             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3074                                                               &error_abort));
3075             usb_create_simple(usb_bus, "usb-kbd");
3076             usb_create_simple(usb_bus, "usb-mouse");
3077         }
3078     }
3079 
3080     if (kernel_filename) {
3081         uint64_t loaded_addr = 0;
3082 
3083         spapr->kernel_size = load_elf(kernel_filename, NULL,
3084                                       translate_kernel_address, spapr,
3085                                       NULL, &loaded_addr, NULL, NULL, 1,
3086                                       PPC_ELF_MACHINE, 0, 0);
3087         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3088             spapr->kernel_size = load_elf(kernel_filename, NULL,
3089                                           translate_kernel_address, spapr,
3090                                           NULL, &loaded_addr, NULL, NULL, 0,
3091                                           PPC_ELF_MACHINE, 0, 0);
3092             spapr->kernel_le = spapr->kernel_size > 0;
3093         }
3094         if (spapr->kernel_size < 0) {
3095             error_report("error loading %s: %s", kernel_filename,
3096                          load_elf_strerror(spapr->kernel_size));
3097             exit(1);
3098         }
3099 
3100         if (spapr->kernel_addr != loaded_addr) {
3101             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3102                         " to 0x%"PRIx64,
3103                         spapr->kernel_addr, loaded_addr);
3104             spapr->kernel_addr = loaded_addr;
3105         }
3106 
3107         /* load initrd */
3108         if (initrd_filename) {
3109             /* Try to locate the initrd in the gap between the kernel
3110              * and the firmware. Add a bit of space just in case
3111              */
3112             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3113                                   + 0x1ffff) & ~0xffff;
3114             spapr->initrd_size = load_image_targphys(initrd_filename,
3115                                                      spapr->initrd_base,
3116                                                      load_limit
3117                                                      - spapr->initrd_base);
3118             if (spapr->initrd_size < 0) {
3119                 error_report("could not load initial ram disk '%s'",
3120                              initrd_filename);
3121                 exit(1);
3122             }
3123         }
3124     }
3125 
3126     /* FIXME: Should register things through the MachineState's qdev
3127      * interface, this is a legacy from the sPAPREnvironment structure
3128      * which predated MachineState but had a similar function */
3129     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3130     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3131                          &savevm_htab_handlers, spapr);
3132 
3133     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3134 
3135     qemu_register_boot_set(spapr_boot_set, spapr);
3136 
3137     /*
3138      * Nothing needs to be done to resume a suspended guest because
3139      * suspending does not change the machine state, so no need for
3140      * a ->wakeup method.
3141      */
3142     qemu_register_wakeup_support();
3143 
3144     if (kvm_enabled()) {
3145         /* to stop and start vmclock */
3146         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3147                                          &spapr->tb);
3148 
3149         kvmppc_spapr_enable_inkernel_multitce();
3150     }
3151 
3152     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3153     if (spapr->vof) {
3154         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3155         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3156     }
3157 
3158     spapr_watchdog_init(spapr);
3159 }
3160 
3161 #define DEFAULT_KVM_TYPE "auto"
3162 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3163 {
3164     /*
3165      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3166      * accommodate the 'HV' and 'PV' formats that exists in the
3167      * wild. The 'auto' mode is being introduced already as
3168      * lower-case, thus we don't need to bother checking for
3169      * "AUTO".
3170      */
3171     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3172         return 0;
3173     }
3174 
3175     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3176         return 1;
3177     }
3178 
3179     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3180         return 2;
3181     }
3182 
3183     error_report("Unknown kvm-type specified '%s'", vm_type);
3184     return -1;
3185 }
3186 
3187 /*
3188  * Implementation of an interface to adjust firmware path
3189  * for the bootindex property handling.
3190  */
3191 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3192                                    DeviceState *dev)
3193 {
3194 #define CAST(type, obj, name) \
3195     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3196     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3197     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3198     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3199     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3200 
3201     if (d && bus) {
3202         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3203         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3204         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3205 
3206         if (spapr) {
3207             /*
3208              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3209              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3210              * 0x8000 | (target << 8) | (bus << 5) | lun
3211              * (see the "Logical unit addressing format" table in SAM5)
3212              */
3213             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3214             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3215                                    (uint64_t)id << 48);
3216         } else if (virtio) {
3217             /*
3218              * We use SRP luns of the form 01000000 | (target << 8) | lun
3219              * in the top 32 bits of the 64-bit LUN
3220              * Note: the quote above is from SLOF and it is wrong,
3221              * the actual binding is:
3222              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3223              */
3224             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3225             if (d->lun >= 256) {
3226                 /* Use the LUN "flat space addressing method" */
3227                 id |= 0x4000;
3228             }
3229             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3230                                    (uint64_t)id << 32);
3231         } else if (usb) {
3232             /*
3233              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3234              * in the top 32 bits of the 64-bit LUN
3235              */
3236             unsigned usb_port = atoi(usb->port->path);
3237             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3238             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3239                                    (uint64_t)id << 32);
3240         }
3241     }
3242 
3243     /*
3244      * SLOF probes the USB devices, and if it recognizes that the device is a
3245      * storage device, it changes its name to "storage" instead of "usb-host",
3246      * and additionally adds a child node for the SCSI LUN, so the correct
3247      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3248      */
3249     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3250         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3251         if (usb_device_is_scsi_storage(usbdev)) {
3252             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3253         }
3254     }
3255 
3256     if (phb) {
3257         /* Replace "pci" with "pci@800000020000000" */
3258         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3259     }
3260 
3261     if (vsc) {
3262         /* Same logic as virtio above */
3263         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3264         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3265     }
3266 
3267     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3268         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3269         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3270         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3271     }
3272 
3273     if (pcidev) {
3274         return spapr_pci_fw_dev_name(pcidev);
3275     }
3276 
3277     return NULL;
3278 }
3279 
3280 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     return g_strdup(spapr->kvm_type);
3285 }
3286 
3287 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3288 {
3289     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3290 
3291     g_free(spapr->kvm_type);
3292     spapr->kvm_type = g_strdup(value);
3293 }
3294 
3295 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 
3299     return spapr->use_hotplug_event_source;
3300 }
3301 
3302 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3303                                             Error **errp)
3304 {
3305     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3306 
3307     spapr->use_hotplug_event_source = value;
3308 }
3309 
3310 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3311 {
3312     return true;
3313 }
3314 
3315 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3316 {
3317     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3318 
3319     switch (spapr->resize_hpt) {
3320     case SPAPR_RESIZE_HPT_DEFAULT:
3321         return g_strdup("default");
3322     case SPAPR_RESIZE_HPT_DISABLED:
3323         return g_strdup("disabled");
3324     case SPAPR_RESIZE_HPT_ENABLED:
3325         return g_strdup("enabled");
3326     case SPAPR_RESIZE_HPT_REQUIRED:
3327         return g_strdup("required");
3328     }
3329     g_assert_not_reached();
3330 }
3331 
3332 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3333 {
3334     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3335 
3336     if (strcmp(value, "default") == 0) {
3337         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3338     } else if (strcmp(value, "disabled") == 0) {
3339         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3340     } else if (strcmp(value, "enabled") == 0) {
3341         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3342     } else if (strcmp(value, "required") == 0) {
3343         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3344     } else {
3345         error_setg(errp, "Bad value for \"resize-hpt\" property");
3346     }
3347 }
3348 
3349 static bool spapr_get_vof(Object *obj, Error **errp)
3350 {
3351     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3352 
3353     return spapr->vof != NULL;
3354 }
3355 
3356 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3357 {
3358     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3359 
3360     if (spapr->vof) {
3361         vof_cleanup(spapr->vof);
3362         g_free(spapr->vof);
3363         spapr->vof = NULL;
3364     }
3365     if (!value) {
3366         return;
3367     }
3368     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3369 }
3370 
3371 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3372 {
3373     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3374 
3375     if (spapr->irq == &spapr_irq_xics_legacy) {
3376         return g_strdup("legacy");
3377     } else if (spapr->irq == &spapr_irq_xics) {
3378         return g_strdup("xics");
3379     } else if (spapr->irq == &spapr_irq_xive) {
3380         return g_strdup("xive");
3381     } else if (spapr->irq == &spapr_irq_dual) {
3382         return g_strdup("dual");
3383     }
3384     g_assert_not_reached();
3385 }
3386 
3387 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3388 {
3389     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3390 
3391     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3392         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3393         return;
3394     }
3395 
3396     /* The legacy IRQ backend can not be set */
3397     if (strcmp(value, "xics") == 0) {
3398         spapr->irq = &spapr_irq_xics;
3399     } else if (strcmp(value, "xive") == 0) {
3400         spapr->irq = &spapr_irq_xive;
3401     } else if (strcmp(value, "dual") == 0) {
3402         spapr->irq = &spapr_irq_dual;
3403     } else {
3404         error_setg(errp, "Bad value for \"ic-mode\" property");
3405     }
3406 }
3407 
3408 static char *spapr_get_host_model(Object *obj, Error **errp)
3409 {
3410     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3411 
3412     return g_strdup(spapr->host_model);
3413 }
3414 
3415 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3416 {
3417     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3418 
3419     g_free(spapr->host_model);
3420     spapr->host_model = g_strdup(value);
3421 }
3422 
3423 static char *spapr_get_host_serial(Object *obj, Error **errp)
3424 {
3425     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3426 
3427     return g_strdup(spapr->host_serial);
3428 }
3429 
3430 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3431 {
3432     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3433 
3434     g_free(spapr->host_serial);
3435     spapr->host_serial = g_strdup(value);
3436 }
3437 
3438 static void spapr_instance_init(Object *obj)
3439 {
3440     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3441     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3442     MachineState *ms = MACHINE(spapr);
3443     MachineClass *mc = MACHINE_GET_CLASS(ms);
3444 
3445     /*
3446      * NVDIMM support went live in 5.1 without considering that, in
3447      * other archs, the user needs to enable NVDIMM support with the
3448      * 'nvdimm' machine option and the default behavior is NVDIMM
3449      * support disabled. It is too late to roll back to the standard
3450      * behavior without breaking 5.1 guests.
3451      */
3452     if (mc->nvdimm_supported) {
3453         ms->nvdimms_state->is_enabled = true;
3454     }
3455 
3456     spapr->htab_fd = -1;
3457     spapr->use_hotplug_event_source = true;
3458     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3459     object_property_add_str(obj, "kvm-type",
3460                             spapr_get_kvm_type, spapr_set_kvm_type);
3461     object_property_set_description(obj, "kvm-type",
3462                                     "Specifies the KVM virtualization mode (auto,"
3463                                     " hv, pr). Defaults to 'auto'. This mode will use"
3464                                     " any available KVM module loaded in the host,"
3465                                     " where kvm_hv takes precedence if both kvm_hv and"
3466                                     " kvm_pr are loaded.");
3467     object_property_add_bool(obj, "modern-hotplug-events",
3468                             spapr_get_modern_hotplug_events,
3469                             spapr_set_modern_hotplug_events);
3470     object_property_set_description(obj, "modern-hotplug-events",
3471                                     "Use dedicated hotplug event mechanism in"
3472                                     " place of standard EPOW events when possible"
3473                                     " (required for memory hot-unplug support)");
3474     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3475                             "Maximum permitted CPU compatibility mode");
3476 
3477     object_property_add_str(obj, "resize-hpt",
3478                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3479     object_property_set_description(obj, "resize-hpt",
3480                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3481     object_property_add_uint32_ptr(obj, "vsmt",
3482                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3483     object_property_set_description(obj, "vsmt",
3484                                     "Virtual SMT: KVM behaves as if this were"
3485                                     " the host's SMT mode");
3486 
3487     object_property_add_bool(obj, "vfio-no-msix-emulation",
3488                              spapr_get_msix_emulation, NULL);
3489 
3490     object_property_add_uint64_ptr(obj, "kernel-addr",
3491                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3492     object_property_set_description(obj, "kernel-addr",
3493                                     stringify(KERNEL_LOAD_ADDR)
3494                                     " for -kernel is the default");
3495     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3496 
3497     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3498     object_property_set_description(obj, "x-vof",
3499                                     "Enable Virtual Open Firmware (experimental)");
3500 
3501     /* The machine class defines the default interrupt controller mode */
3502     spapr->irq = smc->irq;
3503     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3504                             spapr_set_ic_mode);
3505     object_property_set_description(obj, "ic-mode",
3506                  "Specifies the interrupt controller mode (xics, xive, dual)");
3507 
3508     object_property_add_str(obj, "host-model",
3509         spapr_get_host_model, spapr_set_host_model);
3510     object_property_set_description(obj, "host-model",
3511         "Host model to advertise in guest device tree");
3512     object_property_add_str(obj, "host-serial",
3513         spapr_get_host_serial, spapr_set_host_serial);
3514     object_property_set_description(obj, "host-serial",
3515         "Host serial number to advertise in guest device tree");
3516 }
3517 
3518 static void spapr_machine_finalizefn(Object *obj)
3519 {
3520     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3521 
3522     g_free(spapr->kvm_type);
3523 }
3524 
3525 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3526 {
3527     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3528     PowerPCCPU *cpu = POWERPC_CPU(cs);
3529     CPUPPCState *env = &cpu->env;
3530 
3531     cpu_synchronize_state(cs);
3532     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3533     if (spapr->fwnmi_system_reset_addr != -1) {
3534         uint64_t rtas_addr, addr;
3535 
3536         /* get rtas addr from fdt */
3537         rtas_addr = spapr_get_rtas_addr();
3538         if (!rtas_addr) {
3539             qemu_system_guest_panicked(NULL);
3540             return;
3541         }
3542 
3543         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3544         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3545         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3546         env->gpr[3] = addr;
3547     }
3548     ppc_cpu_do_system_reset(cs);
3549     if (spapr->fwnmi_system_reset_addr != -1) {
3550         env->nip = spapr->fwnmi_system_reset_addr;
3551     }
3552 }
3553 
3554 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3555 {
3556     CPUState *cs;
3557 
3558     CPU_FOREACH(cs) {
3559         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3560     }
3561 }
3562 
3563 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3564                           void *fdt, int *fdt_start_offset, Error **errp)
3565 {
3566     uint64_t addr;
3567     uint32_t node;
3568 
3569     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3570     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3571                                     &error_abort);
3572     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3573                                              SPAPR_MEMORY_BLOCK_SIZE);
3574     return 0;
3575 }
3576 
3577 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3578                            bool dedicated_hp_event_source)
3579 {
3580     SpaprDrc *drc;
3581     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3582     int i;
3583     uint64_t addr = addr_start;
3584     bool hotplugged = spapr_drc_hotplugged(dev);
3585 
3586     for (i = 0; i < nr_lmbs; i++) {
3587         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3588                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3589         g_assert(drc);
3590 
3591         /*
3592          * memory_device_get_free_addr() provided a range of free addresses
3593          * that doesn't overlap with any existing mapping at pre-plug. The
3594          * corresponding LMB DRCs are thus assumed to be all attachable.
3595          */
3596         spapr_drc_attach(drc, dev);
3597         if (!hotplugged) {
3598             spapr_drc_reset(drc);
3599         }
3600         addr += SPAPR_MEMORY_BLOCK_SIZE;
3601     }
3602     /* send hotplug notification to the
3603      * guest only in case of hotplugged memory
3604      */
3605     if (hotplugged) {
3606         if (dedicated_hp_event_source) {
3607             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3608                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3609             g_assert(drc);
3610             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3611                                                    nr_lmbs,
3612                                                    spapr_drc_index(drc));
3613         } else {
3614             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3615                                            nr_lmbs);
3616         }
3617     }
3618 }
3619 
3620 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3621 {
3622     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3623     PCDIMMDevice *dimm = PC_DIMM(dev);
3624     uint64_t size, addr;
3625     int64_t slot;
3626     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3627 
3628     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3629 
3630     pc_dimm_plug(dimm, MACHINE(ms));
3631 
3632     if (!is_nvdimm) {
3633         addr = object_property_get_uint(OBJECT(dimm),
3634                                         PC_DIMM_ADDR_PROP, &error_abort);
3635         spapr_add_lmbs(dev, addr, size,
3636                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3637     } else {
3638         slot = object_property_get_int(OBJECT(dimm),
3639                                        PC_DIMM_SLOT_PROP, &error_abort);
3640         /* We should have valid slot number at this point */
3641         g_assert(slot >= 0);
3642         spapr_add_nvdimm(dev, slot);
3643     }
3644 }
3645 
3646 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3647                                   Error **errp)
3648 {
3649     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3650     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3651     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3652     PCDIMMDevice *dimm = PC_DIMM(dev);
3653     Error *local_err = NULL;
3654     uint64_t size;
3655     Object *memdev;
3656     hwaddr pagesize;
3657 
3658     if (!smc->dr_lmb_enabled) {
3659         error_setg(errp, "Memory hotplug not supported for this machine");
3660         return;
3661     }
3662 
3663     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3664     if (local_err) {
3665         error_propagate(errp, local_err);
3666         return;
3667     }
3668 
3669     if (is_nvdimm) {
3670         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3671             return;
3672         }
3673     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3674         error_setg(errp, "Hotplugged memory size must be a multiple of "
3675                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3676         return;
3677     }
3678 
3679     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3680                                       &error_abort);
3681     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3682     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3683         return;
3684     }
3685 
3686     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3687 }
3688 
3689 struct SpaprDimmState {
3690     PCDIMMDevice *dimm;
3691     uint32_t nr_lmbs;
3692     QTAILQ_ENTRY(SpaprDimmState) next;
3693 };
3694 
3695 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3696                                                        PCDIMMDevice *dimm)
3697 {
3698     SpaprDimmState *dimm_state = NULL;
3699 
3700     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3701         if (dimm_state->dimm == dimm) {
3702             break;
3703         }
3704     }
3705     return dimm_state;
3706 }
3707 
3708 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3709                                                       uint32_t nr_lmbs,
3710                                                       PCDIMMDevice *dimm)
3711 {
3712     SpaprDimmState *ds = NULL;
3713 
3714     /*
3715      * If this request is for a DIMM whose removal had failed earlier
3716      * (due to guest's refusal to remove the LMBs), we would have this
3717      * dimm already in the pending_dimm_unplugs list. In that
3718      * case don't add again.
3719      */
3720     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3721     if (!ds) {
3722         ds = g_new0(SpaprDimmState, 1);
3723         ds->nr_lmbs = nr_lmbs;
3724         ds->dimm = dimm;
3725         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3726     }
3727     return ds;
3728 }
3729 
3730 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3731                                               SpaprDimmState *dimm_state)
3732 {
3733     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3734     g_free(dimm_state);
3735 }
3736 
3737 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3738                                                         PCDIMMDevice *dimm)
3739 {
3740     SpaprDrc *drc;
3741     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3742                                                   &error_abort);
3743     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3744     uint32_t avail_lmbs = 0;
3745     uint64_t addr_start, addr;
3746     int i;
3747 
3748     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3749                                           &error_abort);
3750 
3751     addr = addr_start;
3752     for (i = 0; i < nr_lmbs; i++) {
3753         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3754                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3755         g_assert(drc);
3756         if (drc->dev) {
3757             avail_lmbs++;
3758         }
3759         addr += SPAPR_MEMORY_BLOCK_SIZE;
3760     }
3761 
3762     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3763 }
3764 
3765 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3766 {
3767     SpaprDimmState *ds;
3768     PCDIMMDevice *dimm;
3769     SpaprDrc *drc;
3770     uint32_t nr_lmbs;
3771     uint64_t size, addr_start, addr;
3772     g_autofree char *qapi_error = NULL;
3773     int i;
3774 
3775     if (!dev) {
3776         return;
3777     }
3778 
3779     dimm = PC_DIMM(dev);
3780     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3781 
3782     /*
3783      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3784      * unplug state, but one of its DRC is marked as unplug_requested.
3785      * This is bad and weird enough to g_assert() out.
3786      */
3787     g_assert(ds);
3788 
3789     spapr_pending_dimm_unplugs_remove(spapr, ds);
3790 
3791     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3792     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3793 
3794     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3795                                           &error_abort);
3796 
3797     addr = addr_start;
3798     for (i = 0; i < nr_lmbs; i++) {
3799         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3800                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3801         g_assert(drc);
3802 
3803         drc->unplug_requested = false;
3804         addr += SPAPR_MEMORY_BLOCK_SIZE;
3805     }
3806 
3807     /*
3808      * Tell QAPI that something happened and the memory
3809      * hotunplug wasn't successful. Keep sending
3810      * MEM_UNPLUG_ERROR even while sending
3811      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3812      * MEM_UNPLUG_ERROR is due.
3813      */
3814     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3815                                  "for device %s", dev->id);
3816 
3817     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3818 
3819     qapi_event_send_device_unplug_guest_error(dev->id,
3820                                               dev->canonical_path);
3821 }
3822 
3823 /* Callback to be called during DRC release. */
3824 void spapr_lmb_release(DeviceState *dev)
3825 {
3826     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3827     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3828     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3829 
3830     /* This information will get lost if a migration occurs
3831      * during the unplug process. In this case recover it. */
3832     if (ds == NULL) {
3833         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3834         g_assert(ds);
3835         /* The DRC being examined by the caller at least must be counted */
3836         g_assert(ds->nr_lmbs);
3837     }
3838 
3839     if (--ds->nr_lmbs) {
3840         return;
3841     }
3842 
3843     /*
3844      * Now that all the LMBs have been removed by the guest, call the
3845      * unplug handler chain. This can never fail.
3846      */
3847     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3848     object_unparent(OBJECT(dev));
3849 }
3850 
3851 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3852 {
3853     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3854     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3855 
3856     /* We really shouldn't get this far without anything to unplug */
3857     g_assert(ds);
3858 
3859     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3860     qdev_unrealize(dev);
3861     spapr_pending_dimm_unplugs_remove(spapr, ds);
3862 }
3863 
3864 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3865                                         DeviceState *dev, Error **errp)
3866 {
3867     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3868     PCDIMMDevice *dimm = PC_DIMM(dev);
3869     uint32_t nr_lmbs;
3870     uint64_t size, addr_start, addr;
3871     int i;
3872     SpaprDrc *drc;
3873 
3874     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3875         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3876         return;
3877     }
3878 
3879     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3880     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3881 
3882     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3883                                           &error_abort);
3884 
3885     /*
3886      * An existing pending dimm state for this DIMM means that there is an
3887      * unplug operation in progress, waiting for the spapr_lmb_release
3888      * callback to complete the job (BQL can't cover that far). In this case,
3889      * bail out to avoid detaching DRCs that were already released.
3890      */
3891     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3892         error_setg(errp, "Memory unplug already in progress for device %s",
3893                    dev->id);
3894         return;
3895     }
3896 
3897     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3898 
3899     addr = addr_start;
3900     for (i = 0; i < nr_lmbs; i++) {
3901         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3902                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3903         g_assert(drc);
3904 
3905         spapr_drc_unplug_request(drc);
3906         addr += SPAPR_MEMORY_BLOCK_SIZE;
3907     }
3908 
3909     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3910                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3911     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3912                                               nr_lmbs, spapr_drc_index(drc));
3913 }
3914 
3915 /* Callback to be called during DRC release. */
3916 void spapr_core_release(DeviceState *dev)
3917 {
3918     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3919 
3920     /* Call the unplug handler chain. This can never fail. */
3921     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3922     object_unparent(OBJECT(dev));
3923 }
3924 
3925 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3926 {
3927     MachineState *ms = MACHINE(hotplug_dev);
3928     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3929     CPUCore *cc = CPU_CORE(dev);
3930     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3931 
3932     if (smc->pre_2_10_has_unused_icps) {
3933         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3934         int i;
3935 
3936         for (i = 0; i < cc->nr_threads; i++) {
3937             CPUState *cs = CPU(sc->threads[i]);
3938 
3939             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3940         }
3941     }
3942 
3943     assert(core_slot);
3944     core_slot->cpu = NULL;
3945     qdev_unrealize(dev);
3946 }
3947 
3948 static
3949 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3950                                Error **errp)
3951 {
3952     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3953     int index;
3954     SpaprDrc *drc;
3955     CPUCore *cc = CPU_CORE(dev);
3956 
3957     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3958         error_setg(errp, "Unable to find CPU core with core-id: %d",
3959                    cc->core_id);
3960         return;
3961     }
3962     if (index == 0) {
3963         error_setg(errp, "Boot CPU core may not be unplugged");
3964         return;
3965     }
3966 
3967     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3968                           spapr_vcpu_id(spapr, cc->core_id));
3969     g_assert(drc);
3970 
3971     if (!spapr_drc_unplug_requested(drc)) {
3972         spapr_drc_unplug_request(drc);
3973     }
3974 
3975     /*
3976      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3977      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3978      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3979      * attempt (e.g. the kernel will refuse to remove the last online
3980      * CPU), we will never attempt it again because unplug_requested
3981      * will still be 'true' in that case.
3982      */
3983     spapr_hotplug_req_remove_by_index(drc);
3984 }
3985 
3986 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3987                            void *fdt, int *fdt_start_offset, Error **errp)
3988 {
3989     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3990     CPUState *cs = CPU(core->threads[0]);
3991     PowerPCCPU *cpu = POWERPC_CPU(cs);
3992     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3993     int id = spapr_get_vcpu_id(cpu);
3994     g_autofree char *nodename = NULL;
3995     int offset;
3996 
3997     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3998     offset = fdt_add_subnode(fdt, 0, nodename);
3999 
4000     spapr_dt_cpu(cs, fdt, offset, spapr);
4001 
4002     /*
4003      * spapr_dt_cpu() does not fill the 'name' property in the
4004      * CPU node. The function is called during boot process, before
4005      * and after CAS, and overwriting the 'name' property written
4006      * by SLOF is not allowed.
4007      *
4008      * Write it manually after spapr_dt_cpu(). This makes the hotplug
4009      * CPUs more compatible with the coldplugged ones, which have
4010      * the 'name' property. Linux Kernel also relies on this
4011      * property to identify CPU nodes.
4012      */
4013     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
4014 
4015     *fdt_start_offset = offset;
4016     return 0;
4017 }
4018 
4019 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4020 {
4021     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4022     MachineClass *mc = MACHINE_GET_CLASS(spapr);
4023     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4024     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
4025     CPUCore *cc = CPU_CORE(dev);
4026     CPUState *cs;
4027     SpaprDrc *drc;
4028     CPUArchId *core_slot;
4029     int index;
4030     bool hotplugged = spapr_drc_hotplugged(dev);
4031     int i;
4032 
4033     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4034     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4035 
4036     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4037                           spapr_vcpu_id(spapr, cc->core_id));
4038 
4039     g_assert(drc || !mc->has_hotpluggable_cpus);
4040 
4041     if (drc) {
4042         /*
4043          * spapr_core_pre_plug() already buys us this is a brand new
4044          * core being plugged into a free slot. Nothing should already
4045          * be attached to the corresponding DRC.
4046          */
4047         spapr_drc_attach(drc, dev);
4048 
4049         if (hotplugged) {
4050             /*
4051              * Send hotplug notification interrupt to the guest only
4052              * in case of hotplugged CPUs.
4053              */
4054             spapr_hotplug_req_add_by_index(drc);
4055         } else {
4056             spapr_drc_reset(drc);
4057         }
4058     }
4059 
4060     core_slot->cpu = OBJECT(dev);
4061 
4062     /*
4063      * Set compatibility mode to match the boot CPU, which was either set
4064      * by the machine reset code or by CAS. This really shouldn't fail at
4065      * this point.
4066      */
4067     if (hotplugged) {
4068         for (i = 0; i < cc->nr_threads; i++) {
4069             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4070                            &error_abort);
4071         }
4072     }
4073 
4074     if (smc->pre_2_10_has_unused_icps) {
4075         for (i = 0; i < cc->nr_threads; i++) {
4076             cs = CPU(core->threads[i]);
4077             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4078         }
4079     }
4080 }
4081 
4082 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4083                                 Error **errp)
4084 {
4085     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4086     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4087     CPUCore *cc = CPU_CORE(dev);
4088     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4089     const char *type = object_get_typename(OBJECT(dev));
4090     CPUArchId *core_slot;
4091     int index;
4092     unsigned int smp_threads = machine->smp.threads;
4093 
4094     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4095         error_setg(errp, "CPU hotplug not supported for this machine");
4096         return;
4097     }
4098 
4099     if (strcmp(base_core_type, type)) {
4100         error_setg(errp, "CPU core type should be %s", base_core_type);
4101         return;
4102     }
4103 
4104     if (cc->core_id % smp_threads) {
4105         error_setg(errp, "invalid core id %d", cc->core_id);
4106         return;
4107     }
4108 
4109     /*
4110      * In general we should have homogeneous threads-per-core, but old
4111      * (pre hotplug support) machine types allow the last core to have
4112      * reduced threads as a compatibility hack for when we allowed
4113      * total vcpus not a multiple of threads-per-core.
4114      */
4115     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4116         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4117                    smp_threads);
4118         return;
4119     }
4120 
4121     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4122     if (!core_slot) {
4123         error_setg(errp, "core id %d out of range", cc->core_id);
4124         return;
4125     }
4126 
4127     if (core_slot->cpu) {
4128         error_setg(errp, "core %d already populated", cc->core_id);
4129         return;
4130     }
4131 
4132     numa_cpu_pre_plug(core_slot, dev, errp);
4133 }
4134 
4135 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4136                           void *fdt, int *fdt_start_offset, Error **errp)
4137 {
4138     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4139     int intc_phandle;
4140 
4141     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4142     if (intc_phandle <= 0) {
4143         return -1;
4144     }
4145 
4146     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4147         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4148         return -1;
4149     }
4150 
4151     /* generally SLOF creates these, for hotplug it's up to QEMU */
4152     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4153 
4154     return 0;
4155 }
4156 
4157 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4158                                Error **errp)
4159 {
4160     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4161     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4162     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4163     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4164     SpaprDrc *drc;
4165 
4166     if (dev->hotplugged && !smc->dr_phb_enabled) {
4167         error_setg(errp, "PHB hotplug not supported for this machine");
4168         return false;
4169     }
4170 
4171     if (sphb->index == (uint32_t)-1) {
4172         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4173         return false;
4174     }
4175 
4176     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4177     if (drc && drc->dev) {
4178         error_setg(errp, "PHB %d already attached", sphb->index);
4179         return false;
4180     }
4181 
4182     /*
4183      * This will check that sphb->index doesn't exceed the maximum number of
4184      * PHBs for the current machine type.
4185      */
4186     return
4187         smc->phb_placement(spapr, sphb->index,
4188                            &sphb->buid, &sphb->io_win_addr,
4189                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4190                            windows_supported, sphb->dma_liobn,
4191                            errp);
4192 }
4193 
4194 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4195 {
4196     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4197     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4198     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4199     SpaprDrc *drc;
4200     bool hotplugged = spapr_drc_hotplugged(dev);
4201 
4202     if (!smc->dr_phb_enabled) {
4203         return;
4204     }
4205 
4206     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4207     /* hotplug hooks should check it's enabled before getting this far */
4208     assert(drc);
4209 
4210     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4211     spapr_drc_attach(drc, dev);
4212 
4213     if (hotplugged) {
4214         spapr_hotplug_req_add_by_index(drc);
4215     } else {
4216         spapr_drc_reset(drc);
4217     }
4218 }
4219 
4220 void spapr_phb_release(DeviceState *dev)
4221 {
4222     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4223 
4224     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4225     object_unparent(OBJECT(dev));
4226 }
4227 
4228 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4229 {
4230     qdev_unrealize(dev);
4231 }
4232 
4233 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4234                                      DeviceState *dev, Error **errp)
4235 {
4236     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4237     SpaprDrc *drc;
4238 
4239     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4240     assert(drc);
4241 
4242     if (!spapr_drc_unplug_requested(drc)) {
4243         spapr_drc_unplug_request(drc);
4244         spapr_hotplug_req_remove_by_index(drc);
4245     } else {
4246         error_setg(errp,
4247                    "PCI Host Bridge unplug already in progress for device %s",
4248                    dev->id);
4249     }
4250 }
4251 
4252 static
4253 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4254                               Error **errp)
4255 {
4256     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4257 
4258     if (spapr->tpm_proxy != NULL) {
4259         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4260         return false;
4261     }
4262 
4263     return true;
4264 }
4265 
4266 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4267 {
4268     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4269     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4270 
4271     /* Already checked in spapr_tpm_proxy_pre_plug() */
4272     g_assert(spapr->tpm_proxy == NULL);
4273 
4274     spapr->tpm_proxy = tpm_proxy;
4275 }
4276 
4277 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4278 {
4279     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4280 
4281     qdev_unrealize(dev);
4282     object_unparent(OBJECT(dev));
4283     spapr->tpm_proxy = NULL;
4284 }
4285 
4286 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4287                                       DeviceState *dev, Error **errp)
4288 {
4289     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4290         spapr_memory_plug(hotplug_dev, dev);
4291     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4292         spapr_core_plug(hotplug_dev, dev);
4293     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4294         spapr_phb_plug(hotplug_dev, dev);
4295     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4296         spapr_tpm_proxy_plug(hotplug_dev, dev);
4297     }
4298 }
4299 
4300 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4301                                         DeviceState *dev, Error **errp)
4302 {
4303     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4304         spapr_memory_unplug(hotplug_dev, dev);
4305     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4306         spapr_core_unplug(hotplug_dev, dev);
4307     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4308         spapr_phb_unplug(hotplug_dev, dev);
4309     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4310         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4311     }
4312 }
4313 
4314 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4315 {
4316     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4317         /*
4318          * CAS will process all pending unplug requests.
4319          *
4320          * HACK: a guest could theoretically have cleared all bits in OV5,
4321          * but none of the guests we care for do.
4322          */
4323         spapr_ovec_empty(spapr->ov5_cas);
4324 }
4325 
4326 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4327                                                 DeviceState *dev, Error **errp)
4328 {
4329     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4330     MachineClass *mc = MACHINE_GET_CLASS(sms);
4331     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4332 
4333     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4334         if (spapr_memory_hot_unplug_supported(sms)) {
4335             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4336         } else {
4337             error_setg(errp, "Memory hot unplug not supported for this guest");
4338         }
4339     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4340         if (!mc->has_hotpluggable_cpus) {
4341             error_setg(errp, "CPU hot unplug not supported on this machine");
4342             return;
4343         }
4344         spapr_core_unplug_request(hotplug_dev, dev, errp);
4345     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4346         if (!smc->dr_phb_enabled) {
4347             error_setg(errp, "PHB hot unplug not supported on this machine");
4348             return;
4349         }
4350         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4351     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4352         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4353     }
4354 }
4355 
4356 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4357                                           DeviceState *dev, Error **errp)
4358 {
4359     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4360         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4361     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4362         spapr_core_pre_plug(hotplug_dev, dev, errp);
4363     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4364         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4365     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4366         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4367     }
4368 }
4369 
4370 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4371                                                  DeviceState *dev)
4372 {
4373     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4374         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4375         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4376         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4377         return HOTPLUG_HANDLER(machine);
4378     }
4379     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4380         PCIDevice *pcidev = PCI_DEVICE(dev);
4381         PCIBus *root = pci_device_root_bus(pcidev);
4382         SpaprPhbState *phb =
4383             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4384                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4385 
4386         if (phb) {
4387             return HOTPLUG_HANDLER(phb);
4388         }
4389     }
4390     return NULL;
4391 }
4392 
4393 static CpuInstanceProperties
4394 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4395 {
4396     CPUArchId *core_slot;
4397     MachineClass *mc = MACHINE_GET_CLASS(machine);
4398 
4399     /* make sure possible_cpu are initialized */
4400     mc->possible_cpu_arch_ids(machine);
4401     /* get CPU core slot containing thread that matches cpu_index */
4402     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4403     assert(core_slot);
4404     return core_slot->props;
4405 }
4406 
4407 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4408 {
4409     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4410 }
4411 
4412 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4413 {
4414     int i;
4415     unsigned int smp_threads = machine->smp.threads;
4416     unsigned int smp_cpus = machine->smp.cpus;
4417     const char *core_type;
4418     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4419     MachineClass *mc = MACHINE_GET_CLASS(machine);
4420 
4421     if (!mc->has_hotpluggable_cpus) {
4422         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4423     }
4424     if (machine->possible_cpus) {
4425         assert(machine->possible_cpus->len == spapr_max_cores);
4426         return machine->possible_cpus;
4427     }
4428 
4429     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4430     if (!core_type) {
4431         error_report("Unable to find sPAPR CPU Core definition");
4432         exit(1);
4433     }
4434 
4435     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4436                              sizeof(CPUArchId) * spapr_max_cores);
4437     machine->possible_cpus->len = spapr_max_cores;
4438     for (i = 0; i < machine->possible_cpus->len; i++) {
4439         int core_id = i * smp_threads;
4440 
4441         machine->possible_cpus->cpus[i].type = core_type;
4442         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4443         machine->possible_cpus->cpus[i].arch_id = core_id;
4444         machine->possible_cpus->cpus[i].props.has_core_id = true;
4445         machine->possible_cpus->cpus[i].props.core_id = core_id;
4446     }
4447     return machine->possible_cpus;
4448 }
4449 
4450 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4451                                 uint64_t *buid, hwaddr *pio,
4452                                 hwaddr *mmio32, hwaddr *mmio64,
4453                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4454 {
4455     /*
4456      * New-style PHB window placement.
4457      *
4458      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4459      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4460      * windows.
4461      *
4462      * Some guest kernels can't work with MMIO windows above 1<<46
4463      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4464      *
4465      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4466      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4467      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4468      * 1TiB 64-bit MMIO windows for each PHB.
4469      */
4470     const uint64_t base_buid = 0x800000020000000ULL;
4471     int i;
4472 
4473     /* Sanity check natural alignments */
4474     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4475     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4476     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4477     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4478     /* Sanity check bounds */
4479     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4480                       SPAPR_PCI_MEM32_WIN_SIZE);
4481     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4482                       SPAPR_PCI_MEM64_WIN_SIZE);
4483 
4484     if (index >= SPAPR_MAX_PHBS) {
4485         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4486                    SPAPR_MAX_PHBS - 1);
4487         return false;
4488     }
4489 
4490     *buid = base_buid + index;
4491     for (i = 0; i < n_dma; ++i) {
4492         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4493     }
4494 
4495     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4496     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4497     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4498     return true;
4499 }
4500 
4501 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4502 {
4503     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4504 
4505     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4506 }
4507 
4508 static void spapr_ics_resend(XICSFabric *dev)
4509 {
4510     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4511 
4512     ics_resend(spapr->ics);
4513 }
4514 
4515 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4516 {
4517     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4518 
4519     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4520 }
4521 
4522 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4523                                  Monitor *mon)
4524 {
4525     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4526 
4527     spapr_irq_print_info(spapr, mon);
4528     monitor_printf(mon, "irqchip: %s\n",
4529                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4530 }
4531 
4532 /*
4533  * This is a XIVE only operation
4534  */
4535 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4536                            uint8_t nvt_blk, uint32_t nvt_idx,
4537                            bool cam_ignore, uint8_t priority,
4538                            uint32_t logic_serv, XiveTCTXMatch *match)
4539 {
4540     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4541     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4542     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4543     int count;
4544 
4545     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4546                            priority, logic_serv, match);
4547     if (count < 0) {
4548         return count;
4549     }
4550 
4551     /*
4552      * When we implement the save and restore of the thread interrupt
4553      * contexts in the enter/exit CPU handlers of the machine and the
4554      * escalations in QEMU, we should be able to handle non dispatched
4555      * vCPUs.
4556      *
4557      * Until this is done, the sPAPR machine should find at least one
4558      * matching context always.
4559      */
4560     if (count == 0) {
4561         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4562                       nvt_blk, nvt_idx);
4563     }
4564 
4565     return count;
4566 }
4567 
4568 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4569 {
4570     return cpu->vcpu_id;
4571 }
4572 
4573 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4574 {
4575     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4576     MachineState *ms = MACHINE(spapr);
4577     int vcpu_id;
4578 
4579     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4580 
4581     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4582         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4583         error_append_hint(errp, "Adjust the number of cpus to %d "
4584                           "or try to raise the number of threads per core\n",
4585                           vcpu_id * ms->smp.threads / spapr->vsmt);
4586         return false;
4587     }
4588 
4589     cpu->vcpu_id = vcpu_id;
4590     return true;
4591 }
4592 
4593 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4594 {
4595     CPUState *cs;
4596 
4597     CPU_FOREACH(cs) {
4598         PowerPCCPU *cpu = POWERPC_CPU(cs);
4599 
4600         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4601             return cpu;
4602         }
4603     }
4604 
4605     return NULL;
4606 }
4607 
4608 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4609 {
4610     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4611 
4612     return spapr_cpu->in_nested;
4613 }
4614 
4615 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4616 {
4617     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4618 
4619     /* These are only called by TCG, KVM maintains dispatch state */
4620 
4621     spapr_cpu->prod = false;
4622     if (spapr_cpu->vpa_addr) {
4623         CPUState *cs = CPU(cpu);
4624         uint32_t dispatch;
4625 
4626         dispatch = ldl_be_phys(cs->as,
4627                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4628         dispatch++;
4629         if ((dispatch & 1) != 0) {
4630             qemu_log_mask(LOG_GUEST_ERROR,
4631                           "VPA: incorrect dispatch counter value for "
4632                           "dispatched partition %u, correcting.\n", dispatch);
4633             dispatch++;
4634         }
4635         stl_be_phys(cs->as,
4636                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4637     }
4638 }
4639 
4640 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4641 {
4642     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4643 
4644     if (spapr_cpu->vpa_addr) {
4645         CPUState *cs = CPU(cpu);
4646         uint32_t dispatch;
4647 
4648         dispatch = ldl_be_phys(cs->as,
4649                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4650         dispatch++;
4651         if ((dispatch & 1) != 1) {
4652             qemu_log_mask(LOG_GUEST_ERROR,
4653                           "VPA: incorrect dispatch counter value for "
4654                           "preempted partition %u, correcting.\n", dispatch);
4655             dispatch++;
4656         }
4657         stl_be_phys(cs->as,
4658                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4659     }
4660 }
4661 
4662 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4663 {
4664     MachineClass *mc = MACHINE_CLASS(oc);
4665     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4666     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4667     NMIClass *nc = NMI_CLASS(oc);
4668     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4669     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4670     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4671     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4672     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4673     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4674 
4675     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4676     mc->ignore_boot_device_suffixes = true;
4677 
4678     /*
4679      * We set up the default / latest behaviour here.  The class_init
4680      * functions for the specific versioned machine types can override
4681      * these details for backwards compatibility
4682      */
4683     mc->init = spapr_machine_init;
4684     mc->reset = spapr_machine_reset;
4685     mc->block_default_type = IF_SCSI;
4686 
4687     /*
4688      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4689      * In TCG the limit is restricted by the range of CPU IPIs available.
4690      */
4691     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4692 
4693     mc->no_parallel = 1;
4694     mc->default_boot_order = "";
4695     mc->default_ram_size = 512 * MiB;
4696     mc->default_ram_id = "ppc_spapr.ram";
4697     mc->default_display = "std";
4698     mc->kvm_type = spapr_kvm_type;
4699     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4700     mc->pci_allow_0_address = true;
4701     assert(!mc->get_hotplug_handler);
4702     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4703     hc->pre_plug = spapr_machine_device_pre_plug;
4704     hc->plug = spapr_machine_device_plug;
4705     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4706     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4707     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4708     hc->unplug_request = spapr_machine_device_unplug_request;
4709     hc->unplug = spapr_machine_device_unplug;
4710 
4711     smc->dr_lmb_enabled = true;
4712     smc->update_dt_enabled = true;
4713     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4714     mc->has_hotpluggable_cpus = true;
4715     mc->nvdimm_supported = true;
4716     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4717     fwc->get_dev_path = spapr_get_fw_dev_path;
4718     nc->nmi_monitor_handler = spapr_nmi;
4719     smc->phb_placement = spapr_phb_placement;
4720     vhc->cpu_in_nested = spapr_cpu_in_nested;
4721     vhc->deliver_hv_excp = spapr_exit_nested;
4722     vhc->hypercall = emulate_spapr_hypercall;
4723     vhc->hpt_mask = spapr_hpt_mask;
4724     vhc->map_hptes = spapr_map_hptes;
4725     vhc->unmap_hptes = spapr_unmap_hptes;
4726     vhc->hpte_set_c = spapr_hpte_set_c;
4727     vhc->hpte_set_r = spapr_hpte_set_r;
4728     vhc->get_pate = spapr_get_pate;
4729     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4730     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4731     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4732     xic->ics_get = spapr_ics_get;
4733     xic->ics_resend = spapr_ics_resend;
4734     xic->icp_get = spapr_icp_get;
4735     ispc->print_info = spapr_pic_print_info;
4736     /* Force NUMA node memory size to be a multiple of
4737      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4738      * in which LMBs are represented and hot-added
4739      */
4740     mc->numa_mem_align_shift = 28;
4741     mc->auto_enable_numa = true;
4742 
4743     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4744     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4745     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4746     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4747     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4748     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4749     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4750     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4751     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4752     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4753     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4754     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4755 
4756     /*
4757      * This cap specifies whether the AIL 3 mode for
4758      * H_SET_RESOURCE is supported. The default is modified
4759      * by default_caps_with_cpu().
4760      */
4761     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4762     spapr_caps_add_properties(smc);
4763     smc->irq = &spapr_irq_dual;
4764     smc->dr_phb_enabled = true;
4765     smc->linux_pci_probe = true;
4766     smc->smp_threads_vsmt = true;
4767     smc->nr_xirqs = SPAPR_NR_XIRQS;
4768     xfc->match_nvt = spapr_match_nvt;
4769     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4770     vmc->quiesce = spapr_vof_quiesce;
4771     vmc->setprop = spapr_vof_setprop;
4772 }
4773 
4774 static const TypeInfo spapr_machine_info = {
4775     .name          = TYPE_SPAPR_MACHINE,
4776     .parent        = TYPE_MACHINE,
4777     .abstract      = true,
4778     .instance_size = sizeof(SpaprMachineState),
4779     .instance_init = spapr_instance_init,
4780     .instance_finalize = spapr_machine_finalizefn,
4781     .class_size    = sizeof(SpaprMachineClass),
4782     .class_init    = spapr_machine_class_init,
4783     .interfaces = (InterfaceInfo[]) {
4784         { TYPE_FW_PATH_PROVIDER },
4785         { TYPE_NMI },
4786         { TYPE_HOTPLUG_HANDLER },
4787         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4788         { TYPE_XICS_FABRIC },
4789         { TYPE_INTERRUPT_STATS_PROVIDER },
4790         { TYPE_XIVE_FABRIC },
4791         { TYPE_VOF_MACHINE_IF },
4792         { }
4793     },
4794 };
4795 
4796 static void spapr_machine_latest_class_options(MachineClass *mc)
4797 {
4798     mc->alias = "pseries";
4799     mc->is_default = true;
4800 }
4801 
4802 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4803     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4804                                                     void *data)      \
4805     {                                                                \
4806         MachineClass *mc = MACHINE_CLASS(oc);                        \
4807         spapr_machine_##suffix##_class_options(mc);                  \
4808         if (latest) {                                                \
4809             spapr_machine_latest_class_options(mc);                  \
4810         }                                                            \
4811     }                                                                \
4812     static const TypeInfo spapr_machine_##suffix##_info = {          \
4813         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4814         .parent = TYPE_SPAPR_MACHINE,                                \
4815         .class_init = spapr_machine_##suffix##_class_init,           \
4816     };                                                               \
4817     static void spapr_machine_register_##suffix(void)                \
4818     {                                                                \
4819         type_register(&spapr_machine_##suffix##_info);               \
4820     }                                                                \
4821     type_init(spapr_machine_register_##suffix)
4822 
4823 /*
4824  * pseries-9.0
4825  */
4826 static void spapr_machine_9_0_class_options(MachineClass *mc)
4827 {
4828     /* Defaults for the latest behaviour inherited from the base class */
4829 }
4830 
4831 DEFINE_SPAPR_MACHINE(9_0, "9.0", true);
4832 
4833 /*
4834  * pseries-8.2
4835  */
4836 static void spapr_machine_8_2_class_options(MachineClass *mc)
4837 {
4838     spapr_machine_9_0_class_options(mc);
4839     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4840 }
4841 
4842 DEFINE_SPAPR_MACHINE(8_2, "8.2", false);
4843 
4844 /*
4845  * pseries-8.1
4846  */
4847 static void spapr_machine_8_1_class_options(MachineClass *mc)
4848 {
4849     spapr_machine_8_2_class_options(mc);
4850     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4851 }
4852 
4853 DEFINE_SPAPR_MACHINE(8_1, "8.1", false);
4854 
4855 /*
4856  * pseries-8.0
4857  */
4858 static void spapr_machine_8_0_class_options(MachineClass *mc)
4859 {
4860     spapr_machine_8_1_class_options(mc);
4861     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4862 }
4863 
4864 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4865 
4866 /*
4867  * pseries-7.2
4868  */
4869 static void spapr_machine_7_2_class_options(MachineClass *mc)
4870 {
4871     spapr_machine_8_0_class_options(mc);
4872     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4873 }
4874 
4875 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4876 
4877 /*
4878  * pseries-7.1
4879  */
4880 static void spapr_machine_7_1_class_options(MachineClass *mc)
4881 {
4882     spapr_machine_7_2_class_options(mc);
4883     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4884 }
4885 
4886 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4887 
4888 /*
4889  * pseries-7.0
4890  */
4891 static void spapr_machine_7_0_class_options(MachineClass *mc)
4892 {
4893     spapr_machine_7_1_class_options(mc);
4894     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4895 }
4896 
4897 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4898 
4899 /*
4900  * pseries-6.2
4901  */
4902 static void spapr_machine_6_2_class_options(MachineClass *mc)
4903 {
4904     spapr_machine_7_0_class_options(mc);
4905     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4906 }
4907 
4908 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4909 
4910 /*
4911  * pseries-6.1
4912  */
4913 static void spapr_machine_6_1_class_options(MachineClass *mc)
4914 {
4915     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4916 
4917     spapr_machine_6_2_class_options(mc);
4918     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4919     smc->pre_6_2_numa_affinity = true;
4920     mc->smp_props.prefer_sockets = true;
4921 }
4922 
4923 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4924 
4925 /*
4926  * pseries-6.0
4927  */
4928 static void spapr_machine_6_0_class_options(MachineClass *mc)
4929 {
4930     spapr_machine_6_1_class_options(mc);
4931     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4932 }
4933 
4934 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4935 
4936 /*
4937  * pseries-5.2
4938  */
4939 static void spapr_machine_5_2_class_options(MachineClass *mc)
4940 {
4941     spapr_machine_6_0_class_options(mc);
4942     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4943 }
4944 
4945 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4946 
4947 /*
4948  * pseries-5.1
4949  */
4950 static void spapr_machine_5_1_class_options(MachineClass *mc)
4951 {
4952     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4953 
4954     spapr_machine_5_2_class_options(mc);
4955     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4956     smc->pre_5_2_numa_associativity = true;
4957 }
4958 
4959 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4960 
4961 /*
4962  * pseries-5.0
4963  */
4964 static void spapr_machine_5_0_class_options(MachineClass *mc)
4965 {
4966     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4967     static GlobalProperty compat[] = {
4968         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4969     };
4970 
4971     spapr_machine_5_1_class_options(mc);
4972     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4973     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4974     mc->numa_mem_supported = true;
4975     smc->pre_5_1_assoc_refpoints = true;
4976 }
4977 
4978 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4979 
4980 /*
4981  * pseries-4.2
4982  */
4983 static void spapr_machine_4_2_class_options(MachineClass *mc)
4984 {
4985     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4986 
4987     spapr_machine_5_0_class_options(mc);
4988     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4989     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4990     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4991     smc->rma_limit = 16 * GiB;
4992     mc->nvdimm_supported = false;
4993 }
4994 
4995 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4996 
4997 /*
4998  * pseries-4.1
4999  */
5000 static void spapr_machine_4_1_class_options(MachineClass *mc)
5001 {
5002     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5003     static GlobalProperty compat[] = {
5004         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
5005         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
5006     };
5007 
5008     spapr_machine_4_2_class_options(mc);
5009     smc->linux_pci_probe = false;
5010     smc->smp_threads_vsmt = false;
5011     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
5012     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5013 }
5014 
5015 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
5016 
5017 /*
5018  * pseries-4.0
5019  */
5020 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
5021                               uint64_t *buid, hwaddr *pio,
5022                               hwaddr *mmio32, hwaddr *mmio64,
5023                               unsigned n_dma, uint32_t *liobns, Error **errp)
5024 {
5025     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
5026                              liobns, errp)) {
5027         return false;
5028     }
5029     return true;
5030 }
5031 static void spapr_machine_4_0_class_options(MachineClass *mc)
5032 {
5033     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5034 
5035     spapr_machine_4_1_class_options(mc);
5036     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5037     smc->phb_placement = phb_placement_4_0;
5038     smc->irq = &spapr_irq_xics;
5039     smc->pre_4_1_migration = true;
5040 }
5041 
5042 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
5043 
5044 /*
5045  * pseries-3.1
5046  */
5047 static void spapr_machine_3_1_class_options(MachineClass *mc)
5048 {
5049     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5050 
5051     spapr_machine_4_0_class_options(mc);
5052     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5053 
5054     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5055     smc->update_dt_enabled = false;
5056     smc->dr_phb_enabled = false;
5057     smc->broken_host_serial_model = true;
5058     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5059     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5060     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5061     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5062 }
5063 
5064 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
5065 
5066 /*
5067  * pseries-3.0
5068  */
5069 
5070 static void spapr_machine_3_0_class_options(MachineClass *mc)
5071 {
5072     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5073 
5074     spapr_machine_3_1_class_options(mc);
5075     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5076 
5077     smc->legacy_irq_allocation = true;
5078     smc->nr_xirqs = 0x400;
5079     smc->irq = &spapr_irq_xics_legacy;
5080 }
5081 
5082 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
5083 
5084 /*
5085  * pseries-2.12
5086  */
5087 static void spapr_machine_2_12_class_options(MachineClass *mc)
5088 {
5089     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5090     static GlobalProperty compat[] = {
5091         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5092         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5093     };
5094 
5095     spapr_machine_3_0_class_options(mc);
5096     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5097     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5098 
5099     /* We depend on kvm_enabled() to choose a default value for the
5100      * hpt-max-page-size capability. Of course we can't do it here
5101      * because this is too early and the HW accelerator isn't initialized
5102      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5103      */
5104     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5105 }
5106 
5107 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5108 
5109 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5110 {
5111     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5112 
5113     spapr_machine_2_12_class_options(mc);
5114     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5115     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5116     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5117 }
5118 
5119 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5120 
5121 /*
5122  * pseries-2.11
5123  */
5124 
5125 static void spapr_machine_2_11_class_options(MachineClass *mc)
5126 {
5127     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5128 
5129     spapr_machine_2_12_class_options(mc);
5130     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5131     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5132     mc->deprecation_reason = "old and not maintained - use a 2.12+ version";
5133 }
5134 
5135 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5136 
5137 /*
5138  * pseries-2.10
5139  */
5140 
5141 static void spapr_machine_2_10_class_options(MachineClass *mc)
5142 {
5143     spapr_machine_2_11_class_options(mc);
5144     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5145 }
5146 
5147 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5148 
5149 /*
5150  * pseries-2.9
5151  */
5152 
5153 static void spapr_machine_2_9_class_options(MachineClass *mc)
5154 {
5155     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5156     static GlobalProperty compat[] = {
5157         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5158     };
5159 
5160     spapr_machine_2_10_class_options(mc);
5161     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5162     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5163     smc->pre_2_10_has_unused_icps = true;
5164     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5165 }
5166 
5167 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5168 
5169 /*
5170  * pseries-2.8
5171  */
5172 
5173 static void spapr_machine_2_8_class_options(MachineClass *mc)
5174 {
5175     static GlobalProperty compat[] = {
5176         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5177     };
5178 
5179     spapr_machine_2_9_class_options(mc);
5180     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5181     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5182     mc->numa_mem_align_shift = 23;
5183 }
5184 
5185 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5186 
5187 /*
5188  * pseries-2.7
5189  */
5190 
5191 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5192                               uint64_t *buid, hwaddr *pio,
5193                               hwaddr *mmio32, hwaddr *mmio64,
5194                               unsigned n_dma, uint32_t *liobns, Error **errp)
5195 {
5196     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5197     const uint64_t base_buid = 0x800000020000000ULL;
5198     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5199     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5200     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5201     const uint32_t max_index = 255;
5202     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5203 
5204     uint64_t ram_top = MACHINE(spapr)->ram_size;
5205     hwaddr phb0_base, phb_base;
5206     int i;
5207 
5208     /* Do we have device memory? */
5209     if (MACHINE(spapr)->device_memory) {
5210         /* Can't just use maxram_size, because there may be an
5211          * alignment gap between normal and device memory regions
5212          */
5213         ram_top = MACHINE(spapr)->device_memory->base +
5214             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5215     }
5216 
5217     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5218 
5219     if (index > max_index) {
5220         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5221                    max_index);
5222         return false;
5223     }
5224 
5225     *buid = base_buid + index;
5226     for (i = 0; i < n_dma; ++i) {
5227         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5228     }
5229 
5230     phb_base = phb0_base + index * phb_spacing;
5231     *pio = phb_base + pio_offset;
5232     *mmio32 = phb_base + mmio_offset;
5233     /*
5234      * We don't set the 64-bit MMIO window, relying on the PHB's
5235      * fallback behaviour of automatically splitting a large "32-bit"
5236      * window into contiguous 32-bit and 64-bit windows
5237      */
5238 
5239     return true;
5240 }
5241 
5242 static void spapr_machine_2_7_class_options(MachineClass *mc)
5243 {
5244     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5245     static GlobalProperty compat[] = {
5246         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5247         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5248         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5249         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5250     };
5251 
5252     spapr_machine_2_8_class_options(mc);
5253     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5254     mc->default_machine_opts = "modern-hotplug-events=off";
5255     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5256     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5257     smc->phb_placement = phb_placement_2_7;
5258 }
5259 
5260 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5261 
5262 /*
5263  * pseries-2.6
5264  */
5265 
5266 static void spapr_machine_2_6_class_options(MachineClass *mc)
5267 {
5268     static GlobalProperty compat[] = {
5269         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5270     };
5271 
5272     spapr_machine_2_7_class_options(mc);
5273     mc->has_hotpluggable_cpus = false;
5274     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5275     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5276 }
5277 
5278 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5279 
5280 /*
5281  * pseries-2.5
5282  */
5283 
5284 static void spapr_machine_2_5_class_options(MachineClass *mc)
5285 {
5286     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5287     static GlobalProperty compat[] = {
5288         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5289     };
5290 
5291     spapr_machine_2_6_class_options(mc);
5292     smc->use_ohci_by_default = true;
5293     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5294     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5295 }
5296 
5297 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5298 
5299 /*
5300  * pseries-2.4
5301  */
5302 
5303 static void spapr_machine_2_4_class_options(MachineClass *mc)
5304 {
5305     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5306 
5307     spapr_machine_2_5_class_options(mc);
5308     smc->dr_lmb_enabled = false;
5309     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5310 }
5311 
5312 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5313 
5314 /*
5315  * pseries-2.3
5316  */
5317 
5318 static void spapr_machine_2_3_class_options(MachineClass *mc)
5319 {
5320     static GlobalProperty compat[] = {
5321         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5322     };
5323     spapr_machine_2_4_class_options(mc);
5324     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5325     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5326 }
5327 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5328 
5329 /*
5330  * pseries-2.2
5331  */
5332 
5333 static void spapr_machine_2_2_class_options(MachineClass *mc)
5334 {
5335     static GlobalProperty compat[] = {
5336         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5337     };
5338 
5339     spapr_machine_2_3_class_options(mc);
5340     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5341     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5342     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5343 }
5344 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5345 
5346 /*
5347  * pseries-2.1
5348  */
5349 
5350 static void spapr_machine_2_1_class_options(MachineClass *mc)
5351 {
5352     spapr_machine_2_2_class_options(mc);
5353     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5354 }
5355 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5356 
5357 static void spapr_machine_register_types(void)
5358 {
5359     type_register_static(&spapr_machine_info);
5360 }
5361 
5362 type_init(spapr_machine_register_types)
5363