xref: /openbmc/qemu/hw/ppc/spapr.c (revision 9940412ae4ae2d4971959c13311ba8d822974452)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu/datadir.h"
29 #include "qemu/memalign.h"
30 #include "qemu/guest-random.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-events-machine.h"
33 #include "qapi/qapi-events-qdev.h"
34 #include "qapi/visitor.h"
35 #include "sysemu/sysemu.h"
36 #include "sysemu/hostmem.h"
37 #include "sysemu/numa.h"
38 #include "sysemu/qtest.h"
39 #include "sysemu/reset.h"
40 #include "sysemu/runstate.h"
41 #include "qemu/log.h"
42 #include "hw/fw-path-provider.h"
43 #include "elf.h"
44 #include "net/net.h"
45 #include "sysemu/device_tree.h"
46 #include "sysemu/cpus.h"
47 #include "sysemu/hw_accel.h"
48 #include "kvm_ppc.h"
49 #include "migration/misc.h"
50 #include "migration/qemu-file-types.h"
51 #include "migration/global_state.h"
52 #include "migration/register.h"
53 #include "migration/blocker.h"
54 #include "mmu-hash64.h"
55 #include "mmu-book3s-v3.h"
56 #include "cpu-models.h"
57 #include "hw/core/cpu.h"
58 
59 #include "hw/ppc/ppc.h"
60 #include "hw/loader.h"
61 
62 #include "hw/ppc/fdt.h"
63 #include "hw/ppc/spapr.h"
64 #include "hw/ppc/spapr_nested.h"
65 #include "hw/ppc/spapr_vio.h"
66 #include "hw/ppc/vof.h"
67 #include "hw/qdev-properties.h"
68 #include "hw/pci-host/spapr.h"
69 #include "hw/pci/msi.h"
70 
71 #include "hw/pci/pci.h"
72 #include "hw/scsi/scsi.h"
73 #include "hw/virtio/virtio-scsi.h"
74 #include "hw/virtio/vhost-scsi-common.h"
75 
76 #include "exec/ram_addr.h"
77 #include "hw/usb.h"
78 #include "qemu/config-file.h"
79 #include "qemu/error-report.h"
80 #include "trace.h"
81 #include "hw/nmi.h"
82 #include "hw/intc/intc.h"
83 
84 #include "hw/ppc/spapr_cpu_core.h"
85 #include "hw/mem/memory-device.h"
86 #include "hw/ppc/spapr_tpm_proxy.h"
87 #include "hw/ppc/spapr_nvdimm.h"
88 #include "hw/ppc/spapr_numa.h"
89 #include "hw/ppc/pef.h"
90 
91 #include "monitor/monitor.h"
92 
93 #include <libfdt.h>
94 
95 /* SLOF memory layout:
96  *
97  * SLOF raw image loaded at 0, copies its romfs right below the flat
98  * device-tree, then position SLOF itself 31M below that
99  *
100  * So we set FW_OVERHEAD to 40MB which should account for all of that
101  * and more
102  *
103  * We load our kernel at 4M, leaving space for SLOF initial image
104  */
105 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
106 #define FW_MAX_SIZE             0x400000
107 #define FW_FILE_NAME            "slof.bin"
108 #define FW_FILE_NAME_VOF        "vof.bin"
109 #define FW_OVERHEAD             0x2800000
110 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
111 
112 #define MIN_RMA_SLOF            (128 * MiB)
113 
114 #define PHANDLE_INTC            0x00001111
115 
116 /* These two functions implement the VCPU id numbering: one to compute them
117  * all and one to identify thread 0 of a VCORE. Any change to the first one
118  * is likely to have an impact on the second one, so let's keep them close.
119  */
120 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
121 {
122     MachineState *ms = MACHINE(spapr);
123     unsigned int smp_threads = ms->smp.threads;
124 
125     assert(spapr->vsmt);
126     return
127         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
128 }
129 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
130                                       PowerPCCPU *cpu)
131 {
132     assert(spapr->vsmt);
133     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
134 }
135 
136 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
137 {
138     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
139      * and newer QEMUs don't even have them. In both cases, we don't want
140      * to send anything on the wire.
141      */
142     return false;
143 }
144 
145 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
146     /*
147      * Hack ahead.  We can't have two devices with the same name and
148      * instance id.  So I rename this to pass make check.
149      * Real help from people who knows the hardware is needed.
150      */
151     .name = "icp/server",
152     .version_id = 1,
153     .minimum_version_id = 1,
154     .needed = pre_2_10_vmstate_dummy_icp_needed,
155     .fields = (const VMStateField[]) {
156         VMSTATE_UNUSED(4), /* uint32_t xirr */
157         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
158         VMSTATE_UNUSED(1), /* uint8_t mfrr */
159         VMSTATE_END_OF_LIST()
160     },
161 };
162 
163 /*
164  * See comment in hw/intc/xics.c:icp_realize()
165  *
166  * You have to remove vmstate_replace_hack_for_ppc() when you remove
167  * the machine types that need the following function.
168  */
169 static void pre_2_10_vmstate_register_dummy_icp(int i)
170 {
171     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
172                      (void *)(uintptr_t) i);
173 }
174 
175 /*
176  * See comment in hw/intc/xics.c:icp_realize()
177  *
178  * You have to remove vmstate_replace_hack_for_ppc() when you remove
179  * the machine types that need the following function.
180  */
181 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
182 {
183     /*
184      * This used to be:
185      *
186      *    vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
187      *                      (void *)(uintptr_t) i);
188      */
189 }
190 
191 int spapr_max_server_number(SpaprMachineState *spapr)
192 {
193     MachineState *ms = MACHINE(spapr);
194 
195     assert(spapr->vsmt);
196     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
197 }
198 
199 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
200                                   int smt_threads)
201 {
202     int i, ret = 0;
203     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
204     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
205     int index = spapr_get_vcpu_id(cpu);
206 
207     if (cpu->compat_pvr) {
208         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
209         if (ret < 0) {
210             return ret;
211         }
212     }
213 
214     /* Build interrupt servers and gservers properties */
215     for (i = 0; i < smt_threads; i++) {
216         servers_prop[i] = cpu_to_be32(index + i);
217         /* Hack, direct the group queues back to cpu 0 */
218         gservers_prop[i*2] = cpu_to_be32(index + i);
219         gservers_prop[i*2 + 1] = 0;
220     }
221     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
222                       servers_prop, sizeof(*servers_prop) * smt_threads);
223     if (ret < 0) {
224         return ret;
225     }
226     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
227                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
228 
229     return ret;
230 }
231 
232 static void spapr_dt_pa_features(SpaprMachineState *spapr,
233                                  PowerPCCPU *cpu,
234                                  void *fdt, int offset)
235 {
236     /*
237      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
238      * but not MTTCG, so disable it. To advertise it, a cap would have
239      * to be added, or support implemented for MTTCG.
240      *
241      * Copy/paste is not supported by TCG, so it is not advertised. KVM
242      * can execute them but it has no accelerator drivers which are usable,
243      * so there isn't much need for it anyway.
244      */
245 
246     /* These should be kept in sync with pnv */
247     uint8_t pa_features_206[] = { 6, 0,
248         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
249     uint8_t pa_features_207[] = { 24, 0,
250         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
251         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
252         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
254     uint8_t pa_features_300[] = { 66, 0,
255         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
256         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
257         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
258         /* 6: DS207 */
259         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
260         /* 16: Vector */
261         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
262         /* 18: Vec. Scalar, 20: Vec. XOR */
263         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
264         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
265         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
266         /* 32: LE atomic, 34: EBB + ext EBB */
267         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
268         /* 40: Radix MMU */
269         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
270         /* 42: PM, 44: PC RA, 46: SC vec'd */
271         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
272         /* 48: SIMD, 50: QP BFP, 52: String */
273         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
274         /* 54: DecFP, 56: DecI, 58: SHA */
275         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
276         /* 60: NM atomic, 62: RNG */
277         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
278     };
279     /* 3.1 removes SAO, HTM support */
280     uint8_t pa_features_31[] = { 74, 0,
281         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
282         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
283         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
284         /* 6: DS207 */
285         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
286         /* 16: Vector */
287         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
288         /* 18: Vec. Scalar, 20: Vec. XOR */
289         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
290         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
291         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
292         /* 32: LE atomic, 34: EBB + ext EBB */
293         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
294         /* 40: Radix MMU */
295         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
296         /* 42: PM, 44: PC RA, 46: SC vec'd */
297         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
298         /* 48: SIMD, 50: QP BFP, 52: String */
299         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
300         /* 54: DecFP, 56: DecI, 58: SHA */
301         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
302         /* 60: NM atomic, 62: RNG */
303         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
304         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
305         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
306         /* 72: [P]HASHST/[P]HASHCHK */
307         0x80, 0x00,                         /* 72 - 73 */
308     };
309     uint8_t *pa_features = NULL;
310     size_t pa_size;
311 
312     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
313         pa_features = pa_features_206;
314         pa_size = sizeof(pa_features_206);
315     }
316     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
317         pa_features = pa_features_207;
318         pa_size = sizeof(pa_features_207);
319     }
320     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
321         pa_features = pa_features_300;
322         pa_size = sizeof(pa_features_300);
323     }
324     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
325         pa_features = pa_features_31;
326         pa_size = sizeof(pa_features_31);
327     }
328     if (!pa_features) {
329         return;
330     }
331 
332     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
333         /*
334          * Note: we keep CI large pages off by default because a 64K capable
335          * guest provisioned with large pages might otherwise try to map a qemu
336          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
337          * even if that qemu runs on a 4k host.
338          * We dd this bit back here if we are confident this is not an issue
339          */
340         pa_features[3] |= 0x20;
341     }
342     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
343         pa_features[24] |= 0x80;    /* Transactional memory support */
344     }
345     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
346         /* Workaround for broken kernels that attempt (guest) radix
347          * mode when they can't handle it, if they see the radix bit set
348          * in pa-features. So hide it from them. */
349         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
350     }
351 
352     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
353 }
354 
355 static hwaddr spapr_node0_size(MachineState *machine)
356 {
357     if (machine->numa_state->num_nodes) {
358         int i;
359         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
360             if (machine->numa_state->nodes[i].node_mem) {
361                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
362                            machine->ram_size);
363             }
364         }
365     }
366     return machine->ram_size;
367 }
368 
369 static void add_str(GString *s, const gchar *s1)
370 {
371     g_string_append_len(s, s1, strlen(s1) + 1);
372 }
373 
374 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
375                                 hwaddr start, hwaddr size)
376 {
377     char mem_name[32];
378     uint64_t mem_reg_property[2];
379     int off;
380 
381     mem_reg_property[0] = cpu_to_be64(start);
382     mem_reg_property[1] = cpu_to_be64(size);
383 
384     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
385     off = fdt_add_subnode(fdt, 0, mem_name);
386     _FDT(off);
387     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
388     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
389                       sizeof(mem_reg_property))));
390     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
391     return off;
392 }
393 
394 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
395 {
396     MemoryDeviceInfoList *info;
397 
398     for (info = list; info; info = info->next) {
399         MemoryDeviceInfo *value = info->value;
400 
401         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
402             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
403 
404             if (addr >= pcdimm_info->addr &&
405                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
406                 return pcdimm_info->node;
407             }
408         }
409     }
410 
411     return -1;
412 }
413 
414 struct sPAPRDrconfCellV2 {
415      uint32_t seq_lmbs;
416      uint64_t base_addr;
417      uint32_t drc_index;
418      uint32_t aa_index;
419      uint32_t flags;
420 } QEMU_PACKED;
421 
422 typedef struct DrconfCellQueue {
423     struct sPAPRDrconfCellV2 cell;
424     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
425 } DrconfCellQueue;
426 
427 static DrconfCellQueue *
428 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
429                       uint32_t drc_index, uint32_t aa_index,
430                       uint32_t flags)
431 {
432     DrconfCellQueue *elem;
433 
434     elem = g_malloc0(sizeof(*elem));
435     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
436     elem->cell.base_addr = cpu_to_be64(base_addr);
437     elem->cell.drc_index = cpu_to_be32(drc_index);
438     elem->cell.aa_index = cpu_to_be32(aa_index);
439     elem->cell.flags = cpu_to_be32(flags);
440 
441     return elem;
442 }
443 
444 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
445                                       int offset, MemoryDeviceInfoList *dimms)
446 {
447     MachineState *machine = MACHINE(spapr);
448     uint8_t *int_buf, *cur_index;
449     int ret;
450     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
451     uint64_t addr, cur_addr, size;
452     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
453     uint64_t mem_end = machine->device_memory->base +
454                        memory_region_size(&machine->device_memory->mr);
455     uint32_t node, buf_len, nr_entries = 0;
456     SpaprDrc *drc;
457     DrconfCellQueue *elem, *next;
458     MemoryDeviceInfoList *info;
459     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
460         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
461 
462     /* Entry to cover RAM and the gap area */
463     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
464                                  SPAPR_LMB_FLAGS_RESERVED |
465                                  SPAPR_LMB_FLAGS_DRC_INVALID);
466     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
467     nr_entries++;
468 
469     cur_addr = machine->device_memory->base;
470     for (info = dimms; info; info = info->next) {
471         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
472 
473         addr = di->addr;
474         size = di->size;
475         node = di->node;
476 
477         /*
478          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
479          * area is marked hotpluggable in the next iteration for the bigger
480          * chunk including the NVDIMM occupied area.
481          */
482         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
483             continue;
484 
485         /* Entry for hot-pluggable area */
486         if (cur_addr < addr) {
487             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
488             g_assert(drc);
489             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
490                                          cur_addr, spapr_drc_index(drc), -1, 0);
491             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
492             nr_entries++;
493         }
494 
495         /* Entry for DIMM */
496         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
497         g_assert(drc);
498         elem = spapr_get_drconf_cell(size / lmb_size, addr,
499                                      spapr_drc_index(drc), node,
500                                      (SPAPR_LMB_FLAGS_ASSIGNED |
501                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
502         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
503         nr_entries++;
504         cur_addr = addr + size;
505     }
506 
507     /* Entry for remaining hotpluggable area */
508     if (cur_addr < mem_end) {
509         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
510         g_assert(drc);
511         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
512                                      cur_addr, spapr_drc_index(drc), -1, 0);
513         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
514         nr_entries++;
515     }
516 
517     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
518     int_buf = cur_index = g_malloc0(buf_len);
519     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
520     cur_index += sizeof(nr_entries);
521 
522     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
523         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
524         cur_index += sizeof(elem->cell);
525         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
526         g_free(elem);
527     }
528 
529     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
530     g_free(int_buf);
531     if (ret < 0) {
532         return -1;
533     }
534     return 0;
535 }
536 
537 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
538                                    int offset, MemoryDeviceInfoList *dimms)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int i, ret;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
544     uint32_t nr_lmbs = (machine->device_memory->base +
545                        memory_region_size(&machine->device_memory->mr)) /
546                        lmb_size;
547     uint32_t *int_buf, *cur_index, buf_len;
548 
549     /*
550      * Allocate enough buffer size to fit in ibm,dynamic-memory
551      */
552     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
553     cur_index = int_buf = g_malloc0(buf_len);
554     int_buf[0] = cpu_to_be32(nr_lmbs);
555     cur_index++;
556     for (i = 0; i < nr_lmbs; i++) {
557         uint64_t addr = i * lmb_size;
558         uint32_t *dynamic_memory = cur_index;
559 
560         if (i >= device_lmb_start) {
561             SpaprDrc *drc;
562 
563             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
564             g_assert(drc);
565 
566             dynamic_memory[0] = cpu_to_be32(addr >> 32);
567             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
568             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
569             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
570             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
571             if (memory_region_present(get_system_memory(), addr)) {
572                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
573             } else {
574                 dynamic_memory[5] = cpu_to_be32(0);
575             }
576         } else {
577             /*
578              * LMB information for RMA, boot time RAM and gap b/n RAM and
579              * device memory region -- all these are marked as reserved
580              * and as having no valid DRC.
581              */
582             dynamic_memory[0] = cpu_to_be32(addr >> 32);
583             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
584             dynamic_memory[2] = cpu_to_be32(0);
585             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
586             dynamic_memory[4] = cpu_to_be32(-1);
587             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
588                                             SPAPR_LMB_FLAGS_DRC_INVALID);
589         }
590 
591         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
592     }
593     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
594     g_free(int_buf);
595     if (ret < 0) {
596         return -1;
597     }
598     return 0;
599 }
600 
601 /*
602  * Adds ibm,dynamic-reconfiguration-memory node.
603  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
604  * of this device tree node.
605  */
606 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
607                                                    void *fdt)
608 {
609     MachineState *machine = MACHINE(spapr);
610     int ret, offset;
611     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
612     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
613                                 cpu_to_be32(lmb_size & 0xffffffff)};
614     MemoryDeviceInfoList *dimms = NULL;
615 
616     /* Don't create the node if there is no device memory. */
617     if (!machine->device_memory) {
618         return 0;
619     }
620 
621     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
622 
623     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
624                     sizeof(prop_lmb_size));
625     if (ret < 0) {
626         return ret;
627     }
628 
629     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
630     if (ret < 0) {
631         return ret;
632     }
633 
634     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
635     if (ret < 0) {
636         return ret;
637     }
638 
639     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
640     dimms = qmp_memory_device_list();
641     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
642         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
643     } else {
644         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
645     }
646     qapi_free_MemoryDeviceInfoList(dimms);
647 
648     if (ret < 0) {
649         return ret;
650     }
651 
652     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
653 
654     return ret;
655 }
656 
657 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
658 {
659     MachineState *machine = MACHINE(spapr);
660     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
661     hwaddr mem_start, node_size;
662     int i, nb_nodes = machine->numa_state->num_nodes;
663     NodeInfo *nodes = machine->numa_state->nodes;
664 
665     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
666         if (!nodes[i].node_mem) {
667             continue;
668         }
669         if (mem_start >= machine->ram_size) {
670             node_size = 0;
671         } else {
672             node_size = nodes[i].node_mem;
673             if (node_size > machine->ram_size - mem_start) {
674                 node_size = machine->ram_size - mem_start;
675             }
676         }
677         if (!mem_start) {
678             /* spapr_machine_init() checks for rma_size <= node0_size
679              * already */
680             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
681             mem_start += spapr->rma_size;
682             node_size -= spapr->rma_size;
683         }
684         for ( ; node_size; ) {
685             hwaddr sizetmp = pow2floor(node_size);
686 
687             /* mem_start != 0 here */
688             if (ctzl(mem_start) < ctzl(sizetmp)) {
689                 sizetmp = 1ULL << ctzl(mem_start);
690             }
691 
692             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
693             node_size -= sizetmp;
694             mem_start += sizetmp;
695         }
696     }
697 
698     /* Generate ibm,dynamic-reconfiguration-memory node if required */
699     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
700         int ret;
701 
702         g_assert(smc->dr_lmb_enabled);
703         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
704         if (ret) {
705             return ret;
706         }
707     }
708 
709     return 0;
710 }
711 
712 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
713                          SpaprMachineState *spapr)
714 {
715     MachineState *ms = MACHINE(spapr);
716     PowerPCCPU *cpu = POWERPC_CPU(cs);
717     CPUPPCState *env = &cpu->env;
718     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
719     int index = spapr_get_vcpu_id(cpu);
720     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
721                        0xffffffff, 0xffffffff};
722     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
723         : SPAPR_TIMEBASE_FREQ;
724     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
725     uint32_t page_sizes_prop[64];
726     size_t page_sizes_prop_size;
727     unsigned int smp_threads = ms->smp.threads;
728     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
729     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
730     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
731     SpaprDrc *drc;
732     int drc_index;
733     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
734     int i;
735 
736     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
737     if (drc) {
738         drc_index = spapr_drc_index(drc);
739         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
740     }
741 
742     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
743     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
744 
745     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
746     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
747                            env->dcache_line_size)));
748     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
749                            env->dcache_line_size)));
750     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
751                            env->icache_line_size)));
752     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
753                            env->icache_line_size)));
754 
755     if (pcc->l1_dcache_size) {
756         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
757                                pcc->l1_dcache_size)));
758     } else {
759         warn_report("Unknown L1 dcache size for cpu");
760     }
761     if (pcc->l1_icache_size) {
762         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
763                                pcc->l1_icache_size)));
764     } else {
765         warn_report("Unknown L1 icache size for cpu");
766     }
767 
768     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
769     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
770     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
771     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
772     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
773     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
774 
775     if (ppc_has_spr(cpu, SPR_PURR)) {
776         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
777     }
778     if (ppc_has_spr(cpu, SPR_PURR)) {
779         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
780     }
781 
782     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
783         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
784                           segs, sizeof(segs))));
785     }
786 
787     /* Advertise VSX (vector extensions) if available
788      *   1               == VMX / Altivec available
789      *   2               == VSX available
790      *
791      * Only CPUs for which we create core types in spapr_cpu_core.c
792      * are possible, and all of those have VMX */
793     if (env->insns_flags & PPC_ALTIVEC) {
794         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
795             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
796         } else {
797             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
798         }
799     }
800 
801     /* Advertise DFP (Decimal Floating Point) if available
802      *   0 / no property == no DFP
803      *   1               == DFP available */
804     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
805         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
806     }
807 
808     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
809                                                       sizeof(page_sizes_prop));
810     if (page_sizes_prop_size) {
811         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
812                           page_sizes_prop, page_sizes_prop_size)));
813     }
814 
815     spapr_dt_pa_features(spapr, cpu, fdt, offset);
816 
817     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
818                            cs->cpu_index / vcpus_per_socket)));
819 
820     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
821                       pft_size_prop, sizeof(pft_size_prop))));
822 
823     if (ms->numa_state->num_nodes > 1) {
824         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
825     }
826 
827     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
828 
829     if (pcc->radix_page_info) {
830         for (i = 0; i < pcc->radix_page_info->count; i++) {
831             radix_AP_encodings[i] =
832                 cpu_to_be32(pcc->radix_page_info->entries[i]);
833         }
834         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
835                           radix_AP_encodings,
836                           pcc->radix_page_info->count *
837                           sizeof(radix_AP_encodings[0]))));
838     }
839 
840     /*
841      * We set this property to let the guest know that it can use the large
842      * decrementer and its width in bits.
843      */
844     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
845         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
846                               pcc->lrg_decr_bits)));
847 }
848 
849 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
850                              int cpus_offset)
851 {
852     PowerPCCPU *cpu = POWERPC_CPU(cs);
853     int index = spapr_get_vcpu_id(cpu);
854     DeviceClass *dc = DEVICE_GET_CLASS(cs);
855     g_autofree char *nodename = NULL;
856     int offset;
857 
858     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
859         return;
860     }
861 
862     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
863     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
864     _FDT(offset);
865     spapr_dt_cpu(cs, fdt, offset, spapr);
866 }
867 
868 
869 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
870 {
871     CPUState **rev;
872     CPUState *cs;
873     int n_cpus;
874     int cpus_offset;
875     int i;
876 
877     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
878     _FDT(cpus_offset);
879     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
880     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
881 
882     /*
883      * We walk the CPUs in reverse order to ensure that CPU DT nodes
884      * created by fdt_add_subnode() end up in the right order in FDT
885      * for the guest kernel the enumerate the CPUs correctly.
886      *
887      * The CPU list cannot be traversed in reverse order, so we need
888      * to do extra work.
889      */
890     n_cpus = 0;
891     rev = NULL;
892     CPU_FOREACH(cs) {
893         rev = g_renew(CPUState *, rev, n_cpus + 1);
894         rev[n_cpus++] = cs;
895     }
896 
897     for (i = n_cpus - 1; i >= 0; i--) {
898         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
899     }
900 
901     g_free(rev);
902 }
903 
904 static int spapr_dt_rng(void *fdt)
905 {
906     int node;
907     int ret;
908 
909     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
910     if (node <= 0) {
911         return -1;
912     }
913     ret = fdt_setprop_string(fdt, node, "device_type",
914                              "ibm,platform-facilities");
915     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
916     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
917 
918     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
919     if (node <= 0) {
920         return -1;
921     }
922     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
923 
924     return ret ? -1 : 0;
925 }
926 
927 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
928 {
929     MachineState *ms = MACHINE(spapr);
930     int rtas;
931     GString *hypertas = g_string_sized_new(256);
932     GString *qemu_hypertas = g_string_sized_new(256);
933     uint32_t lrdr_capacity[] = {
934         0,
935         0,
936         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
937         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
938         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
939     };
940 
941     /* Do we have device memory? */
942     if (MACHINE(spapr)->device_memory) {
943         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
944             memory_region_size(&MACHINE(spapr)->device_memory->mr);
945 
946         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
947         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
948     }
949 
950     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
951 
952     /* hypertas */
953     add_str(hypertas, "hcall-pft");
954     add_str(hypertas, "hcall-term");
955     add_str(hypertas, "hcall-dabr");
956     add_str(hypertas, "hcall-interrupt");
957     add_str(hypertas, "hcall-tce");
958     add_str(hypertas, "hcall-vio");
959     add_str(hypertas, "hcall-splpar");
960     add_str(hypertas, "hcall-join");
961     add_str(hypertas, "hcall-bulk");
962     add_str(hypertas, "hcall-set-mode");
963     add_str(hypertas, "hcall-sprg0");
964     add_str(hypertas, "hcall-copy");
965     add_str(hypertas, "hcall-debug");
966     add_str(hypertas, "hcall-vphn");
967     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
968         add_str(hypertas, "hcall-rpt-invalidate");
969     }
970 
971     add_str(qemu_hypertas, "hcall-memop1");
972 
973     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
974         add_str(hypertas, "hcall-multi-tce");
975     }
976 
977     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
978         add_str(hypertas, "hcall-hpt-resize");
979     }
980 
981     add_str(hypertas, "hcall-watchdog");
982 
983     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
984                      hypertas->str, hypertas->len));
985     g_string_free(hypertas, TRUE);
986     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
987                      qemu_hypertas->str, qemu_hypertas->len));
988     g_string_free(qemu_hypertas, TRUE);
989 
990     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
991 
992     /*
993      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
994      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
995      *
996      * The system reset requirements are driven by existing Linux and PowerVM
997      * implementation which (contrary to PAPR) saves r3 in the error log
998      * structure like machine check, so Linux expects to find the saved r3
999      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
1000      * does not look at the error value).
1001      *
1002      * System reset interrupts are not subject to interlock like machine
1003      * check, so this memory area could be corrupted if the sreset is
1004      * interrupted by a machine check (or vice versa) if it was shared. To
1005      * prevent this, system reset uses per-CPU areas for the sreset save
1006      * area. A system reset that interrupts a system reset handler could
1007      * still overwrite this area, but Linux doesn't try to recover in that
1008      * case anyway.
1009      *
1010      * The extra 8 bytes is required because Linux's FWNMI error log check
1011      * is off-by-one.
1012      *
1013      * RTAS_MIN_SIZE is required for the RTAS blob itself.
1014      */
1015     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
1016                           RTAS_ERROR_LOG_MAX +
1017                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
1018                           sizeof(uint64_t)));
1019     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1020                           RTAS_ERROR_LOG_MAX));
1021     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1022                           RTAS_EVENT_SCAN_RATE));
1023 
1024     g_assert(msi_nonbroken);
1025     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1026 
1027     /*
1028      * According to PAPR, rtas ibm,os-term does not guarantee a return
1029      * back to the guest cpu.
1030      *
1031      * While an additional ibm,extended-os-term property indicates
1032      * that rtas call return will always occur. Set this property.
1033      */
1034     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1035 
1036     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1037                      lrdr_capacity, sizeof(lrdr_capacity)));
1038 
1039     spapr_dt_rtas_tokens(fdt, rtas);
1040 }
1041 
1042 /*
1043  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1044  * and the XIVE features that the guest may request and thus the valid
1045  * values for bytes 23..26 of option vector 5:
1046  */
1047 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1048                                           int chosen)
1049 {
1050     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1051 
1052     char val[2 * 4] = {
1053         23, 0x00, /* XICS / XIVE mode */
1054         24, 0x00, /* Hash/Radix, filled in below. */
1055         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1056         26, 0x40, /* Radix options: GTSE == yes. */
1057     };
1058 
1059     if (spapr->irq->xics && spapr->irq->xive) {
1060         val[1] = SPAPR_OV5_XIVE_BOTH;
1061     } else if (spapr->irq->xive) {
1062         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1063     } else {
1064         assert(spapr->irq->xics);
1065         val[1] = SPAPR_OV5_XIVE_LEGACY;
1066     }
1067 
1068     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1069                           first_ppc_cpu->compat_pvr)) {
1070         /*
1071          * If we're in a pre POWER9 compat mode then the guest should
1072          * do hash and use the legacy interrupt mode
1073          */
1074         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1075         val[3] = 0x00; /* Hash */
1076         spapr_check_mmu_mode(false);
1077     } else if (kvm_enabled()) {
1078         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1079             val[3] = 0x80; /* OV5_MMU_BOTH */
1080         } else if (kvmppc_has_cap_mmu_radix()) {
1081             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1082         } else {
1083             val[3] = 0x00; /* Hash */
1084         }
1085     } else {
1086         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1087         val[3] = 0xC0;
1088     }
1089     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1090                      val, sizeof(val)));
1091 }
1092 
1093 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1094 {
1095     MachineState *machine = MACHINE(spapr);
1096     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1097     int chosen;
1098 
1099     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1100 
1101     if (reset) {
1102         const char *boot_device = spapr->boot_device;
1103         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1104         size_t cb = 0;
1105         g_autofree char *bootlist = get_boot_devices_list(&cb);
1106 
1107         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1108             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1109                                     machine->kernel_cmdline));
1110         }
1111 
1112         if (spapr->initrd_size) {
1113             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1114                                   spapr->initrd_base));
1115             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1116                                   spapr->initrd_base + spapr->initrd_size));
1117         }
1118 
1119         if (spapr->kernel_size) {
1120             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1121                                   cpu_to_be64(spapr->kernel_size) };
1122 
1123             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1124                          &kprop, sizeof(kprop)));
1125             if (spapr->kernel_le) {
1126                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1127             }
1128         }
1129         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1130             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1131         }
1132         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1133         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1134         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1135 
1136         if (cb && bootlist) {
1137             int i;
1138 
1139             for (i = 0; i < cb; i++) {
1140                 if (bootlist[i] == '\n') {
1141                     bootlist[i] = ' ';
1142                 }
1143             }
1144             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1145         }
1146 
1147         if (boot_device && strlen(boot_device)) {
1148             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1149         }
1150 
1151         if (spapr->want_stdout_path && stdout_path) {
1152             /*
1153              * "linux,stdout-path" and "stdout" properties are
1154              * deprecated by linux kernel. New platforms should only
1155              * use the "stdout-path" property. Set the new property
1156              * and continue using older property to remain compatible
1157              * with the existing firmware.
1158              */
1159             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1160             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1161         }
1162 
1163         /*
1164          * We can deal with BAR reallocation just fine, advertise it
1165          * to the guest
1166          */
1167         if (smc->linux_pci_probe) {
1168             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1169         }
1170 
1171         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1172     }
1173 
1174     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1175 
1176     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1177 }
1178 
1179 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1180 {
1181     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1182      * KVM to work under pHyp with some guest co-operation */
1183     int hypervisor;
1184     uint8_t hypercall[16];
1185 
1186     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1187     /* indicate KVM hypercall interface */
1188     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1189     if (kvmppc_has_cap_fixup_hcalls()) {
1190         /*
1191          * Older KVM versions with older guest kernels were broken
1192          * with the magic page, don't allow the guest to map it.
1193          */
1194         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1195                                   sizeof(hypercall))) {
1196             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1197                              hypercall, sizeof(hypercall)));
1198         }
1199     }
1200 }
1201 
1202 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1203 {
1204     MachineState *machine = MACHINE(spapr);
1205     MachineClass *mc = MACHINE_GET_CLASS(machine);
1206     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1207     uint32_t root_drc_type_mask = 0;
1208     int ret;
1209     void *fdt;
1210     SpaprPhbState *phb;
1211     char *buf;
1212 
1213     fdt = g_malloc0(space);
1214     _FDT((fdt_create_empty_tree(fdt, space)));
1215 
1216     /* Root node */
1217     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1218     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1219     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1220 
1221     /* Guest UUID & Name*/
1222     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1223     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1224     if (qemu_uuid_set) {
1225         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1226     }
1227     g_free(buf);
1228 
1229     if (qemu_get_vm_name()) {
1230         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1231                                 qemu_get_vm_name()));
1232     }
1233 
1234     /* Host Model & Serial Number */
1235     if (spapr->host_model) {
1236         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1237     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1238         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1239         g_free(buf);
1240     }
1241 
1242     if (spapr->host_serial) {
1243         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1244     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1245         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1246         g_free(buf);
1247     }
1248 
1249     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1250     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1251 
1252     /* /interrupt controller */
1253     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1254 
1255     ret = spapr_dt_memory(spapr, fdt);
1256     if (ret < 0) {
1257         error_report("couldn't setup memory nodes in fdt");
1258         exit(1);
1259     }
1260 
1261     /* /vdevice */
1262     spapr_dt_vdevice(spapr->vio_bus, fdt);
1263 
1264     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1265         ret = spapr_dt_rng(fdt);
1266         if (ret < 0) {
1267             error_report("could not set up rng device in the fdt");
1268             exit(1);
1269         }
1270     }
1271 
1272     QLIST_FOREACH(phb, &spapr->phbs, list) {
1273         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1274         if (ret < 0) {
1275             error_report("couldn't setup PCI devices in fdt");
1276             exit(1);
1277         }
1278     }
1279 
1280     spapr_dt_cpus(fdt, spapr);
1281 
1282     /* ibm,drc-indexes and friends */
1283     if (smc->dr_lmb_enabled) {
1284         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1285     }
1286     if (smc->dr_phb_enabled) {
1287         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1288     }
1289     if (mc->nvdimm_supported) {
1290         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1291     }
1292     if (root_drc_type_mask) {
1293         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1294     }
1295 
1296     if (mc->has_hotpluggable_cpus) {
1297         int offset = fdt_path_offset(fdt, "/cpus");
1298         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1299         if (ret < 0) {
1300             error_report("Couldn't set up CPU DR device tree properties");
1301             exit(1);
1302         }
1303     }
1304 
1305     /* /event-sources */
1306     spapr_dt_events(spapr, fdt);
1307 
1308     /* /rtas */
1309     spapr_dt_rtas(spapr, fdt);
1310 
1311     /* /chosen */
1312     spapr_dt_chosen(spapr, fdt, reset);
1313 
1314     /* /hypervisor */
1315     if (kvm_enabled()) {
1316         spapr_dt_hypervisor(spapr, fdt);
1317     }
1318 
1319     /* Build memory reserve map */
1320     if (reset) {
1321         if (spapr->kernel_size) {
1322             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1323                                   spapr->kernel_size)));
1324         }
1325         if (spapr->initrd_size) {
1326             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1327                                   spapr->initrd_size)));
1328         }
1329     }
1330 
1331     /* NVDIMM devices */
1332     if (mc->nvdimm_supported) {
1333         spapr_dt_persistent_memory(spapr, fdt);
1334     }
1335 
1336     return fdt;
1337 }
1338 
1339 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1340 {
1341     SpaprMachineState *spapr = opaque;
1342 
1343     return (addr & 0x0fffffff) + spapr->kernel_addr;
1344 }
1345 
1346 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1347                                     PowerPCCPU *cpu)
1348 {
1349     CPUPPCState *env = &cpu->env;
1350 
1351     /* The TCG path should also be holding the BQL at this point */
1352     g_assert(bql_locked());
1353 
1354     g_assert(!vhyp_cpu_in_nested(cpu));
1355 
1356     if (FIELD_EX64(env->msr, MSR, PR)) {
1357         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1358         env->gpr[3] = H_PRIVILEGE;
1359     } else {
1360         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1361     }
1362 }
1363 
1364 struct LPCRSyncState {
1365     target_ulong value;
1366     target_ulong mask;
1367 };
1368 
1369 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1370 {
1371     struct LPCRSyncState *s = arg.host_ptr;
1372     PowerPCCPU *cpu = POWERPC_CPU(cs);
1373     CPUPPCState *env = &cpu->env;
1374     target_ulong lpcr;
1375 
1376     cpu_synchronize_state(cs);
1377     lpcr = env->spr[SPR_LPCR];
1378     lpcr &= ~s->mask;
1379     lpcr |= s->value;
1380     ppc_store_lpcr(cpu, lpcr);
1381 }
1382 
1383 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1384 {
1385     CPUState *cs;
1386     struct LPCRSyncState s = {
1387         .value = value,
1388         .mask = mask
1389     };
1390     CPU_FOREACH(cs) {
1391         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1392     }
1393 }
1394 
1395 /* May be used when the machine is not running */
1396 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1397 {
1398     CPUState *cs;
1399     CPU_FOREACH(cs) {
1400         PowerPCCPU *cpu = POWERPC_CPU(cs);
1401         CPUPPCState *env = &cpu->env;
1402         target_ulong lpcr;
1403 
1404         lpcr = env->spr[SPR_LPCR];
1405         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1406         ppc_store_lpcr(cpu, lpcr);
1407     }
1408 }
1409 
1410 
1411 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1412                            target_ulong lpid, ppc_v3_pate_t *entry)
1413 {
1414     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1415     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1416 
1417     if (!spapr_cpu->in_nested) {
1418         assert(lpid == 0);
1419 
1420         /* Copy PATE1:GR into PATE0:HR */
1421         entry->dw0 = spapr->patb_entry & PATE0_HR;
1422         entry->dw1 = spapr->patb_entry;
1423 
1424     } else {
1425         uint64_t patb, pats;
1426 
1427         assert(lpid != 0);
1428 
1429         patb = spapr->nested_ptcr & PTCR_PATB;
1430         pats = spapr->nested_ptcr & PTCR_PATS;
1431 
1432         /* Check if partition table is properly aligned */
1433         if (patb & MAKE_64BIT_MASK(0, pats + 12)) {
1434             return false;
1435         }
1436 
1437         /* Calculate number of entries */
1438         pats = 1ull << (pats + 12 - 4);
1439         if (pats <= lpid) {
1440             return false;
1441         }
1442 
1443         /* Grab entry */
1444         patb += 16 * lpid;
1445         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1446         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1447     }
1448 
1449     return true;
1450 }
1451 
1452 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1453 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1454 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1455 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1456 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1457 
1458 /*
1459  * Get the fd to access the kernel htab, re-opening it if necessary
1460  */
1461 static int get_htab_fd(SpaprMachineState *spapr)
1462 {
1463     Error *local_err = NULL;
1464 
1465     if (spapr->htab_fd >= 0) {
1466         return spapr->htab_fd;
1467     }
1468 
1469     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1470     if (spapr->htab_fd < 0) {
1471         error_report_err(local_err);
1472     }
1473 
1474     return spapr->htab_fd;
1475 }
1476 
1477 void close_htab_fd(SpaprMachineState *spapr)
1478 {
1479     if (spapr->htab_fd >= 0) {
1480         close(spapr->htab_fd);
1481     }
1482     spapr->htab_fd = -1;
1483 }
1484 
1485 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1486 {
1487     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1488 
1489     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1490 }
1491 
1492 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1493 {
1494     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1495 
1496     assert(kvm_enabled());
1497 
1498     if (!spapr->htab) {
1499         return 0;
1500     }
1501 
1502     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1503 }
1504 
1505 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1506                                                 hwaddr ptex, int n)
1507 {
1508     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1509     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1510 
1511     if (!spapr->htab) {
1512         /*
1513          * HTAB is controlled by KVM. Fetch into temporary buffer
1514          */
1515         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1516         kvmppc_read_hptes(hptes, ptex, n);
1517         return hptes;
1518     }
1519 
1520     /*
1521      * HTAB is controlled by QEMU. Just point to the internally
1522      * accessible PTEG.
1523      */
1524     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1525 }
1526 
1527 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1528                               const ppc_hash_pte64_t *hptes,
1529                               hwaddr ptex, int n)
1530 {
1531     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1532 
1533     if (!spapr->htab) {
1534         g_free((void *)hptes);
1535     }
1536 
1537     /* Nothing to do for qemu managed HPT */
1538 }
1539 
1540 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1541                       uint64_t pte0, uint64_t pte1)
1542 {
1543     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1544     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1545 
1546     if (!spapr->htab) {
1547         kvmppc_write_hpte(ptex, pte0, pte1);
1548     } else {
1549         if (pte0 & HPTE64_V_VALID) {
1550             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1551             /*
1552              * When setting valid, we write PTE1 first. This ensures
1553              * proper synchronization with the reading code in
1554              * ppc_hash64_pteg_search()
1555              */
1556             smp_wmb();
1557             stq_p(spapr->htab + offset, pte0);
1558         } else {
1559             stq_p(spapr->htab + offset, pte0);
1560             /*
1561              * When clearing it we set PTE0 first. This ensures proper
1562              * synchronization with the reading code in
1563              * ppc_hash64_pteg_search()
1564              */
1565             smp_wmb();
1566             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1567         }
1568     }
1569 }
1570 
1571 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1572                              uint64_t pte1)
1573 {
1574     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1575     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1576 
1577     if (!spapr->htab) {
1578         /* There should always be a hash table when this is called */
1579         error_report("spapr_hpte_set_c called with no hash table !");
1580         return;
1581     }
1582 
1583     /* The HW performs a non-atomic byte update */
1584     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1585 }
1586 
1587 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1588                              uint64_t pte1)
1589 {
1590     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1591     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1592 
1593     if (!spapr->htab) {
1594         /* There should always be a hash table when this is called */
1595         error_report("spapr_hpte_set_r called with no hash table !");
1596         return;
1597     }
1598 
1599     /* The HW performs a non-atomic byte update */
1600     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1601 }
1602 
1603 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1604 {
1605     int shift;
1606 
1607     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1608      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1609      * that's much more than is needed for Linux guests */
1610     shift = ctz64(pow2ceil(ramsize)) - 7;
1611     shift = MAX(shift, 18); /* Minimum architected size */
1612     shift = MIN(shift, 46); /* Maximum architected size */
1613     return shift;
1614 }
1615 
1616 void spapr_free_hpt(SpaprMachineState *spapr)
1617 {
1618     qemu_vfree(spapr->htab);
1619     spapr->htab = NULL;
1620     spapr->htab_shift = 0;
1621     close_htab_fd(spapr);
1622 }
1623 
1624 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1625 {
1626     ERRP_GUARD();
1627     long rc;
1628 
1629     /* Clean up any HPT info from a previous boot */
1630     spapr_free_hpt(spapr);
1631 
1632     rc = kvmppc_reset_htab(shift);
1633 
1634     if (rc == -EOPNOTSUPP) {
1635         error_setg(errp, "HPT not supported in nested guests");
1636         return -EOPNOTSUPP;
1637     }
1638 
1639     if (rc < 0) {
1640         /* kernel-side HPT needed, but couldn't allocate one */
1641         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1642                          shift);
1643         error_append_hint(errp, "Try smaller maxmem?\n");
1644         return -errno;
1645     } else if (rc > 0) {
1646         /* kernel-side HPT allocated */
1647         if (rc != shift) {
1648             error_setg(errp,
1649                        "Requested order %d HPT, but kernel allocated order %ld",
1650                        shift, rc);
1651             error_append_hint(errp, "Try smaller maxmem?\n");
1652             return -ENOSPC;
1653         }
1654 
1655         spapr->htab_shift = shift;
1656         spapr->htab = NULL;
1657     } else {
1658         /* kernel-side HPT not needed, allocate in userspace instead */
1659         size_t size = 1ULL << shift;
1660         int i;
1661 
1662         spapr->htab = qemu_memalign(size, size);
1663         memset(spapr->htab, 0, size);
1664         spapr->htab_shift = shift;
1665 
1666         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1667             DIRTY_HPTE(HPTE(spapr->htab, i));
1668         }
1669     }
1670     /* We're setting up a hash table, so that means we're not radix */
1671     spapr->patb_entry = 0;
1672     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1673     return 0;
1674 }
1675 
1676 void spapr_setup_hpt(SpaprMachineState *spapr)
1677 {
1678     int hpt_shift;
1679 
1680     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1681         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1682     } else {
1683         uint64_t current_ram_size;
1684 
1685         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1686         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1687     }
1688     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1689 
1690     if (kvm_enabled()) {
1691         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1692 
1693         /* Check our RMA fits in the possible VRMA */
1694         if (vrma_limit < spapr->rma_size) {
1695             error_report("Unable to create %" HWADDR_PRIu
1696                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1697                          spapr->rma_size / MiB, vrma_limit / MiB);
1698             exit(EXIT_FAILURE);
1699         }
1700     }
1701 }
1702 
1703 void spapr_check_mmu_mode(bool guest_radix)
1704 {
1705     if (guest_radix) {
1706         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1707             error_report("Guest requested unavailable MMU mode (radix).");
1708             exit(EXIT_FAILURE);
1709         }
1710     } else {
1711         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1712             && !kvmppc_has_cap_mmu_hash_v3()) {
1713             error_report("Guest requested unavailable MMU mode (hash).");
1714             exit(EXIT_FAILURE);
1715         }
1716     }
1717 }
1718 
1719 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1720 {
1721     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1722     PowerPCCPU *first_ppc_cpu;
1723     hwaddr fdt_addr;
1724     void *fdt;
1725     int rc;
1726 
1727     if (reason != SHUTDOWN_CAUSE_SNAPSHOT_LOAD) {
1728         /*
1729          * Record-replay snapshot load must not consume random, this was
1730          * already replayed from initial machine reset.
1731          */
1732         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1733     }
1734 
1735     pef_kvm_reset(machine->cgs, &error_fatal);
1736     spapr_caps_apply(spapr);
1737 
1738     first_ppc_cpu = POWERPC_CPU(first_cpu);
1739     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1740         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1741                               spapr->max_compat_pvr)) {
1742         /*
1743          * If using KVM with radix mode available, VCPUs can be started
1744          * without a HPT because KVM will start them in radix mode.
1745          * Set the GR bit in PATE so that we know there is no HPT.
1746          */
1747         spapr->patb_entry = PATE1_GR;
1748         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1749     } else {
1750         spapr_setup_hpt(spapr);
1751     }
1752 
1753     qemu_devices_reset(reason);
1754 
1755     spapr_ovec_cleanup(spapr->ov5_cas);
1756     spapr->ov5_cas = spapr_ovec_new();
1757 
1758     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1759 
1760     /*
1761      * This is fixing some of the default configuration of the XIVE
1762      * devices. To be called after the reset of the machine devices.
1763      */
1764     spapr_irq_reset(spapr, &error_fatal);
1765 
1766     /*
1767      * There is no CAS under qtest. Simulate one to please the code that
1768      * depends on spapr->ov5_cas. This is especially needed to test device
1769      * unplug, so we do that before resetting the DRCs.
1770      */
1771     if (qtest_enabled()) {
1772         spapr_ovec_cleanup(spapr->ov5_cas);
1773         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1774     }
1775 
1776     spapr_nvdimm_finish_flushes();
1777 
1778     /* DRC reset may cause a device to be unplugged. This will cause troubles
1779      * if this device is used by another device (eg, a running vhost backend
1780      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1781      * situations, we reset DRCs after all devices have been reset.
1782      */
1783     spapr_drc_reset_all(spapr);
1784 
1785     spapr_clear_pending_events(spapr);
1786 
1787     /*
1788      * We place the device tree just below either the top of the RMA,
1789      * or just below 2GB, whichever is lower, so that it can be
1790      * processed with 32-bit real mode code if necessary
1791      */
1792     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1793 
1794     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1795     if (spapr->vof) {
1796         spapr_vof_reset(spapr, fdt, &error_fatal);
1797         /*
1798          * Do not pack the FDT as the client may change properties.
1799          * VOF client does not expect the FDT so we do not load it to the VM.
1800          */
1801     } else {
1802         rc = fdt_pack(fdt);
1803         /* Should only fail if we've built a corrupted tree */
1804         assert(rc == 0);
1805 
1806         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1807                                   0, fdt_addr, 0);
1808         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1809     }
1810     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1811 
1812     g_free(spapr->fdt_blob);
1813     spapr->fdt_size = fdt_totalsize(fdt);
1814     spapr->fdt_initial_size = spapr->fdt_size;
1815     spapr->fdt_blob = fdt;
1816 
1817     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1818     machine->fdt = fdt;
1819 
1820     /* Set up the entry state */
1821     first_ppc_cpu->env.gpr[5] = 0;
1822 
1823     spapr->fwnmi_system_reset_addr = -1;
1824     spapr->fwnmi_machine_check_addr = -1;
1825     spapr->fwnmi_machine_check_interlock = -1;
1826 
1827     /* Signal all vCPUs waiting on this condition */
1828     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1829 
1830     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1831 }
1832 
1833 static void spapr_create_nvram(SpaprMachineState *spapr)
1834 {
1835     DeviceState *dev = qdev_new("spapr-nvram");
1836     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1837 
1838     if (dinfo) {
1839         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1840                                 &error_fatal);
1841     }
1842 
1843     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1844 
1845     spapr->nvram = (struct SpaprNvram *)dev;
1846 }
1847 
1848 static void spapr_rtc_create(SpaprMachineState *spapr)
1849 {
1850     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1851                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1852                                        &error_fatal, NULL);
1853     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1854     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1855                               "date");
1856 }
1857 
1858 /* Returns whether we want to use VGA or not */
1859 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1860 {
1861     vga_interface_created = true;
1862     switch (vga_interface_type) {
1863     case VGA_NONE:
1864         return false;
1865     case VGA_DEVICE:
1866         return true;
1867     case VGA_STD:
1868     case VGA_VIRTIO:
1869     case VGA_CIRRUS:
1870         return pci_vga_init(pci_bus) != NULL;
1871     default:
1872         error_setg(errp,
1873                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1874         return false;
1875     }
1876 }
1877 
1878 static int spapr_pre_load(void *opaque)
1879 {
1880     int rc;
1881 
1882     rc = spapr_caps_pre_load(opaque);
1883     if (rc) {
1884         return rc;
1885     }
1886 
1887     return 0;
1888 }
1889 
1890 static int spapr_post_load(void *opaque, int version_id)
1891 {
1892     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1893     int err = 0;
1894 
1895     err = spapr_caps_post_migration(spapr);
1896     if (err) {
1897         return err;
1898     }
1899 
1900     /*
1901      * In earlier versions, there was no separate qdev for the PAPR
1902      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1903      * So when migrating from those versions, poke the incoming offset
1904      * value into the RTC device
1905      */
1906     if (version_id < 3) {
1907         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1908         if (err) {
1909             return err;
1910         }
1911     }
1912 
1913     if (kvm_enabled() && spapr->patb_entry) {
1914         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1915         bool radix = !!(spapr->patb_entry & PATE1_GR);
1916         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1917 
1918         /*
1919          * Update LPCR:HR and UPRT as they may not be set properly in
1920          * the stream
1921          */
1922         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1923                             LPCR_HR | LPCR_UPRT);
1924 
1925         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1926         if (err) {
1927             error_report("Process table config unsupported by the host");
1928             return -EINVAL;
1929         }
1930     }
1931 
1932     err = spapr_irq_post_load(spapr, version_id);
1933     if (err) {
1934         return err;
1935     }
1936 
1937     return err;
1938 }
1939 
1940 static int spapr_pre_save(void *opaque)
1941 {
1942     int rc;
1943 
1944     rc = spapr_caps_pre_save(opaque);
1945     if (rc) {
1946         return rc;
1947     }
1948 
1949     return 0;
1950 }
1951 
1952 static bool version_before_3(void *opaque, int version_id)
1953 {
1954     return version_id < 3;
1955 }
1956 
1957 static bool spapr_pending_events_needed(void *opaque)
1958 {
1959     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1960     return !QTAILQ_EMPTY(&spapr->pending_events);
1961 }
1962 
1963 static const VMStateDescription vmstate_spapr_event_entry = {
1964     .name = "spapr_event_log_entry",
1965     .version_id = 1,
1966     .minimum_version_id = 1,
1967     .fields = (const VMStateField[]) {
1968         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1969         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1970         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1971                                      NULL, extended_length),
1972         VMSTATE_END_OF_LIST()
1973     },
1974 };
1975 
1976 static const VMStateDescription vmstate_spapr_pending_events = {
1977     .name = "spapr_pending_events",
1978     .version_id = 1,
1979     .minimum_version_id = 1,
1980     .needed = spapr_pending_events_needed,
1981     .fields = (const VMStateField[]) {
1982         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1983                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1984         VMSTATE_END_OF_LIST()
1985     },
1986 };
1987 
1988 static bool spapr_ov5_cas_needed(void *opaque)
1989 {
1990     SpaprMachineState *spapr = opaque;
1991     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1992     bool cas_needed;
1993 
1994     /* Prior to the introduction of SpaprOptionVector, we had two option
1995      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1996      * Both of these options encode machine topology into the device-tree
1997      * in such a way that the now-booted OS should still be able to interact
1998      * appropriately with QEMU regardless of what options were actually
1999      * negotiatied on the source side.
2000      *
2001      * As such, we can avoid migrating the CAS-negotiated options if these
2002      * are the only options available on the current machine/platform.
2003      * Since these are the only options available for pseries-2.7 and
2004      * earlier, this allows us to maintain old->new/new->old migration
2005      * compatibility.
2006      *
2007      * For QEMU 2.8+, there are additional CAS-negotiatable options available
2008      * via default pseries-2.8 machines and explicit command-line parameters.
2009      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
2010      * of the actual CAS-negotiated values to continue working properly. For
2011      * example, availability of memory unplug depends on knowing whether
2012      * OV5_HP_EVT was negotiated via CAS.
2013      *
2014      * Thus, for any cases where the set of available CAS-negotiatable
2015      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2016      * include the CAS-negotiated options in the migration stream, unless
2017      * if they affect boot time behaviour only.
2018      */
2019     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2020     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2021     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2022 
2023     /* We need extra information if we have any bits outside the mask
2024      * defined above */
2025     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
2026 
2027     spapr_ovec_cleanup(ov5_mask);
2028 
2029     return cas_needed;
2030 }
2031 
2032 static const VMStateDescription vmstate_spapr_ov5_cas = {
2033     .name = "spapr_option_vector_ov5_cas",
2034     .version_id = 1,
2035     .minimum_version_id = 1,
2036     .needed = spapr_ov5_cas_needed,
2037     .fields = (const VMStateField[]) {
2038         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2039                                  vmstate_spapr_ovec, SpaprOptionVector),
2040         VMSTATE_END_OF_LIST()
2041     },
2042 };
2043 
2044 static bool spapr_patb_entry_needed(void *opaque)
2045 {
2046     SpaprMachineState *spapr = opaque;
2047 
2048     return !!spapr->patb_entry;
2049 }
2050 
2051 static const VMStateDescription vmstate_spapr_patb_entry = {
2052     .name = "spapr_patb_entry",
2053     .version_id = 1,
2054     .minimum_version_id = 1,
2055     .needed = spapr_patb_entry_needed,
2056     .fields = (const VMStateField[]) {
2057         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2058         VMSTATE_END_OF_LIST()
2059     },
2060 };
2061 
2062 static bool spapr_irq_map_needed(void *opaque)
2063 {
2064     SpaprMachineState *spapr = opaque;
2065 
2066     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2067 }
2068 
2069 static const VMStateDescription vmstate_spapr_irq_map = {
2070     .name = "spapr_irq_map",
2071     .version_id = 1,
2072     .minimum_version_id = 1,
2073     .needed = spapr_irq_map_needed,
2074     .fields = (const VMStateField[]) {
2075         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2076         VMSTATE_END_OF_LIST()
2077     },
2078 };
2079 
2080 static bool spapr_dtb_needed(void *opaque)
2081 {
2082     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2083 
2084     return smc->update_dt_enabled;
2085 }
2086 
2087 static int spapr_dtb_pre_load(void *opaque)
2088 {
2089     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2090 
2091     g_free(spapr->fdt_blob);
2092     spapr->fdt_blob = NULL;
2093     spapr->fdt_size = 0;
2094 
2095     return 0;
2096 }
2097 
2098 static const VMStateDescription vmstate_spapr_dtb = {
2099     .name = "spapr_dtb",
2100     .version_id = 1,
2101     .minimum_version_id = 1,
2102     .needed = spapr_dtb_needed,
2103     .pre_load = spapr_dtb_pre_load,
2104     .fields = (const VMStateField[]) {
2105         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2106         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2107         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2108                                      fdt_size),
2109         VMSTATE_END_OF_LIST()
2110     },
2111 };
2112 
2113 static bool spapr_fwnmi_needed(void *opaque)
2114 {
2115     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2116 
2117     return spapr->fwnmi_machine_check_addr != -1;
2118 }
2119 
2120 static int spapr_fwnmi_pre_save(void *opaque)
2121 {
2122     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2123 
2124     /*
2125      * Check if machine check handling is in progress and print a
2126      * warning message.
2127      */
2128     if (spapr->fwnmi_machine_check_interlock != -1) {
2129         warn_report("A machine check is being handled during migration. The"
2130                 "handler may run and log hardware error on the destination");
2131     }
2132 
2133     return 0;
2134 }
2135 
2136 static const VMStateDescription vmstate_spapr_fwnmi = {
2137     .name = "spapr_fwnmi",
2138     .version_id = 1,
2139     .minimum_version_id = 1,
2140     .needed = spapr_fwnmi_needed,
2141     .pre_save = spapr_fwnmi_pre_save,
2142     .fields = (const VMStateField[]) {
2143         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2144         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2145         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2146         VMSTATE_END_OF_LIST()
2147     },
2148 };
2149 
2150 static const VMStateDescription vmstate_spapr = {
2151     .name = "spapr",
2152     .version_id = 3,
2153     .minimum_version_id = 1,
2154     .pre_load = spapr_pre_load,
2155     .post_load = spapr_post_load,
2156     .pre_save = spapr_pre_save,
2157     .fields = (const VMStateField[]) {
2158         /* used to be @next_irq */
2159         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2160 
2161         /* RTC offset */
2162         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2163 
2164         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2165         VMSTATE_END_OF_LIST()
2166     },
2167     .subsections = (const VMStateDescription * const []) {
2168         &vmstate_spapr_ov5_cas,
2169         &vmstate_spapr_patb_entry,
2170         &vmstate_spapr_pending_events,
2171         &vmstate_spapr_cap_htm,
2172         &vmstate_spapr_cap_vsx,
2173         &vmstate_spapr_cap_dfp,
2174         &vmstate_spapr_cap_cfpc,
2175         &vmstate_spapr_cap_sbbc,
2176         &vmstate_spapr_cap_ibs,
2177         &vmstate_spapr_cap_hpt_maxpagesize,
2178         &vmstate_spapr_irq_map,
2179         &vmstate_spapr_cap_nested_kvm_hv,
2180         &vmstate_spapr_dtb,
2181         &vmstate_spapr_cap_large_decr,
2182         &vmstate_spapr_cap_ccf_assist,
2183         &vmstate_spapr_cap_fwnmi,
2184         &vmstate_spapr_fwnmi,
2185         &vmstate_spapr_cap_rpt_invalidate,
2186         NULL
2187     }
2188 };
2189 
2190 static int htab_save_setup(QEMUFile *f, void *opaque)
2191 {
2192     SpaprMachineState *spapr = opaque;
2193 
2194     /* "Iteration" header */
2195     if (!spapr->htab_shift) {
2196         qemu_put_be32(f, -1);
2197     } else {
2198         qemu_put_be32(f, spapr->htab_shift);
2199     }
2200 
2201     if (spapr->htab) {
2202         spapr->htab_save_index = 0;
2203         spapr->htab_first_pass = true;
2204     } else {
2205         if (spapr->htab_shift) {
2206             assert(kvm_enabled());
2207         }
2208     }
2209 
2210 
2211     return 0;
2212 }
2213 
2214 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2215                             int chunkstart, int n_valid, int n_invalid)
2216 {
2217     qemu_put_be32(f, chunkstart);
2218     qemu_put_be16(f, n_valid);
2219     qemu_put_be16(f, n_invalid);
2220     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2221                     HASH_PTE_SIZE_64 * n_valid);
2222 }
2223 
2224 static void htab_save_end_marker(QEMUFile *f)
2225 {
2226     qemu_put_be32(f, 0);
2227     qemu_put_be16(f, 0);
2228     qemu_put_be16(f, 0);
2229 }
2230 
2231 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2232                                  int64_t max_ns)
2233 {
2234     bool has_timeout = max_ns != -1;
2235     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2236     int index = spapr->htab_save_index;
2237     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2238 
2239     assert(spapr->htab_first_pass);
2240 
2241     do {
2242         int chunkstart;
2243 
2244         /* Consume invalid HPTEs */
2245         while ((index < htabslots)
2246                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2247             CLEAN_HPTE(HPTE(spapr->htab, index));
2248             index++;
2249         }
2250 
2251         /* Consume valid HPTEs */
2252         chunkstart = index;
2253         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2254                && HPTE_VALID(HPTE(spapr->htab, index))) {
2255             CLEAN_HPTE(HPTE(spapr->htab, index));
2256             index++;
2257         }
2258 
2259         if (index > chunkstart) {
2260             int n_valid = index - chunkstart;
2261 
2262             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2263 
2264             if (has_timeout &&
2265                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2266                 break;
2267             }
2268         }
2269     } while ((index < htabslots) && !migration_rate_exceeded(f));
2270 
2271     if (index >= htabslots) {
2272         assert(index == htabslots);
2273         index = 0;
2274         spapr->htab_first_pass = false;
2275     }
2276     spapr->htab_save_index = index;
2277 }
2278 
2279 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2280                                 int64_t max_ns)
2281 {
2282     bool final = max_ns < 0;
2283     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2284     int examined = 0, sent = 0;
2285     int index = spapr->htab_save_index;
2286     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2287 
2288     assert(!spapr->htab_first_pass);
2289 
2290     do {
2291         int chunkstart, invalidstart;
2292 
2293         /* Consume non-dirty HPTEs */
2294         while ((index < htabslots)
2295                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2296             index++;
2297             examined++;
2298         }
2299 
2300         chunkstart = index;
2301         /* Consume valid dirty HPTEs */
2302         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2303                && HPTE_DIRTY(HPTE(spapr->htab, index))
2304                && HPTE_VALID(HPTE(spapr->htab, index))) {
2305             CLEAN_HPTE(HPTE(spapr->htab, index));
2306             index++;
2307             examined++;
2308         }
2309 
2310         invalidstart = index;
2311         /* Consume invalid dirty HPTEs */
2312         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2313                && HPTE_DIRTY(HPTE(spapr->htab, index))
2314                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2315             CLEAN_HPTE(HPTE(spapr->htab, index));
2316             index++;
2317             examined++;
2318         }
2319 
2320         if (index > chunkstart) {
2321             int n_valid = invalidstart - chunkstart;
2322             int n_invalid = index - invalidstart;
2323 
2324             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2325             sent += index - chunkstart;
2326 
2327             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2328                 break;
2329             }
2330         }
2331 
2332         if (examined >= htabslots) {
2333             break;
2334         }
2335 
2336         if (index >= htabslots) {
2337             assert(index == htabslots);
2338             index = 0;
2339         }
2340     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2341 
2342     if (index >= htabslots) {
2343         assert(index == htabslots);
2344         index = 0;
2345     }
2346 
2347     spapr->htab_save_index = index;
2348 
2349     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2350 }
2351 
2352 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2353 #define MAX_KVM_BUF_SIZE    2048
2354 
2355 static int htab_save_iterate(QEMUFile *f, void *opaque)
2356 {
2357     SpaprMachineState *spapr = opaque;
2358     int fd;
2359     int rc = 0;
2360 
2361     /* Iteration header */
2362     if (!spapr->htab_shift) {
2363         qemu_put_be32(f, -1);
2364         return 1;
2365     } else {
2366         qemu_put_be32(f, 0);
2367     }
2368 
2369     if (!spapr->htab) {
2370         assert(kvm_enabled());
2371 
2372         fd = get_htab_fd(spapr);
2373         if (fd < 0) {
2374             return fd;
2375         }
2376 
2377         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2378         if (rc < 0) {
2379             return rc;
2380         }
2381     } else  if (spapr->htab_first_pass) {
2382         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2383     } else {
2384         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2385     }
2386 
2387     htab_save_end_marker(f);
2388 
2389     return rc;
2390 }
2391 
2392 static int htab_save_complete(QEMUFile *f, void *opaque)
2393 {
2394     SpaprMachineState *spapr = opaque;
2395     int fd;
2396 
2397     /* Iteration header */
2398     if (!spapr->htab_shift) {
2399         qemu_put_be32(f, -1);
2400         return 0;
2401     } else {
2402         qemu_put_be32(f, 0);
2403     }
2404 
2405     if (!spapr->htab) {
2406         int rc;
2407 
2408         assert(kvm_enabled());
2409 
2410         fd = get_htab_fd(spapr);
2411         if (fd < 0) {
2412             return fd;
2413         }
2414 
2415         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2416         if (rc < 0) {
2417             return rc;
2418         }
2419     } else {
2420         if (spapr->htab_first_pass) {
2421             htab_save_first_pass(f, spapr, -1);
2422         }
2423         htab_save_later_pass(f, spapr, -1);
2424     }
2425 
2426     /* End marker */
2427     htab_save_end_marker(f);
2428 
2429     return 0;
2430 }
2431 
2432 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2433 {
2434     SpaprMachineState *spapr = opaque;
2435     uint32_t section_hdr;
2436     int fd = -1;
2437     Error *local_err = NULL;
2438 
2439     if (version_id < 1 || version_id > 1) {
2440         error_report("htab_load() bad version");
2441         return -EINVAL;
2442     }
2443 
2444     section_hdr = qemu_get_be32(f);
2445 
2446     if (section_hdr == -1) {
2447         spapr_free_hpt(spapr);
2448         return 0;
2449     }
2450 
2451     if (section_hdr) {
2452         int ret;
2453 
2454         /* First section gives the htab size */
2455         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2456         if (ret < 0) {
2457             error_report_err(local_err);
2458             return ret;
2459         }
2460         return 0;
2461     }
2462 
2463     if (!spapr->htab) {
2464         assert(kvm_enabled());
2465 
2466         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2467         if (fd < 0) {
2468             error_report_err(local_err);
2469             return fd;
2470         }
2471     }
2472 
2473     while (true) {
2474         uint32_t index;
2475         uint16_t n_valid, n_invalid;
2476 
2477         index = qemu_get_be32(f);
2478         n_valid = qemu_get_be16(f);
2479         n_invalid = qemu_get_be16(f);
2480 
2481         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2482             /* End of Stream */
2483             break;
2484         }
2485 
2486         if ((index + n_valid + n_invalid) >
2487             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2488             /* Bad index in stream */
2489             error_report(
2490                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2491                 index, n_valid, n_invalid, spapr->htab_shift);
2492             return -EINVAL;
2493         }
2494 
2495         if (spapr->htab) {
2496             if (n_valid) {
2497                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2498                                 HASH_PTE_SIZE_64 * n_valid);
2499             }
2500             if (n_invalid) {
2501                 memset(HPTE(spapr->htab, index + n_valid), 0,
2502                        HASH_PTE_SIZE_64 * n_invalid);
2503             }
2504         } else {
2505             int rc;
2506 
2507             assert(fd >= 0);
2508 
2509             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2510                                         &local_err);
2511             if (rc < 0) {
2512                 error_report_err(local_err);
2513                 return rc;
2514             }
2515         }
2516     }
2517 
2518     if (!spapr->htab) {
2519         assert(fd >= 0);
2520         close(fd);
2521     }
2522 
2523     return 0;
2524 }
2525 
2526 static void htab_save_cleanup(void *opaque)
2527 {
2528     SpaprMachineState *spapr = opaque;
2529 
2530     close_htab_fd(spapr);
2531 }
2532 
2533 static SaveVMHandlers savevm_htab_handlers = {
2534     .save_setup = htab_save_setup,
2535     .save_live_iterate = htab_save_iterate,
2536     .save_live_complete_precopy = htab_save_complete,
2537     .save_cleanup = htab_save_cleanup,
2538     .load_state = htab_load,
2539 };
2540 
2541 static void spapr_boot_set(void *opaque, const char *boot_device,
2542                            Error **errp)
2543 {
2544     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2545 
2546     g_free(spapr->boot_device);
2547     spapr->boot_device = g_strdup(boot_device);
2548 }
2549 
2550 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2551 {
2552     MachineState *machine = MACHINE(spapr);
2553     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2554     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2555     int i;
2556 
2557     g_assert(!nr_lmbs || machine->device_memory);
2558     for (i = 0; i < nr_lmbs; i++) {
2559         uint64_t addr;
2560 
2561         addr = i * lmb_size + machine->device_memory->base;
2562         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2563                                addr / lmb_size);
2564     }
2565 }
2566 
2567 /*
2568  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2569  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2570  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2571  */
2572 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2573 {
2574     int i;
2575 
2576     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2577         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2578                    " is not aligned to %" PRIu64 " MiB",
2579                    machine->ram_size,
2580                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2581         return;
2582     }
2583 
2584     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2585         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2586                    " is not aligned to %" PRIu64 " MiB",
2587                    machine->ram_size,
2588                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2589         return;
2590     }
2591 
2592     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2593         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2594             error_setg(errp,
2595                        "Node %d memory size 0x%" PRIx64
2596                        " is not aligned to %" PRIu64 " MiB",
2597                        i, machine->numa_state->nodes[i].node_mem,
2598                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2599             return;
2600         }
2601     }
2602 }
2603 
2604 /* find cpu slot in machine->possible_cpus by core_id */
2605 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2606 {
2607     int index = id / ms->smp.threads;
2608 
2609     if (index >= ms->possible_cpus->len) {
2610         return NULL;
2611     }
2612     if (idx) {
2613         *idx = index;
2614     }
2615     return &ms->possible_cpus->cpus[index];
2616 }
2617 
2618 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2619 {
2620     MachineState *ms = MACHINE(spapr);
2621     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2622     Error *local_err = NULL;
2623     bool vsmt_user = !!spapr->vsmt;
2624     int kvm_smt = kvmppc_smt_threads();
2625     int ret;
2626     unsigned int smp_threads = ms->smp.threads;
2627 
2628     if (tcg_enabled()) {
2629         if (smp_threads > 1 &&
2630             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2631                                    spapr->max_compat_pvr)) {
2632             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2633             return;
2634         }
2635 
2636         if (smp_threads > 8) {
2637             error_setg(errp, "TCG cannot support more than 8 threads/core "
2638                        "on a pseries machine");
2639             return;
2640         }
2641     }
2642     if (!is_power_of_2(smp_threads)) {
2643         error_setg(errp, "Cannot support %d threads/core on a pseries "
2644                    "machine because it must be a power of 2", smp_threads);
2645         return;
2646     }
2647 
2648     /* Determine the VSMT mode to use: */
2649     if (vsmt_user) {
2650         if (spapr->vsmt < smp_threads) {
2651             error_setg(errp, "Cannot support VSMT mode %d"
2652                        " because it must be >= threads/core (%d)",
2653                        spapr->vsmt, smp_threads);
2654             return;
2655         }
2656         /* In this case, spapr->vsmt has been set by the command line */
2657     } else if (!smc->smp_threads_vsmt) {
2658         /*
2659          * Default VSMT value is tricky, because we need it to be as
2660          * consistent as possible (for migration), but this requires
2661          * changing it for at least some existing cases.  We pick 8 as
2662          * the value that we'd get with KVM on POWER8, the
2663          * overwhelmingly common case in production systems.
2664          */
2665         spapr->vsmt = MAX(8, smp_threads);
2666     } else {
2667         spapr->vsmt = smp_threads;
2668     }
2669 
2670     /* KVM: If necessary, set the SMT mode: */
2671     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2672         ret = kvmppc_set_smt_threads(spapr->vsmt);
2673         if (ret) {
2674             /* Looks like KVM isn't able to change VSMT mode */
2675             error_setg(&local_err,
2676                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2677                        spapr->vsmt, ret);
2678             /* We can live with that if the default one is big enough
2679              * for the number of threads, and a submultiple of the one
2680              * we want.  In this case we'll waste some vcpu ids, but
2681              * behaviour will be correct */
2682             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2683                 warn_report_err(local_err);
2684             } else {
2685                 if (!vsmt_user) {
2686                     error_append_hint(&local_err,
2687                                       "On PPC, a VM with %d threads/core"
2688                                       " on a host with %d threads/core"
2689                                       " requires the use of VSMT mode %d.\n",
2690                                       smp_threads, kvm_smt, spapr->vsmt);
2691                 }
2692                 kvmppc_error_append_smt_possible_hint(&local_err);
2693                 error_propagate(errp, local_err);
2694             }
2695         }
2696     }
2697     /* else TCG: nothing to do currently */
2698 }
2699 
2700 static void spapr_init_cpus(SpaprMachineState *spapr)
2701 {
2702     MachineState *machine = MACHINE(spapr);
2703     MachineClass *mc = MACHINE_GET_CLASS(machine);
2704     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2705     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2706     const CPUArchIdList *possible_cpus;
2707     unsigned int smp_cpus = machine->smp.cpus;
2708     unsigned int smp_threads = machine->smp.threads;
2709     unsigned int max_cpus = machine->smp.max_cpus;
2710     int boot_cores_nr = smp_cpus / smp_threads;
2711     int i;
2712 
2713     possible_cpus = mc->possible_cpu_arch_ids(machine);
2714     if (mc->has_hotpluggable_cpus) {
2715         if (smp_cpus % smp_threads) {
2716             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2717                          smp_cpus, smp_threads);
2718             exit(1);
2719         }
2720         if (max_cpus % smp_threads) {
2721             error_report("max_cpus (%u) must be multiple of threads (%u)",
2722                          max_cpus, smp_threads);
2723             exit(1);
2724         }
2725     } else {
2726         if (max_cpus != smp_cpus) {
2727             error_report("This machine version does not support CPU hotplug");
2728             exit(1);
2729         }
2730         boot_cores_nr = possible_cpus->len;
2731     }
2732 
2733     if (smc->pre_2_10_has_unused_icps) {
2734         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2735             /* Dummy entries get deregistered when real ICPState objects
2736              * are registered during CPU core hotplug.
2737              */
2738             pre_2_10_vmstate_register_dummy_icp(i);
2739         }
2740     }
2741 
2742     for (i = 0; i < possible_cpus->len; i++) {
2743         int core_id = i * smp_threads;
2744 
2745         if (mc->has_hotpluggable_cpus) {
2746             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2747                                    spapr_vcpu_id(spapr, core_id));
2748         }
2749 
2750         if (i < boot_cores_nr) {
2751             Object *core  = object_new(type);
2752             int nr_threads = smp_threads;
2753 
2754             /* Handle the partially filled core for older machine types */
2755             if ((i + 1) * smp_threads >= smp_cpus) {
2756                 nr_threads = smp_cpus - i * smp_threads;
2757             }
2758 
2759             object_property_set_int(core, "nr-threads", nr_threads,
2760                                     &error_fatal);
2761             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2762                                     &error_fatal);
2763             qdev_realize(DEVICE(core), NULL, &error_fatal);
2764 
2765             object_unref(core);
2766         }
2767     }
2768 }
2769 
2770 static PCIHostState *spapr_create_default_phb(void)
2771 {
2772     DeviceState *dev;
2773 
2774     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2775     qdev_prop_set_uint32(dev, "index", 0);
2776     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2777 
2778     return PCI_HOST_BRIDGE(dev);
2779 }
2780 
2781 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2782 {
2783     MachineState *machine = MACHINE(spapr);
2784     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2785     hwaddr rma_size = machine->ram_size;
2786     hwaddr node0_size = spapr_node0_size(machine);
2787 
2788     /* RMA has to fit in the first NUMA node */
2789     rma_size = MIN(rma_size, node0_size);
2790 
2791     /*
2792      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2793      * never exceed that
2794      */
2795     rma_size = MIN(rma_size, 1 * TiB);
2796 
2797     /*
2798      * Clamp the RMA size based on machine type.  This is for
2799      * migration compatibility with older qemu versions, which limited
2800      * the RMA size for complicated and mostly bad reasons.
2801      */
2802     if (smc->rma_limit) {
2803         rma_size = MIN(rma_size, smc->rma_limit);
2804     }
2805 
2806     if (rma_size < MIN_RMA_SLOF) {
2807         error_setg(errp,
2808                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2809                    "ldMiB guest RMA (Real Mode Area memory)",
2810                    MIN_RMA_SLOF / MiB);
2811         return 0;
2812     }
2813 
2814     return rma_size;
2815 }
2816 
2817 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2818 {
2819     MachineState *machine = MACHINE(spapr);
2820     int i;
2821 
2822     for (i = 0; i < machine->ram_slots; i++) {
2823         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2824     }
2825 }
2826 
2827 /* pSeries LPAR / sPAPR hardware init */
2828 static void spapr_machine_init(MachineState *machine)
2829 {
2830     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2831     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2832     MachineClass *mc = MACHINE_GET_CLASS(machine);
2833     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2834     const char *bios_name = machine->firmware ?: bios_default;
2835     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2836     const char *kernel_filename = machine->kernel_filename;
2837     const char *initrd_filename = machine->initrd_filename;
2838     PCIHostState *phb;
2839     bool has_vga;
2840     int i;
2841     MemoryRegion *sysmem = get_system_memory();
2842     long load_limit, fw_size;
2843     Error *resize_hpt_err = NULL;
2844     NICInfo *nd;
2845 
2846     if (!filename) {
2847         error_report("Could not find LPAR firmware '%s'", bios_name);
2848         exit(1);
2849     }
2850     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2851     if (fw_size <= 0) {
2852         error_report("Could not load LPAR firmware '%s'", filename);
2853         exit(1);
2854     }
2855 
2856     /*
2857      * if Secure VM (PEF) support is configured, then initialize it
2858      */
2859     pef_kvm_init(machine->cgs, &error_fatal);
2860 
2861     msi_nonbroken = true;
2862 
2863     QLIST_INIT(&spapr->phbs);
2864     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2865 
2866     /* Determine capabilities to run with */
2867     spapr_caps_init(spapr);
2868 
2869     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2870     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2871         /*
2872          * If the user explicitly requested a mode we should either
2873          * supply it, or fail completely (which we do below).  But if
2874          * it's not set explicitly, we reset our mode to something
2875          * that works
2876          */
2877         if (resize_hpt_err) {
2878             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2879             error_free(resize_hpt_err);
2880             resize_hpt_err = NULL;
2881         } else {
2882             spapr->resize_hpt = smc->resize_hpt_default;
2883         }
2884     }
2885 
2886     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2887 
2888     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2889         /*
2890          * User requested HPT resize, but this host can't supply it.  Bail out
2891          */
2892         error_report_err(resize_hpt_err);
2893         exit(1);
2894     }
2895     error_free(resize_hpt_err);
2896 
2897     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2898 
2899     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2900     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2901 
2902     /*
2903      * VSMT must be set in order to be able to compute VCPU ids, ie to
2904      * call spapr_max_server_number() or spapr_vcpu_id().
2905      */
2906     spapr_set_vsmt_mode(spapr, &error_fatal);
2907 
2908     /* Set up Interrupt Controller before we create the VCPUs */
2909     spapr_irq_init(spapr, &error_fatal);
2910 
2911     /* Set up containers for ibm,client-architecture-support negotiated options
2912      */
2913     spapr->ov5 = spapr_ovec_new();
2914     spapr->ov5_cas = spapr_ovec_new();
2915 
2916     if (smc->dr_lmb_enabled) {
2917         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2918         spapr_validate_node_memory(machine, &error_fatal);
2919     }
2920 
2921     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2922 
2923     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2924     if (!smc->pre_6_2_numa_affinity) {
2925         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2926     }
2927 
2928     /* advertise support for dedicated HP event source to guests */
2929     if (spapr->use_hotplug_event_source) {
2930         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2931     }
2932 
2933     /* advertise support for HPT resizing */
2934     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2935         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2936     }
2937 
2938     /* advertise support for ibm,dyamic-memory-v2 */
2939     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2940 
2941     /* advertise XIVE on POWER9 machines */
2942     if (spapr->irq->xive) {
2943         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2944     }
2945 
2946     /* init CPUs */
2947     spapr_init_cpus(spapr);
2948 
2949     /* Init numa_assoc_array */
2950     spapr_numa_associativity_init(spapr, machine);
2951 
2952     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2953         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2954                               spapr->max_compat_pvr)) {
2955         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2956         /* KVM and TCG always allow GTSE with radix... */
2957         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2958     }
2959     /* ... but not with hash (currently). */
2960 
2961     if (kvm_enabled()) {
2962         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2963         kvmppc_enable_logical_ci_hcalls();
2964         kvmppc_enable_set_mode_hcall();
2965 
2966         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2967         kvmppc_enable_clear_ref_mod_hcalls();
2968 
2969         /* Enable H_PAGE_INIT */
2970         kvmppc_enable_h_page_init();
2971     }
2972 
2973     /* map RAM */
2974     memory_region_add_subregion(sysmem, 0, machine->ram);
2975 
2976     /* initialize hotplug memory address space */
2977     if (machine->ram_size < machine->maxram_size) {
2978         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2979         hwaddr device_mem_base;
2980 
2981         /*
2982          * Limit the number of hotpluggable memory slots to half the number
2983          * slots that KVM supports, leaving the other half for PCI and other
2984          * devices. However ensure that number of slots doesn't drop below 32.
2985          */
2986         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2987                            SPAPR_MAX_RAM_SLOTS;
2988 
2989         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2990             max_memslots = SPAPR_MAX_RAM_SLOTS;
2991         }
2992         if (machine->ram_slots > max_memslots) {
2993             error_report("Specified number of memory slots %"
2994                          PRIu64" exceeds max supported %d",
2995                          machine->ram_slots, max_memslots);
2996             exit(1);
2997         }
2998 
2999         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
3000         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
3001     }
3002 
3003     if (smc->dr_lmb_enabled) {
3004         spapr_create_lmb_dr_connectors(spapr);
3005     }
3006 
3007     if (mc->nvdimm_supported) {
3008         spapr_create_nvdimm_dr_connectors(spapr);
3009     }
3010 
3011     /* Set up RTAS event infrastructure */
3012     spapr_events_init(spapr);
3013 
3014     /* Set up the RTC RTAS interfaces */
3015     spapr_rtc_create(spapr);
3016 
3017     /* Set up VIO bus */
3018     spapr->vio_bus = spapr_vio_bus_init();
3019 
3020     for (i = 0; serial_hd(i); i++) {
3021         spapr_vty_create(spapr->vio_bus, serial_hd(i));
3022     }
3023 
3024     /* We always have at least the nvram device on VIO */
3025     spapr_create_nvram(spapr);
3026 
3027     /*
3028      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3029      * connectors (described in root DT node's "ibm,drc-types" property)
3030      * are pre-initialized here. additional child connectors (such as
3031      * connectors for a PHBs PCI slots) are added as needed during their
3032      * parent's realization.
3033      */
3034     if (smc->dr_phb_enabled) {
3035         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3036             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3037         }
3038     }
3039 
3040     /* Set up PCI */
3041     spapr_pci_rtas_init();
3042 
3043     phb = spapr_create_default_phb();
3044 
3045     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3046         spapr_vlan_create(spapr->vio_bus, nd);
3047     }
3048 
3049     pci_init_nic_devices(phb->bus, NULL);
3050 
3051     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3052         spapr_vscsi_create(spapr->vio_bus);
3053     }
3054 
3055     /* Graphics */
3056     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3057     if (has_vga) {
3058         spapr->want_stdout_path = !machine->enable_graphics;
3059         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3060     } else {
3061         spapr->want_stdout_path = true;
3062     }
3063 
3064     if (machine->usb) {
3065         if (smc->use_ohci_by_default) {
3066             pci_create_simple(phb->bus, -1, "pci-ohci");
3067         } else {
3068             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3069         }
3070 
3071         if (has_vga) {
3072             USBBus *usb_bus;
3073 
3074             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3075                                                               &error_abort));
3076             usb_create_simple(usb_bus, "usb-kbd");
3077             usb_create_simple(usb_bus, "usb-mouse");
3078         }
3079     }
3080 
3081     if (kernel_filename) {
3082         uint64_t loaded_addr = 0;
3083 
3084         spapr->kernel_size = load_elf(kernel_filename, NULL,
3085                                       translate_kernel_address, spapr,
3086                                       NULL, &loaded_addr, NULL, NULL, 1,
3087                                       PPC_ELF_MACHINE, 0, 0);
3088         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3089             spapr->kernel_size = load_elf(kernel_filename, NULL,
3090                                           translate_kernel_address, spapr,
3091                                           NULL, &loaded_addr, NULL, NULL, 0,
3092                                           PPC_ELF_MACHINE, 0, 0);
3093             spapr->kernel_le = spapr->kernel_size > 0;
3094         }
3095         if (spapr->kernel_size < 0) {
3096             error_report("error loading %s: %s", kernel_filename,
3097                          load_elf_strerror(spapr->kernel_size));
3098             exit(1);
3099         }
3100 
3101         if (spapr->kernel_addr != loaded_addr) {
3102             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3103                         " to 0x%"PRIx64,
3104                         spapr->kernel_addr, loaded_addr);
3105             spapr->kernel_addr = loaded_addr;
3106         }
3107 
3108         /* load initrd */
3109         if (initrd_filename) {
3110             /* Try to locate the initrd in the gap between the kernel
3111              * and the firmware. Add a bit of space just in case
3112              */
3113             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3114                                   + 0x1ffff) & ~0xffff;
3115             spapr->initrd_size = load_image_targphys(initrd_filename,
3116                                                      spapr->initrd_base,
3117                                                      load_limit
3118                                                      - spapr->initrd_base);
3119             if (spapr->initrd_size < 0) {
3120                 error_report("could not load initial ram disk '%s'",
3121                              initrd_filename);
3122                 exit(1);
3123             }
3124         }
3125     }
3126 
3127     /* FIXME: Should register things through the MachineState's qdev
3128      * interface, this is a legacy from the sPAPREnvironment structure
3129      * which predated MachineState but had a similar function */
3130     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3131     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3132                          &savevm_htab_handlers, spapr);
3133 
3134     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3135 
3136     qemu_register_boot_set(spapr_boot_set, spapr);
3137 
3138     /*
3139      * Nothing needs to be done to resume a suspended guest because
3140      * suspending does not change the machine state, so no need for
3141      * a ->wakeup method.
3142      */
3143     qemu_register_wakeup_support();
3144 
3145     if (kvm_enabled()) {
3146         /* to stop and start vmclock */
3147         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3148                                          &spapr->tb);
3149 
3150         kvmppc_spapr_enable_inkernel_multitce();
3151     }
3152 
3153     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3154     if (spapr->vof) {
3155         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3156         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3157     }
3158 
3159     spapr_watchdog_init(spapr);
3160 }
3161 
3162 #define DEFAULT_KVM_TYPE "auto"
3163 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3164 {
3165     /*
3166      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3167      * accommodate the 'HV' and 'PV' formats that exists in the
3168      * wild. The 'auto' mode is being introduced already as
3169      * lower-case, thus we don't need to bother checking for
3170      * "AUTO".
3171      */
3172     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3173         return 0;
3174     }
3175 
3176     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3177         return 1;
3178     }
3179 
3180     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3181         return 2;
3182     }
3183 
3184     error_report("Unknown kvm-type specified '%s'", vm_type);
3185     return -1;
3186 }
3187 
3188 /*
3189  * Implementation of an interface to adjust firmware path
3190  * for the bootindex property handling.
3191  */
3192 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3193                                    DeviceState *dev)
3194 {
3195 #define CAST(type, obj, name) \
3196     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3197     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3198     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3199     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3200     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3201 
3202     if (d && bus) {
3203         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3204         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3205         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3206 
3207         if (spapr) {
3208             /*
3209              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3210              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3211              * 0x8000 | (target << 8) | (bus << 5) | lun
3212              * (see the "Logical unit addressing format" table in SAM5)
3213              */
3214             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3215             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3216                                    (uint64_t)id << 48);
3217         } else if (virtio) {
3218             /*
3219              * We use SRP luns of the form 01000000 | (target << 8) | lun
3220              * in the top 32 bits of the 64-bit LUN
3221              * Note: the quote above is from SLOF and it is wrong,
3222              * the actual binding is:
3223              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3224              */
3225             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3226             if (d->lun >= 256) {
3227                 /* Use the LUN "flat space addressing method" */
3228                 id |= 0x4000;
3229             }
3230             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3231                                    (uint64_t)id << 32);
3232         } else if (usb) {
3233             /*
3234              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3235              * in the top 32 bits of the 64-bit LUN
3236              */
3237             unsigned usb_port = atoi(usb->port->path);
3238             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3239             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3240                                    (uint64_t)id << 32);
3241         }
3242     }
3243 
3244     /*
3245      * SLOF probes the USB devices, and if it recognizes that the device is a
3246      * storage device, it changes its name to "storage" instead of "usb-host",
3247      * and additionally adds a child node for the SCSI LUN, so the correct
3248      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3249      */
3250     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3251         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3252         if (usb_device_is_scsi_storage(usbdev)) {
3253             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3254         }
3255     }
3256 
3257     if (phb) {
3258         /* Replace "pci" with "pci@800000020000000" */
3259         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3260     }
3261 
3262     if (vsc) {
3263         /* Same logic as virtio above */
3264         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3265         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3266     }
3267 
3268     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3269         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3270         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3271         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3272     }
3273 
3274     if (pcidev) {
3275         return spapr_pci_fw_dev_name(pcidev);
3276     }
3277 
3278     return NULL;
3279 }
3280 
3281 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3282 {
3283     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3284 
3285     return g_strdup(spapr->kvm_type);
3286 }
3287 
3288 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3289 {
3290     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3291 
3292     g_free(spapr->kvm_type);
3293     spapr->kvm_type = g_strdup(value);
3294 }
3295 
3296 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3297 {
3298     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3299 
3300     return spapr->use_hotplug_event_source;
3301 }
3302 
3303 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3304                                             Error **errp)
3305 {
3306     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3307 
3308     spapr->use_hotplug_event_source = value;
3309 }
3310 
3311 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3312 {
3313     return true;
3314 }
3315 
3316 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3317 {
3318     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3319 
3320     switch (spapr->resize_hpt) {
3321     case SPAPR_RESIZE_HPT_DEFAULT:
3322         return g_strdup("default");
3323     case SPAPR_RESIZE_HPT_DISABLED:
3324         return g_strdup("disabled");
3325     case SPAPR_RESIZE_HPT_ENABLED:
3326         return g_strdup("enabled");
3327     case SPAPR_RESIZE_HPT_REQUIRED:
3328         return g_strdup("required");
3329     }
3330     g_assert_not_reached();
3331 }
3332 
3333 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3334 {
3335     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3336 
3337     if (strcmp(value, "default") == 0) {
3338         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3339     } else if (strcmp(value, "disabled") == 0) {
3340         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3341     } else if (strcmp(value, "enabled") == 0) {
3342         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3343     } else if (strcmp(value, "required") == 0) {
3344         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3345     } else {
3346         error_setg(errp, "Bad value for \"resize-hpt\" property");
3347     }
3348 }
3349 
3350 static bool spapr_get_vof(Object *obj, Error **errp)
3351 {
3352     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3353 
3354     return spapr->vof != NULL;
3355 }
3356 
3357 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3358 {
3359     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3360 
3361     if (spapr->vof) {
3362         vof_cleanup(spapr->vof);
3363         g_free(spapr->vof);
3364         spapr->vof = NULL;
3365     }
3366     if (!value) {
3367         return;
3368     }
3369     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3370 }
3371 
3372 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3373 {
3374     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3375 
3376     if (spapr->irq == &spapr_irq_xics_legacy) {
3377         return g_strdup("legacy");
3378     } else if (spapr->irq == &spapr_irq_xics) {
3379         return g_strdup("xics");
3380     } else if (spapr->irq == &spapr_irq_xive) {
3381         return g_strdup("xive");
3382     } else if (spapr->irq == &spapr_irq_dual) {
3383         return g_strdup("dual");
3384     }
3385     g_assert_not_reached();
3386 }
3387 
3388 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3389 {
3390     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3391 
3392     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3393         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3394         return;
3395     }
3396 
3397     /* The legacy IRQ backend can not be set */
3398     if (strcmp(value, "xics") == 0) {
3399         spapr->irq = &spapr_irq_xics;
3400     } else if (strcmp(value, "xive") == 0) {
3401         spapr->irq = &spapr_irq_xive;
3402     } else if (strcmp(value, "dual") == 0) {
3403         spapr->irq = &spapr_irq_dual;
3404     } else {
3405         error_setg(errp, "Bad value for \"ic-mode\" property");
3406     }
3407 }
3408 
3409 static char *spapr_get_host_model(Object *obj, Error **errp)
3410 {
3411     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3412 
3413     return g_strdup(spapr->host_model);
3414 }
3415 
3416 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3417 {
3418     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3419 
3420     g_free(spapr->host_model);
3421     spapr->host_model = g_strdup(value);
3422 }
3423 
3424 static char *spapr_get_host_serial(Object *obj, Error **errp)
3425 {
3426     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3427 
3428     return g_strdup(spapr->host_serial);
3429 }
3430 
3431 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3432 {
3433     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3434 
3435     g_free(spapr->host_serial);
3436     spapr->host_serial = g_strdup(value);
3437 }
3438 
3439 static void spapr_instance_init(Object *obj)
3440 {
3441     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3442     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3443     MachineState *ms = MACHINE(spapr);
3444     MachineClass *mc = MACHINE_GET_CLASS(ms);
3445 
3446     /*
3447      * NVDIMM support went live in 5.1 without considering that, in
3448      * other archs, the user needs to enable NVDIMM support with the
3449      * 'nvdimm' machine option and the default behavior is NVDIMM
3450      * support disabled. It is too late to roll back to the standard
3451      * behavior without breaking 5.1 guests.
3452      */
3453     if (mc->nvdimm_supported) {
3454         ms->nvdimms_state->is_enabled = true;
3455     }
3456 
3457     spapr->htab_fd = -1;
3458     spapr->use_hotplug_event_source = true;
3459     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3460     object_property_add_str(obj, "kvm-type",
3461                             spapr_get_kvm_type, spapr_set_kvm_type);
3462     object_property_set_description(obj, "kvm-type",
3463                                     "Specifies the KVM virtualization mode (auto,"
3464                                     " hv, pr). Defaults to 'auto'. This mode will use"
3465                                     " any available KVM module loaded in the host,"
3466                                     " where kvm_hv takes precedence if both kvm_hv and"
3467                                     " kvm_pr are loaded.");
3468     object_property_add_bool(obj, "modern-hotplug-events",
3469                             spapr_get_modern_hotplug_events,
3470                             spapr_set_modern_hotplug_events);
3471     object_property_set_description(obj, "modern-hotplug-events",
3472                                     "Use dedicated hotplug event mechanism in"
3473                                     " place of standard EPOW events when possible"
3474                                     " (required for memory hot-unplug support)");
3475     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3476                             "Maximum permitted CPU compatibility mode");
3477 
3478     object_property_add_str(obj, "resize-hpt",
3479                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3480     object_property_set_description(obj, "resize-hpt",
3481                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3482     object_property_add_uint32_ptr(obj, "vsmt",
3483                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3484     object_property_set_description(obj, "vsmt",
3485                                     "Virtual SMT: KVM behaves as if this were"
3486                                     " the host's SMT mode");
3487 
3488     object_property_add_bool(obj, "vfio-no-msix-emulation",
3489                              spapr_get_msix_emulation, NULL);
3490 
3491     object_property_add_uint64_ptr(obj, "kernel-addr",
3492                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3493     object_property_set_description(obj, "kernel-addr",
3494                                     stringify(KERNEL_LOAD_ADDR)
3495                                     " for -kernel is the default");
3496     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3497 
3498     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3499     object_property_set_description(obj, "x-vof",
3500                                     "Enable Virtual Open Firmware (experimental)");
3501 
3502     /* The machine class defines the default interrupt controller mode */
3503     spapr->irq = smc->irq;
3504     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3505                             spapr_set_ic_mode);
3506     object_property_set_description(obj, "ic-mode",
3507                  "Specifies the interrupt controller mode (xics, xive, dual)");
3508 
3509     object_property_add_str(obj, "host-model",
3510         spapr_get_host_model, spapr_set_host_model);
3511     object_property_set_description(obj, "host-model",
3512         "Host model to advertise in guest device tree");
3513     object_property_add_str(obj, "host-serial",
3514         spapr_get_host_serial, spapr_set_host_serial);
3515     object_property_set_description(obj, "host-serial",
3516         "Host serial number to advertise in guest device tree");
3517 }
3518 
3519 static void spapr_machine_finalizefn(Object *obj)
3520 {
3521     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3522 
3523     g_free(spapr->kvm_type);
3524 }
3525 
3526 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3527 {
3528     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3529     PowerPCCPU *cpu = POWERPC_CPU(cs);
3530     CPUPPCState *env = &cpu->env;
3531 
3532     cpu_synchronize_state(cs);
3533     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3534     if (spapr->fwnmi_system_reset_addr != -1) {
3535         uint64_t rtas_addr, addr;
3536 
3537         /* get rtas addr from fdt */
3538         rtas_addr = spapr_get_rtas_addr();
3539         if (!rtas_addr) {
3540             qemu_system_guest_panicked(NULL);
3541             return;
3542         }
3543 
3544         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3545         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3546         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3547         env->gpr[3] = addr;
3548     }
3549     ppc_cpu_do_system_reset(cs);
3550     if (spapr->fwnmi_system_reset_addr != -1) {
3551         env->nip = spapr->fwnmi_system_reset_addr;
3552     }
3553 }
3554 
3555 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3556 {
3557     CPUState *cs;
3558 
3559     CPU_FOREACH(cs) {
3560         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3561     }
3562 }
3563 
3564 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3565                           void *fdt, int *fdt_start_offset, Error **errp)
3566 {
3567     uint64_t addr;
3568     uint32_t node;
3569 
3570     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3571     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3572                                     &error_abort);
3573     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3574                                              SPAPR_MEMORY_BLOCK_SIZE);
3575     return 0;
3576 }
3577 
3578 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3579                            bool dedicated_hp_event_source)
3580 {
3581     SpaprDrc *drc;
3582     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3583     int i;
3584     uint64_t addr = addr_start;
3585     bool hotplugged = spapr_drc_hotplugged(dev);
3586 
3587     for (i = 0; i < nr_lmbs; i++) {
3588         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3589                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3590         g_assert(drc);
3591 
3592         /*
3593          * memory_device_get_free_addr() provided a range of free addresses
3594          * that doesn't overlap with any existing mapping at pre-plug. The
3595          * corresponding LMB DRCs are thus assumed to be all attachable.
3596          */
3597         spapr_drc_attach(drc, dev);
3598         if (!hotplugged) {
3599             spapr_drc_reset(drc);
3600         }
3601         addr += SPAPR_MEMORY_BLOCK_SIZE;
3602     }
3603     /* send hotplug notification to the
3604      * guest only in case of hotplugged memory
3605      */
3606     if (hotplugged) {
3607         if (dedicated_hp_event_source) {
3608             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3609                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3610             g_assert(drc);
3611             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3612                                                    nr_lmbs,
3613                                                    spapr_drc_index(drc));
3614         } else {
3615             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3616                                            nr_lmbs);
3617         }
3618     }
3619 }
3620 
3621 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3622 {
3623     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3624     PCDIMMDevice *dimm = PC_DIMM(dev);
3625     uint64_t size, addr;
3626     int64_t slot;
3627     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3628 
3629     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3630 
3631     pc_dimm_plug(dimm, MACHINE(ms));
3632 
3633     if (!is_nvdimm) {
3634         addr = object_property_get_uint(OBJECT(dimm),
3635                                         PC_DIMM_ADDR_PROP, &error_abort);
3636         spapr_add_lmbs(dev, addr, size,
3637                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3638     } else {
3639         slot = object_property_get_int(OBJECT(dimm),
3640                                        PC_DIMM_SLOT_PROP, &error_abort);
3641         /* We should have valid slot number at this point */
3642         g_assert(slot >= 0);
3643         spapr_add_nvdimm(dev, slot);
3644     }
3645 }
3646 
3647 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3648                                   Error **errp)
3649 {
3650     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3651     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3652     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3653     PCDIMMDevice *dimm = PC_DIMM(dev);
3654     Error *local_err = NULL;
3655     uint64_t size;
3656     Object *memdev;
3657     hwaddr pagesize;
3658 
3659     if (!smc->dr_lmb_enabled) {
3660         error_setg(errp, "Memory hotplug not supported for this machine");
3661         return;
3662     }
3663 
3664     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3665     if (local_err) {
3666         error_propagate(errp, local_err);
3667         return;
3668     }
3669 
3670     if (is_nvdimm) {
3671         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3672             return;
3673         }
3674     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3675         error_setg(errp, "Hotplugged memory size must be a multiple of "
3676                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3677         return;
3678     }
3679 
3680     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3681                                       &error_abort);
3682     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3683     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3684         return;
3685     }
3686 
3687     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3688 }
3689 
3690 struct SpaprDimmState {
3691     PCDIMMDevice *dimm;
3692     uint32_t nr_lmbs;
3693     QTAILQ_ENTRY(SpaprDimmState) next;
3694 };
3695 
3696 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3697                                                        PCDIMMDevice *dimm)
3698 {
3699     SpaprDimmState *dimm_state = NULL;
3700 
3701     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3702         if (dimm_state->dimm == dimm) {
3703             break;
3704         }
3705     }
3706     return dimm_state;
3707 }
3708 
3709 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3710                                                       uint32_t nr_lmbs,
3711                                                       PCDIMMDevice *dimm)
3712 {
3713     SpaprDimmState *ds = NULL;
3714 
3715     /*
3716      * If this request is for a DIMM whose removal had failed earlier
3717      * (due to guest's refusal to remove the LMBs), we would have this
3718      * dimm already in the pending_dimm_unplugs list. In that
3719      * case don't add again.
3720      */
3721     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3722     if (!ds) {
3723         ds = g_new0(SpaprDimmState, 1);
3724         ds->nr_lmbs = nr_lmbs;
3725         ds->dimm = dimm;
3726         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3727     }
3728     return ds;
3729 }
3730 
3731 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3732                                               SpaprDimmState *dimm_state)
3733 {
3734     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3735     g_free(dimm_state);
3736 }
3737 
3738 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3739                                                         PCDIMMDevice *dimm)
3740 {
3741     SpaprDrc *drc;
3742     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3743                                                   &error_abort);
3744     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3745     uint32_t avail_lmbs = 0;
3746     uint64_t addr_start, addr;
3747     int i;
3748 
3749     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3750                                           &error_abort);
3751 
3752     addr = addr_start;
3753     for (i = 0; i < nr_lmbs; i++) {
3754         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3755                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3756         g_assert(drc);
3757         if (drc->dev) {
3758             avail_lmbs++;
3759         }
3760         addr += SPAPR_MEMORY_BLOCK_SIZE;
3761     }
3762 
3763     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3764 }
3765 
3766 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3767 {
3768     SpaprDimmState *ds;
3769     PCDIMMDevice *dimm;
3770     SpaprDrc *drc;
3771     uint32_t nr_lmbs;
3772     uint64_t size, addr_start, addr;
3773     g_autofree char *qapi_error = NULL;
3774     int i;
3775 
3776     if (!dev) {
3777         return;
3778     }
3779 
3780     dimm = PC_DIMM(dev);
3781     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3782 
3783     /*
3784      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3785      * unplug state, but one of its DRC is marked as unplug_requested.
3786      * This is bad and weird enough to g_assert() out.
3787      */
3788     g_assert(ds);
3789 
3790     spapr_pending_dimm_unplugs_remove(spapr, ds);
3791 
3792     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3793     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3794 
3795     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3796                                           &error_abort);
3797 
3798     addr = addr_start;
3799     for (i = 0; i < nr_lmbs; i++) {
3800         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3801                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3802         g_assert(drc);
3803 
3804         drc->unplug_requested = false;
3805         addr += SPAPR_MEMORY_BLOCK_SIZE;
3806     }
3807 
3808     /*
3809      * Tell QAPI that something happened and the memory
3810      * hotunplug wasn't successful. Keep sending
3811      * MEM_UNPLUG_ERROR even while sending
3812      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3813      * MEM_UNPLUG_ERROR is due.
3814      */
3815     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3816                                  "for device %s", dev->id);
3817 
3818     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3819 
3820     qapi_event_send_device_unplug_guest_error(dev->id,
3821                                               dev->canonical_path);
3822 }
3823 
3824 /* Callback to be called during DRC release. */
3825 void spapr_lmb_release(DeviceState *dev)
3826 {
3827     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3828     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3829     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3830 
3831     /* This information will get lost if a migration occurs
3832      * during the unplug process. In this case recover it. */
3833     if (ds == NULL) {
3834         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3835         g_assert(ds);
3836         /* The DRC being examined by the caller at least must be counted */
3837         g_assert(ds->nr_lmbs);
3838     }
3839 
3840     if (--ds->nr_lmbs) {
3841         return;
3842     }
3843 
3844     /*
3845      * Now that all the LMBs have been removed by the guest, call the
3846      * unplug handler chain. This can never fail.
3847      */
3848     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3849     object_unparent(OBJECT(dev));
3850 }
3851 
3852 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3853 {
3854     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3855     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3856 
3857     /* We really shouldn't get this far without anything to unplug */
3858     g_assert(ds);
3859 
3860     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3861     qdev_unrealize(dev);
3862     spapr_pending_dimm_unplugs_remove(spapr, ds);
3863 }
3864 
3865 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3866                                         DeviceState *dev, Error **errp)
3867 {
3868     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3869     PCDIMMDevice *dimm = PC_DIMM(dev);
3870     uint32_t nr_lmbs;
3871     uint64_t size, addr_start, addr;
3872     int i;
3873     SpaprDrc *drc;
3874 
3875     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3876         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3877         return;
3878     }
3879 
3880     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3881     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3882 
3883     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3884                                           &error_abort);
3885 
3886     /*
3887      * An existing pending dimm state for this DIMM means that there is an
3888      * unplug operation in progress, waiting for the spapr_lmb_release
3889      * callback to complete the job (BQL can't cover that far). In this case,
3890      * bail out to avoid detaching DRCs that were already released.
3891      */
3892     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3893         error_setg(errp, "Memory unplug already in progress for device %s",
3894                    dev->id);
3895         return;
3896     }
3897 
3898     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3899 
3900     addr = addr_start;
3901     for (i = 0; i < nr_lmbs; i++) {
3902         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3903                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3904         g_assert(drc);
3905 
3906         spapr_drc_unplug_request(drc);
3907         addr += SPAPR_MEMORY_BLOCK_SIZE;
3908     }
3909 
3910     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3911                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3912     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3913                                               nr_lmbs, spapr_drc_index(drc));
3914 }
3915 
3916 /* Callback to be called during DRC release. */
3917 void spapr_core_release(DeviceState *dev)
3918 {
3919     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3920 
3921     /* Call the unplug handler chain. This can never fail. */
3922     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3923     object_unparent(OBJECT(dev));
3924 }
3925 
3926 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3927 {
3928     MachineState *ms = MACHINE(hotplug_dev);
3929     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3930     CPUCore *cc = CPU_CORE(dev);
3931     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3932 
3933     if (smc->pre_2_10_has_unused_icps) {
3934         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3935         int i;
3936 
3937         for (i = 0; i < cc->nr_threads; i++) {
3938             CPUState *cs = CPU(sc->threads[i]);
3939 
3940             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3941         }
3942     }
3943 
3944     assert(core_slot);
3945     core_slot->cpu = NULL;
3946     qdev_unrealize(dev);
3947 }
3948 
3949 static
3950 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3951                                Error **errp)
3952 {
3953     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3954     int index;
3955     SpaprDrc *drc;
3956     CPUCore *cc = CPU_CORE(dev);
3957 
3958     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3959         error_setg(errp, "Unable to find CPU core with core-id: %d",
3960                    cc->core_id);
3961         return;
3962     }
3963     if (index == 0) {
3964         error_setg(errp, "Boot CPU core may not be unplugged");
3965         return;
3966     }
3967 
3968     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3969                           spapr_vcpu_id(spapr, cc->core_id));
3970     g_assert(drc);
3971 
3972     if (!spapr_drc_unplug_requested(drc)) {
3973         spapr_drc_unplug_request(drc);
3974     }
3975 
3976     /*
3977      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3978      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3979      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3980      * attempt (e.g. the kernel will refuse to remove the last online
3981      * CPU), we will never attempt it again because unplug_requested
3982      * will still be 'true' in that case.
3983      */
3984     spapr_hotplug_req_remove_by_index(drc);
3985 }
3986 
3987 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3988                            void *fdt, int *fdt_start_offset, Error **errp)
3989 {
3990     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3991     CPUState *cs = CPU(core->threads[0]);
3992     PowerPCCPU *cpu = POWERPC_CPU(cs);
3993     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3994     int id = spapr_get_vcpu_id(cpu);
3995     g_autofree char *nodename = NULL;
3996     int offset;
3997 
3998     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3999     offset = fdt_add_subnode(fdt, 0, nodename);
4000 
4001     spapr_dt_cpu(cs, fdt, offset, spapr);
4002 
4003     /*
4004      * spapr_dt_cpu() does not fill the 'name' property in the
4005      * CPU node. The function is called during boot process, before
4006      * and after CAS, and overwriting the 'name' property written
4007      * by SLOF is not allowed.
4008      *
4009      * Write it manually after spapr_dt_cpu(). This makes the hotplug
4010      * CPUs more compatible with the coldplugged ones, which have
4011      * the 'name' property. Linux Kernel also relies on this
4012      * property to identify CPU nodes.
4013      */
4014     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
4015 
4016     *fdt_start_offset = offset;
4017     return 0;
4018 }
4019 
4020 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4021 {
4022     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4023     MachineClass *mc = MACHINE_GET_CLASS(spapr);
4024     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4025     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
4026     CPUCore *cc = CPU_CORE(dev);
4027     CPUState *cs;
4028     SpaprDrc *drc;
4029     CPUArchId *core_slot;
4030     int index;
4031     bool hotplugged = spapr_drc_hotplugged(dev);
4032     int i;
4033 
4034     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4035     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4036 
4037     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4038                           spapr_vcpu_id(spapr, cc->core_id));
4039 
4040     g_assert(drc || !mc->has_hotpluggable_cpus);
4041 
4042     if (drc) {
4043         /*
4044          * spapr_core_pre_plug() already buys us this is a brand new
4045          * core being plugged into a free slot. Nothing should already
4046          * be attached to the corresponding DRC.
4047          */
4048         spapr_drc_attach(drc, dev);
4049 
4050         if (hotplugged) {
4051             /*
4052              * Send hotplug notification interrupt to the guest only
4053              * in case of hotplugged CPUs.
4054              */
4055             spapr_hotplug_req_add_by_index(drc);
4056         } else {
4057             spapr_drc_reset(drc);
4058         }
4059     }
4060 
4061     core_slot->cpu = OBJECT(dev);
4062 
4063     /*
4064      * Set compatibility mode to match the boot CPU, which was either set
4065      * by the machine reset code or by CAS. This really shouldn't fail at
4066      * this point.
4067      */
4068     if (hotplugged) {
4069         for (i = 0; i < cc->nr_threads; i++) {
4070             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4071                            &error_abort);
4072         }
4073     }
4074 
4075     if (smc->pre_2_10_has_unused_icps) {
4076         for (i = 0; i < cc->nr_threads; i++) {
4077             cs = CPU(core->threads[i]);
4078             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4079         }
4080     }
4081 }
4082 
4083 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4084                                 Error **errp)
4085 {
4086     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4087     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4088     CPUCore *cc = CPU_CORE(dev);
4089     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4090     const char *type = object_get_typename(OBJECT(dev));
4091     CPUArchId *core_slot;
4092     int index;
4093     unsigned int smp_threads = machine->smp.threads;
4094 
4095     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4096         error_setg(errp, "CPU hotplug not supported for this machine");
4097         return;
4098     }
4099 
4100     if (strcmp(base_core_type, type)) {
4101         error_setg(errp, "CPU core type should be %s", base_core_type);
4102         return;
4103     }
4104 
4105     if (cc->core_id % smp_threads) {
4106         error_setg(errp, "invalid core id %d", cc->core_id);
4107         return;
4108     }
4109 
4110     /*
4111      * In general we should have homogeneous threads-per-core, but old
4112      * (pre hotplug support) machine types allow the last core to have
4113      * reduced threads as a compatibility hack for when we allowed
4114      * total vcpus not a multiple of threads-per-core.
4115      */
4116     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4117         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4118                    smp_threads);
4119         return;
4120     }
4121 
4122     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4123     if (!core_slot) {
4124         error_setg(errp, "core id %d out of range", cc->core_id);
4125         return;
4126     }
4127 
4128     if (core_slot->cpu) {
4129         error_setg(errp, "core %d already populated", cc->core_id);
4130         return;
4131     }
4132 
4133     numa_cpu_pre_plug(core_slot, dev, errp);
4134 }
4135 
4136 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4137                           void *fdt, int *fdt_start_offset, Error **errp)
4138 {
4139     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4140     int intc_phandle;
4141 
4142     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4143     if (intc_phandle <= 0) {
4144         return -1;
4145     }
4146 
4147     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4148         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4149         return -1;
4150     }
4151 
4152     /* generally SLOF creates these, for hotplug it's up to QEMU */
4153     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4154 
4155     return 0;
4156 }
4157 
4158 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4159                                Error **errp)
4160 {
4161     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4162     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4163     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4164     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4165     SpaprDrc *drc;
4166 
4167     if (dev->hotplugged && !smc->dr_phb_enabled) {
4168         error_setg(errp, "PHB hotplug not supported for this machine");
4169         return false;
4170     }
4171 
4172     if (sphb->index == (uint32_t)-1) {
4173         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4174         return false;
4175     }
4176 
4177     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4178     if (drc && drc->dev) {
4179         error_setg(errp, "PHB %d already attached", sphb->index);
4180         return false;
4181     }
4182 
4183     /*
4184      * This will check that sphb->index doesn't exceed the maximum number of
4185      * PHBs for the current machine type.
4186      */
4187     return
4188         smc->phb_placement(spapr, sphb->index,
4189                            &sphb->buid, &sphb->io_win_addr,
4190                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4191                            windows_supported, sphb->dma_liobn,
4192                            errp);
4193 }
4194 
4195 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4196 {
4197     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4198     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4199     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4200     SpaprDrc *drc;
4201     bool hotplugged = spapr_drc_hotplugged(dev);
4202 
4203     if (!smc->dr_phb_enabled) {
4204         return;
4205     }
4206 
4207     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4208     /* hotplug hooks should check it's enabled before getting this far */
4209     assert(drc);
4210 
4211     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4212     spapr_drc_attach(drc, dev);
4213 
4214     if (hotplugged) {
4215         spapr_hotplug_req_add_by_index(drc);
4216     } else {
4217         spapr_drc_reset(drc);
4218     }
4219 }
4220 
4221 void spapr_phb_release(DeviceState *dev)
4222 {
4223     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4224 
4225     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4226     object_unparent(OBJECT(dev));
4227 }
4228 
4229 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4230 {
4231     qdev_unrealize(dev);
4232 }
4233 
4234 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4235                                      DeviceState *dev, Error **errp)
4236 {
4237     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4238     SpaprDrc *drc;
4239 
4240     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4241     assert(drc);
4242 
4243     if (!spapr_drc_unplug_requested(drc)) {
4244         spapr_drc_unplug_request(drc);
4245         spapr_hotplug_req_remove_by_index(drc);
4246     } else {
4247         error_setg(errp,
4248                    "PCI Host Bridge unplug already in progress for device %s",
4249                    dev->id);
4250     }
4251 }
4252 
4253 static
4254 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4255                               Error **errp)
4256 {
4257     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4258 
4259     if (spapr->tpm_proxy != NULL) {
4260         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4261         return false;
4262     }
4263 
4264     return true;
4265 }
4266 
4267 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4268 {
4269     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4270     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4271 
4272     /* Already checked in spapr_tpm_proxy_pre_plug() */
4273     g_assert(spapr->tpm_proxy == NULL);
4274 
4275     spapr->tpm_proxy = tpm_proxy;
4276 }
4277 
4278 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4279 {
4280     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4281 
4282     qdev_unrealize(dev);
4283     object_unparent(OBJECT(dev));
4284     spapr->tpm_proxy = NULL;
4285 }
4286 
4287 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4288                                       DeviceState *dev, Error **errp)
4289 {
4290     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4291         spapr_memory_plug(hotplug_dev, dev);
4292     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4293         spapr_core_plug(hotplug_dev, dev);
4294     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4295         spapr_phb_plug(hotplug_dev, dev);
4296     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4297         spapr_tpm_proxy_plug(hotplug_dev, dev);
4298     }
4299 }
4300 
4301 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4302                                         DeviceState *dev, Error **errp)
4303 {
4304     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4305         spapr_memory_unplug(hotplug_dev, dev);
4306     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4307         spapr_core_unplug(hotplug_dev, dev);
4308     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4309         spapr_phb_unplug(hotplug_dev, dev);
4310     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4311         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4312     }
4313 }
4314 
4315 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4316 {
4317     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4318         /*
4319          * CAS will process all pending unplug requests.
4320          *
4321          * HACK: a guest could theoretically have cleared all bits in OV5,
4322          * but none of the guests we care for do.
4323          */
4324         spapr_ovec_empty(spapr->ov5_cas);
4325 }
4326 
4327 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4328                                                 DeviceState *dev, Error **errp)
4329 {
4330     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4331     MachineClass *mc = MACHINE_GET_CLASS(sms);
4332     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4333 
4334     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4335         if (spapr_memory_hot_unplug_supported(sms)) {
4336             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4337         } else {
4338             error_setg(errp, "Memory hot unplug not supported for this guest");
4339         }
4340     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4341         if (!mc->has_hotpluggable_cpus) {
4342             error_setg(errp, "CPU hot unplug not supported on this machine");
4343             return;
4344         }
4345         spapr_core_unplug_request(hotplug_dev, dev, errp);
4346     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4347         if (!smc->dr_phb_enabled) {
4348             error_setg(errp, "PHB hot unplug not supported on this machine");
4349             return;
4350         }
4351         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4352     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4353         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4354     }
4355 }
4356 
4357 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4358                                           DeviceState *dev, Error **errp)
4359 {
4360     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4361         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4362     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4363         spapr_core_pre_plug(hotplug_dev, dev, errp);
4364     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4365         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4366     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4367         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4368     }
4369 }
4370 
4371 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4372                                                  DeviceState *dev)
4373 {
4374     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4375         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4376         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4377         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4378         return HOTPLUG_HANDLER(machine);
4379     }
4380     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4381         PCIDevice *pcidev = PCI_DEVICE(dev);
4382         PCIBus *root = pci_device_root_bus(pcidev);
4383         SpaprPhbState *phb =
4384             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4385                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4386 
4387         if (phb) {
4388             return HOTPLUG_HANDLER(phb);
4389         }
4390     }
4391     return NULL;
4392 }
4393 
4394 static CpuInstanceProperties
4395 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4396 {
4397     CPUArchId *core_slot;
4398     MachineClass *mc = MACHINE_GET_CLASS(machine);
4399 
4400     /* make sure possible_cpu are initialized */
4401     mc->possible_cpu_arch_ids(machine);
4402     /* get CPU core slot containing thread that matches cpu_index */
4403     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4404     assert(core_slot);
4405     return core_slot->props;
4406 }
4407 
4408 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4409 {
4410     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4411 }
4412 
4413 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4414 {
4415     int i;
4416     unsigned int smp_threads = machine->smp.threads;
4417     unsigned int smp_cpus = machine->smp.cpus;
4418     const char *core_type;
4419     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4420     MachineClass *mc = MACHINE_GET_CLASS(machine);
4421 
4422     if (!mc->has_hotpluggable_cpus) {
4423         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4424     }
4425     if (machine->possible_cpus) {
4426         assert(machine->possible_cpus->len == spapr_max_cores);
4427         return machine->possible_cpus;
4428     }
4429 
4430     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4431     if (!core_type) {
4432         error_report("Unable to find sPAPR CPU Core definition");
4433         exit(1);
4434     }
4435 
4436     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4437                              sizeof(CPUArchId) * spapr_max_cores);
4438     machine->possible_cpus->len = spapr_max_cores;
4439     for (i = 0; i < machine->possible_cpus->len; i++) {
4440         int core_id = i * smp_threads;
4441 
4442         machine->possible_cpus->cpus[i].type = core_type;
4443         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4444         machine->possible_cpus->cpus[i].arch_id = core_id;
4445         machine->possible_cpus->cpus[i].props.has_core_id = true;
4446         machine->possible_cpus->cpus[i].props.core_id = core_id;
4447     }
4448     return machine->possible_cpus;
4449 }
4450 
4451 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4452                                 uint64_t *buid, hwaddr *pio,
4453                                 hwaddr *mmio32, hwaddr *mmio64,
4454                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4455 {
4456     /*
4457      * New-style PHB window placement.
4458      *
4459      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4460      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4461      * windows.
4462      *
4463      * Some guest kernels can't work with MMIO windows above 1<<46
4464      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4465      *
4466      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4467      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4468      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4469      * 1TiB 64-bit MMIO windows for each PHB.
4470      */
4471     const uint64_t base_buid = 0x800000020000000ULL;
4472     int i;
4473 
4474     /* Sanity check natural alignments */
4475     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4476     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4477     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4478     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4479     /* Sanity check bounds */
4480     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4481                       SPAPR_PCI_MEM32_WIN_SIZE);
4482     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4483                       SPAPR_PCI_MEM64_WIN_SIZE);
4484 
4485     if (index >= SPAPR_MAX_PHBS) {
4486         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4487                    SPAPR_MAX_PHBS - 1);
4488         return false;
4489     }
4490 
4491     *buid = base_buid + index;
4492     for (i = 0; i < n_dma; ++i) {
4493         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4494     }
4495 
4496     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4497     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4498     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4499     return true;
4500 }
4501 
4502 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4503 {
4504     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4505 
4506     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4507 }
4508 
4509 static void spapr_ics_resend(XICSFabric *dev)
4510 {
4511     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4512 
4513     ics_resend(spapr->ics);
4514 }
4515 
4516 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4517 {
4518     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4519 
4520     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4521 }
4522 
4523 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4524                                  Monitor *mon)
4525 {
4526     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4527 
4528     spapr_irq_print_info(spapr, mon);
4529     monitor_printf(mon, "irqchip: %s\n",
4530                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4531 }
4532 
4533 /*
4534  * This is a XIVE only operation
4535  */
4536 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4537                            uint8_t nvt_blk, uint32_t nvt_idx,
4538                            bool cam_ignore, uint8_t priority,
4539                            uint32_t logic_serv, XiveTCTXMatch *match)
4540 {
4541     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4542     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4543     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4544     int count;
4545 
4546     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4547                            priority, logic_serv, match);
4548     if (count < 0) {
4549         return count;
4550     }
4551 
4552     /*
4553      * When we implement the save and restore of the thread interrupt
4554      * contexts in the enter/exit CPU handlers of the machine and the
4555      * escalations in QEMU, we should be able to handle non dispatched
4556      * vCPUs.
4557      *
4558      * Until this is done, the sPAPR machine should find at least one
4559      * matching context always.
4560      */
4561     if (count == 0) {
4562         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4563                       nvt_blk, nvt_idx);
4564     }
4565 
4566     return count;
4567 }
4568 
4569 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4570 {
4571     return cpu->vcpu_id;
4572 }
4573 
4574 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4575 {
4576     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4577     MachineState *ms = MACHINE(spapr);
4578     int vcpu_id;
4579 
4580     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4581 
4582     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4583         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4584         error_append_hint(errp, "Adjust the number of cpus to %d "
4585                           "or try to raise the number of threads per core\n",
4586                           vcpu_id * ms->smp.threads / spapr->vsmt);
4587         return false;
4588     }
4589 
4590     cpu->vcpu_id = vcpu_id;
4591     return true;
4592 }
4593 
4594 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4595 {
4596     CPUState *cs;
4597 
4598     CPU_FOREACH(cs) {
4599         PowerPCCPU *cpu = POWERPC_CPU(cs);
4600 
4601         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4602             return cpu;
4603         }
4604     }
4605 
4606     return NULL;
4607 }
4608 
4609 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4610 {
4611     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4612 
4613     return spapr_cpu->in_nested;
4614 }
4615 
4616 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4617 {
4618     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4619 
4620     /* These are only called by TCG, KVM maintains dispatch state */
4621 
4622     spapr_cpu->prod = false;
4623     if (spapr_cpu->vpa_addr) {
4624         CPUState *cs = CPU(cpu);
4625         uint32_t dispatch;
4626 
4627         dispatch = ldl_be_phys(cs->as,
4628                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4629         dispatch++;
4630         if ((dispatch & 1) != 0) {
4631             qemu_log_mask(LOG_GUEST_ERROR,
4632                           "VPA: incorrect dispatch counter value for "
4633                           "dispatched partition %u, correcting.\n", dispatch);
4634             dispatch++;
4635         }
4636         stl_be_phys(cs->as,
4637                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4638     }
4639 }
4640 
4641 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4642 {
4643     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4644 
4645     if (spapr_cpu->vpa_addr) {
4646         CPUState *cs = CPU(cpu);
4647         uint32_t dispatch;
4648 
4649         dispatch = ldl_be_phys(cs->as,
4650                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4651         dispatch++;
4652         if ((dispatch & 1) != 1) {
4653             qemu_log_mask(LOG_GUEST_ERROR,
4654                           "VPA: incorrect dispatch counter value for "
4655                           "preempted partition %u, correcting.\n", dispatch);
4656             dispatch++;
4657         }
4658         stl_be_phys(cs->as,
4659                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4660     }
4661 }
4662 
4663 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4664 {
4665     MachineClass *mc = MACHINE_CLASS(oc);
4666     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4667     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4668     NMIClass *nc = NMI_CLASS(oc);
4669     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4670     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4671     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4672     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4673     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4674     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4675 
4676     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4677     mc->ignore_boot_device_suffixes = true;
4678 
4679     /*
4680      * We set up the default / latest behaviour here.  The class_init
4681      * functions for the specific versioned machine types can override
4682      * these details for backwards compatibility
4683      */
4684     mc->init = spapr_machine_init;
4685     mc->reset = spapr_machine_reset;
4686     mc->block_default_type = IF_SCSI;
4687 
4688     /*
4689      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4690      * In TCG the limit is restricted by the range of CPU IPIs available.
4691      */
4692     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4693 
4694     mc->no_parallel = 1;
4695     mc->default_boot_order = "";
4696     mc->default_ram_size = 512 * MiB;
4697     mc->default_ram_id = "ppc_spapr.ram";
4698     mc->default_display = "std";
4699     mc->kvm_type = spapr_kvm_type;
4700     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4701     mc->pci_allow_0_address = true;
4702     assert(!mc->get_hotplug_handler);
4703     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4704     hc->pre_plug = spapr_machine_device_pre_plug;
4705     hc->plug = spapr_machine_device_plug;
4706     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4707     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4708     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4709     hc->unplug_request = spapr_machine_device_unplug_request;
4710     hc->unplug = spapr_machine_device_unplug;
4711 
4712     smc->dr_lmb_enabled = true;
4713     smc->update_dt_enabled = true;
4714     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4715     mc->has_hotpluggable_cpus = true;
4716     mc->nvdimm_supported = true;
4717     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4718     fwc->get_dev_path = spapr_get_fw_dev_path;
4719     nc->nmi_monitor_handler = spapr_nmi;
4720     smc->phb_placement = spapr_phb_placement;
4721     vhc->cpu_in_nested = spapr_cpu_in_nested;
4722     vhc->deliver_hv_excp = spapr_exit_nested;
4723     vhc->hypercall = emulate_spapr_hypercall;
4724     vhc->hpt_mask = spapr_hpt_mask;
4725     vhc->map_hptes = spapr_map_hptes;
4726     vhc->unmap_hptes = spapr_unmap_hptes;
4727     vhc->hpte_set_c = spapr_hpte_set_c;
4728     vhc->hpte_set_r = spapr_hpte_set_r;
4729     vhc->get_pate = spapr_get_pate;
4730     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4731     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4732     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4733     xic->ics_get = spapr_ics_get;
4734     xic->ics_resend = spapr_ics_resend;
4735     xic->icp_get = spapr_icp_get;
4736     ispc->print_info = spapr_pic_print_info;
4737     /* Force NUMA node memory size to be a multiple of
4738      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4739      * in which LMBs are represented and hot-added
4740      */
4741     mc->numa_mem_align_shift = 28;
4742     mc->auto_enable_numa = true;
4743 
4744     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4745     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4746     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4747     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4748     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4749     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4750     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4751     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4752     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4753     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4754     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4755     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4756 
4757     /*
4758      * This cap specifies whether the AIL 3 mode for
4759      * H_SET_RESOURCE is supported. The default is modified
4760      * by default_caps_with_cpu().
4761      */
4762     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4763     spapr_caps_add_properties(smc);
4764     smc->irq = &spapr_irq_dual;
4765     smc->dr_phb_enabled = true;
4766     smc->linux_pci_probe = true;
4767     smc->smp_threads_vsmt = true;
4768     smc->nr_xirqs = SPAPR_NR_XIRQS;
4769     xfc->match_nvt = spapr_match_nvt;
4770     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4771     vmc->quiesce = spapr_vof_quiesce;
4772     vmc->setprop = spapr_vof_setprop;
4773 }
4774 
4775 static const TypeInfo spapr_machine_info = {
4776     .name          = TYPE_SPAPR_MACHINE,
4777     .parent        = TYPE_MACHINE,
4778     .abstract      = true,
4779     .instance_size = sizeof(SpaprMachineState),
4780     .instance_init = spapr_instance_init,
4781     .instance_finalize = spapr_machine_finalizefn,
4782     .class_size    = sizeof(SpaprMachineClass),
4783     .class_init    = spapr_machine_class_init,
4784     .interfaces = (InterfaceInfo[]) {
4785         { TYPE_FW_PATH_PROVIDER },
4786         { TYPE_NMI },
4787         { TYPE_HOTPLUG_HANDLER },
4788         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4789         { TYPE_XICS_FABRIC },
4790         { TYPE_INTERRUPT_STATS_PROVIDER },
4791         { TYPE_XIVE_FABRIC },
4792         { TYPE_VOF_MACHINE_IF },
4793         { }
4794     },
4795 };
4796 
4797 static void spapr_machine_latest_class_options(MachineClass *mc)
4798 {
4799     mc->alias = "pseries";
4800     mc->is_default = true;
4801 }
4802 
4803 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4804     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4805                                                     void *data)      \
4806     {                                                                \
4807         MachineClass *mc = MACHINE_CLASS(oc);                        \
4808         spapr_machine_##suffix##_class_options(mc);                  \
4809         if (latest) {                                                \
4810             spapr_machine_latest_class_options(mc);                  \
4811         }                                                            \
4812     }                                                                \
4813     static const TypeInfo spapr_machine_##suffix##_info = {          \
4814         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4815         .parent = TYPE_SPAPR_MACHINE,                                \
4816         .class_init = spapr_machine_##suffix##_class_init,           \
4817     };                                                               \
4818     static void spapr_machine_register_##suffix(void)                \
4819     {                                                                \
4820         type_register(&spapr_machine_##suffix##_info);               \
4821     }                                                                \
4822     type_init(spapr_machine_register_##suffix)
4823 
4824 /*
4825  * pseries-9.0
4826  */
4827 static void spapr_machine_9_0_class_options(MachineClass *mc)
4828 {
4829     /* Defaults for the latest behaviour inherited from the base class */
4830 }
4831 
4832 DEFINE_SPAPR_MACHINE(9_0, "9.0", true);
4833 
4834 /*
4835  * pseries-8.2
4836  */
4837 static void spapr_machine_8_2_class_options(MachineClass *mc)
4838 {
4839     spapr_machine_9_0_class_options(mc);
4840     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4841 }
4842 
4843 DEFINE_SPAPR_MACHINE(8_2, "8.2", false);
4844 
4845 /*
4846  * pseries-8.1
4847  */
4848 static void spapr_machine_8_1_class_options(MachineClass *mc)
4849 {
4850     spapr_machine_8_2_class_options(mc);
4851     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4852 }
4853 
4854 DEFINE_SPAPR_MACHINE(8_1, "8.1", false);
4855 
4856 /*
4857  * pseries-8.0
4858  */
4859 static void spapr_machine_8_0_class_options(MachineClass *mc)
4860 {
4861     spapr_machine_8_1_class_options(mc);
4862     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4863 }
4864 
4865 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4866 
4867 /*
4868  * pseries-7.2
4869  */
4870 static void spapr_machine_7_2_class_options(MachineClass *mc)
4871 {
4872     spapr_machine_8_0_class_options(mc);
4873     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4874 }
4875 
4876 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4877 
4878 /*
4879  * pseries-7.1
4880  */
4881 static void spapr_machine_7_1_class_options(MachineClass *mc)
4882 {
4883     spapr_machine_7_2_class_options(mc);
4884     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4885 }
4886 
4887 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4888 
4889 /*
4890  * pseries-7.0
4891  */
4892 static void spapr_machine_7_0_class_options(MachineClass *mc)
4893 {
4894     spapr_machine_7_1_class_options(mc);
4895     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4896 }
4897 
4898 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4899 
4900 /*
4901  * pseries-6.2
4902  */
4903 static void spapr_machine_6_2_class_options(MachineClass *mc)
4904 {
4905     spapr_machine_7_0_class_options(mc);
4906     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4907 }
4908 
4909 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4910 
4911 /*
4912  * pseries-6.1
4913  */
4914 static void spapr_machine_6_1_class_options(MachineClass *mc)
4915 {
4916     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4917 
4918     spapr_machine_6_2_class_options(mc);
4919     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4920     smc->pre_6_2_numa_affinity = true;
4921     mc->smp_props.prefer_sockets = true;
4922 }
4923 
4924 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4925 
4926 /*
4927  * pseries-6.0
4928  */
4929 static void spapr_machine_6_0_class_options(MachineClass *mc)
4930 {
4931     spapr_machine_6_1_class_options(mc);
4932     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4933 }
4934 
4935 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4936 
4937 /*
4938  * pseries-5.2
4939  */
4940 static void spapr_machine_5_2_class_options(MachineClass *mc)
4941 {
4942     spapr_machine_6_0_class_options(mc);
4943     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4944 }
4945 
4946 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4947 
4948 /*
4949  * pseries-5.1
4950  */
4951 static void spapr_machine_5_1_class_options(MachineClass *mc)
4952 {
4953     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4954 
4955     spapr_machine_5_2_class_options(mc);
4956     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4957     smc->pre_5_2_numa_associativity = true;
4958 }
4959 
4960 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4961 
4962 /*
4963  * pseries-5.0
4964  */
4965 static void spapr_machine_5_0_class_options(MachineClass *mc)
4966 {
4967     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4968     static GlobalProperty compat[] = {
4969         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4970     };
4971 
4972     spapr_machine_5_1_class_options(mc);
4973     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4974     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4975     mc->numa_mem_supported = true;
4976     smc->pre_5_1_assoc_refpoints = true;
4977 }
4978 
4979 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4980 
4981 /*
4982  * pseries-4.2
4983  */
4984 static void spapr_machine_4_2_class_options(MachineClass *mc)
4985 {
4986     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4987 
4988     spapr_machine_5_0_class_options(mc);
4989     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4990     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4991     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4992     smc->rma_limit = 16 * GiB;
4993     mc->nvdimm_supported = false;
4994 }
4995 
4996 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4997 
4998 /*
4999  * pseries-4.1
5000  */
5001 static void spapr_machine_4_1_class_options(MachineClass *mc)
5002 {
5003     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5004     static GlobalProperty compat[] = {
5005         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
5006         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
5007     };
5008 
5009     spapr_machine_4_2_class_options(mc);
5010     smc->linux_pci_probe = false;
5011     smc->smp_threads_vsmt = false;
5012     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
5013     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5014 }
5015 
5016 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
5017 
5018 /*
5019  * pseries-4.0
5020  */
5021 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
5022                               uint64_t *buid, hwaddr *pio,
5023                               hwaddr *mmio32, hwaddr *mmio64,
5024                               unsigned n_dma, uint32_t *liobns, Error **errp)
5025 {
5026     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
5027                              liobns, errp)) {
5028         return false;
5029     }
5030     return true;
5031 }
5032 static void spapr_machine_4_0_class_options(MachineClass *mc)
5033 {
5034     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5035 
5036     spapr_machine_4_1_class_options(mc);
5037     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5038     smc->phb_placement = phb_placement_4_0;
5039     smc->irq = &spapr_irq_xics;
5040     smc->pre_4_1_migration = true;
5041 }
5042 
5043 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
5044 
5045 /*
5046  * pseries-3.1
5047  */
5048 static void spapr_machine_3_1_class_options(MachineClass *mc)
5049 {
5050     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5051 
5052     spapr_machine_4_0_class_options(mc);
5053     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5054 
5055     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5056     smc->update_dt_enabled = false;
5057     smc->dr_phb_enabled = false;
5058     smc->broken_host_serial_model = true;
5059     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5060     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5061     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5062     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5063 }
5064 
5065 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
5066 
5067 /*
5068  * pseries-3.0
5069  */
5070 
5071 static void spapr_machine_3_0_class_options(MachineClass *mc)
5072 {
5073     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5074 
5075     spapr_machine_3_1_class_options(mc);
5076     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5077 
5078     smc->legacy_irq_allocation = true;
5079     smc->nr_xirqs = 0x400;
5080     smc->irq = &spapr_irq_xics_legacy;
5081 }
5082 
5083 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
5084 
5085 /*
5086  * pseries-2.12
5087  */
5088 static void spapr_machine_2_12_class_options(MachineClass *mc)
5089 {
5090     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5091     static GlobalProperty compat[] = {
5092         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5093         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5094     };
5095 
5096     spapr_machine_3_0_class_options(mc);
5097     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5098     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5099 
5100     /* We depend on kvm_enabled() to choose a default value for the
5101      * hpt-max-page-size capability. Of course we can't do it here
5102      * because this is too early and the HW accelerator isn't initialized
5103      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5104      */
5105     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5106 }
5107 
5108 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5109 
5110 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5111 {
5112     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5113 
5114     spapr_machine_2_12_class_options(mc);
5115     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5116     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5117     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5118 }
5119 
5120 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5121 
5122 /*
5123  * pseries-2.11
5124  */
5125 
5126 static void spapr_machine_2_11_class_options(MachineClass *mc)
5127 {
5128     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5129 
5130     spapr_machine_2_12_class_options(mc);
5131     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5132     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5133     mc->deprecation_reason = "old and not maintained - use a 2.12+ version";
5134 }
5135 
5136 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5137 
5138 /*
5139  * pseries-2.10
5140  */
5141 
5142 static void spapr_machine_2_10_class_options(MachineClass *mc)
5143 {
5144     spapr_machine_2_11_class_options(mc);
5145     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5146 }
5147 
5148 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5149 
5150 /*
5151  * pseries-2.9
5152  */
5153 
5154 static void spapr_machine_2_9_class_options(MachineClass *mc)
5155 {
5156     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5157     static GlobalProperty compat[] = {
5158         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5159     };
5160 
5161     spapr_machine_2_10_class_options(mc);
5162     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5163     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5164     smc->pre_2_10_has_unused_icps = true;
5165     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5166 }
5167 
5168 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5169 
5170 /*
5171  * pseries-2.8
5172  */
5173 
5174 static void spapr_machine_2_8_class_options(MachineClass *mc)
5175 {
5176     static GlobalProperty compat[] = {
5177         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5178     };
5179 
5180     spapr_machine_2_9_class_options(mc);
5181     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5182     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5183     mc->numa_mem_align_shift = 23;
5184 }
5185 
5186 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5187 
5188 /*
5189  * pseries-2.7
5190  */
5191 
5192 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5193                               uint64_t *buid, hwaddr *pio,
5194                               hwaddr *mmio32, hwaddr *mmio64,
5195                               unsigned n_dma, uint32_t *liobns, Error **errp)
5196 {
5197     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5198     const uint64_t base_buid = 0x800000020000000ULL;
5199     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5200     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5201     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5202     const uint32_t max_index = 255;
5203     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5204 
5205     uint64_t ram_top = MACHINE(spapr)->ram_size;
5206     hwaddr phb0_base, phb_base;
5207     int i;
5208 
5209     /* Do we have device memory? */
5210     if (MACHINE(spapr)->device_memory) {
5211         /* Can't just use maxram_size, because there may be an
5212          * alignment gap between normal and device memory regions
5213          */
5214         ram_top = MACHINE(spapr)->device_memory->base +
5215             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5216     }
5217 
5218     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5219 
5220     if (index > max_index) {
5221         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5222                    max_index);
5223         return false;
5224     }
5225 
5226     *buid = base_buid + index;
5227     for (i = 0; i < n_dma; ++i) {
5228         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5229     }
5230 
5231     phb_base = phb0_base + index * phb_spacing;
5232     *pio = phb_base + pio_offset;
5233     *mmio32 = phb_base + mmio_offset;
5234     /*
5235      * We don't set the 64-bit MMIO window, relying on the PHB's
5236      * fallback behaviour of automatically splitting a large "32-bit"
5237      * window into contiguous 32-bit and 64-bit windows
5238      */
5239 
5240     return true;
5241 }
5242 
5243 static void spapr_machine_2_7_class_options(MachineClass *mc)
5244 {
5245     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5246     static GlobalProperty compat[] = {
5247         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5248         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5249         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5250         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5251     };
5252 
5253     spapr_machine_2_8_class_options(mc);
5254     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5255     mc->default_machine_opts = "modern-hotplug-events=off";
5256     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5257     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5258     smc->phb_placement = phb_placement_2_7;
5259 }
5260 
5261 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5262 
5263 /*
5264  * pseries-2.6
5265  */
5266 
5267 static void spapr_machine_2_6_class_options(MachineClass *mc)
5268 {
5269     static GlobalProperty compat[] = {
5270         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5271     };
5272 
5273     spapr_machine_2_7_class_options(mc);
5274     mc->has_hotpluggable_cpus = false;
5275     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5276     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5277 }
5278 
5279 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5280 
5281 /*
5282  * pseries-2.5
5283  */
5284 
5285 static void spapr_machine_2_5_class_options(MachineClass *mc)
5286 {
5287     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5288     static GlobalProperty compat[] = {
5289         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5290     };
5291 
5292     spapr_machine_2_6_class_options(mc);
5293     smc->use_ohci_by_default = true;
5294     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5295     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5296 }
5297 
5298 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5299 
5300 /*
5301  * pseries-2.4
5302  */
5303 
5304 static void spapr_machine_2_4_class_options(MachineClass *mc)
5305 {
5306     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5307 
5308     spapr_machine_2_5_class_options(mc);
5309     smc->dr_lmb_enabled = false;
5310     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5311 }
5312 
5313 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5314 
5315 /*
5316  * pseries-2.3
5317  */
5318 
5319 static void spapr_machine_2_3_class_options(MachineClass *mc)
5320 {
5321     static GlobalProperty compat[] = {
5322         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5323     };
5324     spapr_machine_2_4_class_options(mc);
5325     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5326     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5327 }
5328 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5329 
5330 /*
5331  * pseries-2.2
5332  */
5333 
5334 static void spapr_machine_2_2_class_options(MachineClass *mc)
5335 {
5336     static GlobalProperty compat[] = {
5337         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5338     };
5339 
5340     spapr_machine_2_3_class_options(mc);
5341     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5342     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5343     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5344 }
5345 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5346 
5347 /*
5348  * pseries-2.1
5349  */
5350 
5351 static void spapr_machine_2_1_class_options(MachineClass *mc)
5352 {
5353     spapr_machine_2_2_class_options(mc);
5354     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5355 }
5356 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5357 
5358 static void spapr_machine_register_types(void)
5359 {
5360     type_register_static(&spapr_machine_info);
5361 }
5362 
5363 type_init(spapr_machine_register_types)
5364