xref: /qemu/hw/ppc/spapr.c (revision c447bb1ad4ead9ca8620f468b0324a338877f6bf)
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/tcg.h"
39 #include "sysemu/qtest.h"
40 #include "sysemu/reset.h"
41 #include "sysemu/runstate.h"
42 #include "qemu/log.h"
43 #include "hw/fw-path-provider.h"
44 #include "elf.h"
45 #include "net/net.h"
46 #include "sysemu/device_tree.h"
47 #include "sysemu/cpus.h"
48 #include "sysemu/hw_accel.h"
49 #include "kvm_ppc.h"
50 #include "migration/misc.h"
51 #include "migration/qemu-file-types.h"
52 #include "migration/global_state.h"
53 #include "migration/register.h"
54 #include "migration/blocker.h"
55 #include "mmu-hash64.h"
56 #include "mmu-book3s-v3.h"
57 #include "cpu-models.h"
58 #include "hw/core/cpu.h"
59 
60 #include "hw/ppc/ppc.h"
61 #include "hw/loader.h"
62 
63 #include "hw/ppc/fdt.h"
64 #include "hw/ppc/spapr.h"
65 #include "hw/ppc/spapr_nested.h"
66 #include "hw/ppc/spapr_vio.h"
67 #include "hw/ppc/vof.h"
68 #include "hw/qdev-properties.h"
69 #include "hw/pci-host/spapr.h"
70 #include "hw/pci/msi.h"
71 
72 #include "hw/pci/pci.h"
73 #include "hw/scsi/scsi.h"
74 #include "hw/virtio/virtio-scsi.h"
75 #include "hw/virtio/vhost-scsi-common.h"
76 
77 #include "exec/ram_addr.h"
78 #include "exec/confidential-guest-support.h"
79 #include "hw/usb.h"
80 #include "qemu/config-file.h"
81 #include "qemu/error-report.h"
82 #include "trace.h"
83 #include "hw/nmi.h"
84 #include "hw/intc/intc.h"
85 
86 #include "hw/ppc/spapr_cpu_core.h"
87 #include "hw/mem/memory-device.h"
88 #include "hw/ppc/spapr_tpm_proxy.h"
89 #include "hw/ppc/spapr_nvdimm.h"
90 #include "hw/ppc/spapr_numa.h"
91 
92 #include <libfdt.h>
93 
94 /* SLOF memory layout:
95  *
96  * SLOF raw image loaded at 0, copies its romfs right below the flat
97  * device-tree, then position SLOF itself 31M below that
98  *
99  * So we set FW_OVERHEAD to 40MB which should account for all of that
100  * and more
101  *
102  * We load our kernel at 4M, leaving space for SLOF initial image
103  */
104 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
105 #define FW_MAX_SIZE             0x400000
106 #define FW_FILE_NAME            "slof.bin"
107 #define FW_FILE_NAME_VOF        "vof.bin"
108 #define FW_OVERHEAD             0x2800000
109 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
110 
111 #define MIN_RMA_SLOF            (128 * MiB)
112 
113 #define PHANDLE_INTC            0x00001111
114 
115 /* These two functions implement the VCPU id numbering: one to compute them
116  * all and one to identify thread 0 of a VCORE. Any change to the first one
117  * is likely to have an impact on the second one, so let's keep them close.
118  */
119 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
120 {
121     MachineState *ms = MACHINE(spapr);
122     unsigned int smp_threads = ms->smp.threads;
123 
124     assert(spapr->vsmt);
125     return
126         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
127 }
128 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
129                                       PowerPCCPU *cpu)
130 {
131     assert(spapr->vsmt);
132     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
133 }
134 
135 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
136 {
137     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
138      * and newer QEMUs don't even have them. In both cases, we don't want
139      * to send anything on the wire.
140      */
141     return false;
142 }
143 
144 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
145     /*
146      * Hack ahead.  We can't have two devices with the same name and
147      * instance id.  So I rename this to pass make check.
148      * Real help from people who knows the hardware is needed.
149      */
150     .name = "icp/server",
151     .version_id = 1,
152     .minimum_version_id = 1,
153     .needed = pre_2_10_vmstate_dummy_icp_needed,
154     .fields = (const VMStateField[]) {
155         VMSTATE_UNUSED(4), /* uint32_t xirr */
156         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
157         VMSTATE_UNUSED(1), /* uint8_t mfrr */
158         VMSTATE_END_OF_LIST()
159     },
160 };
161 
162 /*
163  * See comment in hw/intc/xics.c:icp_realize()
164  *
165  * You have to remove vmstate_replace_hack_for_ppc() when you remove
166  * the machine types that need the following function.
167  */
168 static void pre_2_10_vmstate_register_dummy_icp(int i)
169 {
170     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
171                      (void *)(uintptr_t) i);
172 }
173 
174 /*
175  * See comment in hw/intc/xics.c:icp_realize()
176  *
177  * You have to remove vmstate_replace_hack_for_ppc() when you remove
178  * the machine types that need the following function.
179  */
180 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
181 {
182     /*
183      * This used to be:
184      *
185      *    vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
186      *                      (void *)(uintptr_t) i);
187      */
188 }
189 
190 int spapr_max_server_number(SpaprMachineState *spapr)
191 {
192     MachineState *ms = MACHINE(spapr);
193 
194     assert(spapr->vsmt);
195     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
196 }
197 
198 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
199                                   int smt_threads)
200 {
201     int i, ret = 0;
202     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
203     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
204     int index = spapr_get_vcpu_id(cpu);
205 
206     if (cpu->compat_pvr) {
207         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
208         if (ret < 0) {
209             return ret;
210         }
211     }
212 
213     /* Build interrupt servers and gservers properties */
214     for (i = 0; i < smt_threads; i++) {
215         servers_prop[i] = cpu_to_be32(index + i);
216         /* Hack, direct the group queues back to cpu 0 */
217         gservers_prop[i*2] = cpu_to_be32(index + i);
218         gservers_prop[i*2 + 1] = 0;
219     }
220     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
221                       servers_prop, sizeof(*servers_prop) * smt_threads);
222     if (ret < 0) {
223         return ret;
224     }
225     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
226                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
227 
228     return ret;
229 }
230 
231 static void spapr_dt_pa_features(SpaprMachineState *spapr,
232                                  PowerPCCPU *cpu,
233                                  void *fdt, int offset)
234 {
235     /*
236      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
237      * but not MTTCG, so disable it. To advertise it, a cap would have
238      * to be added, or support implemented for MTTCG.
239      *
240      * Copy/paste is not supported by TCG, so it is not advertised. KVM
241      * can execute them but it has no accelerator drivers which are usable,
242      * so there isn't much need for it anyway.
243      */
244 
245     /* These should be kept in sync with pnv */
246     uint8_t pa_features_206[] = { 6, 0,
247         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
248     uint8_t pa_features_207[] = { 24, 0,
249         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
250         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
251         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
252         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
253     uint8_t pa_features_300[] = { 66, 0,
254         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
255         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
256         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
257         /* 6: DS207 */
258         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
259         /* 16: Vector */
260         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
261         /* 18: Vec. Scalar, 20: Vec. XOR */
262         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
263         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
264         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
265         /* 32: LE atomic, 34: EBB + ext EBB */
266         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
267         /* 40: Radix MMU */
268         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
269         /* 42: PM, 44: PC RA, 46: SC vec'd */
270         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
271         /* 48: SIMD, 50: QP BFP, 52: String */
272         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
273         /* 54: DecFP, 56: DecI, 58: SHA */
274         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
275         /* 60: NM atomic, 62: RNG */
276         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
277     };
278     /* 3.1 removes SAO, HTM support */
279     uint8_t pa_features_31[] = { 74, 0,
280         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
281         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
282         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
283         /* 6: DS207 */
284         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
285         /* 16: Vector */
286         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
287         /* 18: Vec. Scalar, 20: Vec. XOR */
288         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
289         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
290         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
291         /* 32: LE atomic, 34: EBB + ext EBB */
292         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
293         /* 40: Radix MMU */
294         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
295         /* 42: PM, 44: PC RA, 46: SC vec'd */
296         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
297         /* 48: SIMD, 50: QP BFP, 52: String */
298         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
299         /* 54: DecFP, 56: DecI, 58: SHA */
300         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
301         /* 60: NM atomic, 62: RNG */
302         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
303         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
304         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
305         /* 72: [P]HASHST/[P]HASHCHK */
306         0x80, 0x00,                         /* 72 - 73 */
307     };
308     uint8_t *pa_features = NULL;
309     size_t pa_size;
310 
311     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
312         pa_features = pa_features_206;
313         pa_size = sizeof(pa_features_206);
314     }
315     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
316         pa_features = pa_features_207;
317         pa_size = sizeof(pa_features_207);
318     }
319     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
320         pa_features = pa_features_300;
321         pa_size = sizeof(pa_features_300);
322     }
323     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
324         pa_features = pa_features_31;
325         pa_size = sizeof(pa_features_31);
326     }
327     if (!pa_features) {
328         return;
329     }
330 
331     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
332         /*
333          * Note: we keep CI large pages off by default because a 64K capable
334          * guest provisioned with large pages might otherwise try to map a qemu
335          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
336          * even if that qemu runs on a 4k host.
337          * We dd this bit back here if we are confident this is not an issue
338          */
339         pa_features[3] |= 0x20;
340     }
341     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
342         pa_features[24] |= 0x80;    /* Transactional memory support */
343     }
344     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
345         /* Workaround for broken kernels that attempt (guest) radix
346          * mode when they can't handle it, if they see the radix bit set
347          * in pa-features. So hide it from them. */
348         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
349     }
350 
351     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
352 }
353 
354 static void spapr_dt_pi_features(SpaprMachineState *spapr,
355                                  PowerPCCPU *cpu,
356                                  void *fdt, int offset)
357 {
358     uint8_t pi_features[] = { 1, 0,
359         0x00 };
360 
361     if (kvm_enabled() && ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00,
362                                           0, cpu->compat_pvr)) {
363         /*
364          * POWER9 and later CPUs with KVM run in LPAR-per-thread mode where
365          * all threads are essentially independent CPUs, and msgsndp does not
366          * work (because it is physically-addressed) and therefore is
367          * emulated by KVM, so disable it here to ensure XIVE will be used.
368          * This is both KVM and CPU implementation-specific behaviour so a KVM
369          * cap would be cleanest, but for now this works. If KVM ever permits
370          * native msgsndp execution by guests, a cap could be added at that
371          * time.
372          */
373         pi_features[2] |= 0x08; /* 4: No msgsndp */
374     }
375 
376     _FDT((fdt_setprop(fdt, offset, "ibm,pi-features", pi_features,
377                       sizeof(pi_features))));
378 }
379 
380 static hwaddr spapr_node0_size(MachineState *machine)
381 {
382     if (machine->numa_state->num_nodes) {
383         int i;
384         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
385             if (machine->numa_state->nodes[i].node_mem) {
386                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
387                            machine->ram_size);
388             }
389         }
390     }
391     return machine->ram_size;
392 }
393 
394 static void add_str(GString *s, const gchar *s1)
395 {
396     g_string_append_len(s, s1, strlen(s1) + 1);
397 }
398 
399 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
400                                 hwaddr start, hwaddr size)
401 {
402     char mem_name[32];
403     uint64_t mem_reg_property[2];
404     int off;
405 
406     mem_reg_property[0] = cpu_to_be64(start);
407     mem_reg_property[1] = cpu_to_be64(size);
408 
409     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
410     off = fdt_add_subnode(fdt, 0, mem_name);
411     _FDT(off);
412     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
413     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
414                       sizeof(mem_reg_property))));
415     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
416     return off;
417 }
418 
419 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
420 {
421     MemoryDeviceInfoList *info;
422 
423     for (info = list; info; info = info->next) {
424         MemoryDeviceInfo *value = info->value;
425 
426         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
427             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
428 
429             if (addr >= pcdimm_info->addr &&
430                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
431                 return pcdimm_info->node;
432             }
433         }
434     }
435 
436     return -1;
437 }
438 
439 struct sPAPRDrconfCellV2 {
440      uint32_t seq_lmbs;
441      uint64_t base_addr;
442      uint32_t drc_index;
443      uint32_t aa_index;
444      uint32_t flags;
445 } QEMU_PACKED;
446 
447 typedef struct DrconfCellQueue {
448     struct sPAPRDrconfCellV2 cell;
449     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
450 } DrconfCellQueue;
451 
452 static DrconfCellQueue *
453 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
454                       uint32_t drc_index, uint32_t aa_index,
455                       uint32_t flags)
456 {
457     DrconfCellQueue *elem;
458 
459     elem = g_malloc0(sizeof(*elem));
460     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
461     elem->cell.base_addr = cpu_to_be64(base_addr);
462     elem->cell.drc_index = cpu_to_be32(drc_index);
463     elem->cell.aa_index = cpu_to_be32(aa_index);
464     elem->cell.flags = cpu_to_be32(flags);
465 
466     return elem;
467 }
468 
469 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
470                                       int offset, MemoryDeviceInfoList *dimms)
471 {
472     MachineState *machine = MACHINE(spapr);
473     uint8_t *int_buf, *cur_index;
474     int ret;
475     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
476     uint64_t addr, cur_addr, size;
477     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
478     uint64_t mem_end = machine->device_memory->base +
479                        memory_region_size(&machine->device_memory->mr);
480     uint32_t node, buf_len, nr_entries = 0;
481     SpaprDrc *drc;
482     DrconfCellQueue *elem, *next;
483     MemoryDeviceInfoList *info;
484     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
485         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
486 
487     /* Entry to cover RAM and the gap area */
488     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
489                                  SPAPR_LMB_FLAGS_RESERVED |
490                                  SPAPR_LMB_FLAGS_DRC_INVALID);
491     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
492     nr_entries++;
493 
494     cur_addr = machine->device_memory->base;
495     for (info = dimms; info; info = info->next) {
496         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
497 
498         addr = di->addr;
499         size = di->size;
500         node = di->node;
501 
502         /*
503          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
504          * area is marked hotpluggable in the next iteration for the bigger
505          * chunk including the NVDIMM occupied area.
506          */
507         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
508             continue;
509 
510         /* Entry for hot-pluggable area */
511         if (cur_addr < addr) {
512             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
513             g_assert(drc);
514             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
515                                          cur_addr, spapr_drc_index(drc), -1, 0);
516             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
517             nr_entries++;
518         }
519 
520         /* Entry for DIMM */
521         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
522         g_assert(drc);
523         elem = spapr_get_drconf_cell(size / lmb_size, addr,
524                                      spapr_drc_index(drc), node,
525                                      (SPAPR_LMB_FLAGS_ASSIGNED |
526                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
527         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
528         nr_entries++;
529         cur_addr = addr + size;
530     }
531 
532     /* Entry for remaining hotpluggable area */
533     if (cur_addr < mem_end) {
534         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
535         g_assert(drc);
536         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
537                                      cur_addr, spapr_drc_index(drc), -1, 0);
538         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
539         nr_entries++;
540     }
541 
542     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
543     int_buf = cur_index = g_malloc0(buf_len);
544     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
545     cur_index += sizeof(nr_entries);
546 
547     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
548         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
549         cur_index += sizeof(elem->cell);
550         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
551         g_free(elem);
552     }
553 
554     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
555     g_free(int_buf);
556     if (ret < 0) {
557         return -1;
558     }
559     return 0;
560 }
561 
562 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
563                                    int offset, MemoryDeviceInfoList *dimms)
564 {
565     MachineState *machine = MACHINE(spapr);
566     int i, ret;
567     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
568     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
569     uint32_t nr_lmbs = (machine->device_memory->base +
570                        memory_region_size(&machine->device_memory->mr)) /
571                        lmb_size;
572     uint32_t *int_buf, *cur_index, buf_len;
573 
574     /*
575      * Allocate enough buffer size to fit in ibm,dynamic-memory
576      */
577     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
578     cur_index = int_buf = g_malloc0(buf_len);
579     int_buf[0] = cpu_to_be32(nr_lmbs);
580     cur_index++;
581     for (i = 0; i < nr_lmbs; i++) {
582         uint64_t addr = i * lmb_size;
583         uint32_t *dynamic_memory = cur_index;
584 
585         if (i >= device_lmb_start) {
586             SpaprDrc *drc;
587 
588             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
589             g_assert(drc);
590 
591             dynamic_memory[0] = cpu_to_be32(addr >> 32);
592             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
593             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
594             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
595             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
596             if (memory_region_present(get_system_memory(), addr)) {
597                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
598             } else {
599                 dynamic_memory[5] = cpu_to_be32(0);
600             }
601         } else {
602             /*
603              * LMB information for RMA, boot time RAM and gap b/n RAM and
604              * device memory region -- all these are marked as reserved
605              * and as having no valid DRC.
606              */
607             dynamic_memory[0] = cpu_to_be32(addr >> 32);
608             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
609             dynamic_memory[2] = cpu_to_be32(0);
610             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
611             dynamic_memory[4] = cpu_to_be32(-1);
612             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
613                                             SPAPR_LMB_FLAGS_DRC_INVALID);
614         }
615 
616         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
617     }
618     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
619     g_free(int_buf);
620     if (ret < 0) {
621         return -1;
622     }
623     return 0;
624 }
625 
626 /*
627  * Adds ibm,dynamic-reconfiguration-memory node.
628  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
629  * of this device tree node.
630  */
631 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
632                                                    void *fdt)
633 {
634     MachineState *machine = MACHINE(spapr);
635     int ret, offset;
636     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
637     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
638                                 cpu_to_be32(lmb_size & 0xffffffff)};
639     MemoryDeviceInfoList *dimms = NULL;
640 
641     /* Don't create the node if there is no device memory. */
642     if (!machine->device_memory) {
643         return 0;
644     }
645 
646     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
647 
648     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
649                     sizeof(prop_lmb_size));
650     if (ret < 0) {
651         return ret;
652     }
653 
654     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
655     if (ret < 0) {
656         return ret;
657     }
658 
659     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
660     if (ret < 0) {
661         return ret;
662     }
663 
664     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
665     dimms = qmp_memory_device_list();
666     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
667         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
668     } else {
669         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
670     }
671     qapi_free_MemoryDeviceInfoList(dimms);
672 
673     if (ret < 0) {
674         return ret;
675     }
676 
677     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
678 
679     return ret;
680 }
681 
682 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
683 {
684     MachineState *machine = MACHINE(spapr);
685     hwaddr mem_start, node_size;
686     int i, nb_nodes = machine->numa_state->num_nodes;
687     NodeInfo *nodes = machine->numa_state->nodes;
688 
689     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
690         if (!nodes[i].node_mem) {
691             continue;
692         }
693         if (mem_start >= machine->ram_size) {
694             node_size = 0;
695         } else {
696             node_size = nodes[i].node_mem;
697             if (node_size > machine->ram_size - mem_start) {
698                 node_size = machine->ram_size - mem_start;
699             }
700         }
701         if (!mem_start) {
702             /* spapr_machine_init() checks for rma_size <= node0_size
703              * already */
704             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
705             mem_start += spapr->rma_size;
706             node_size -= spapr->rma_size;
707         }
708         for ( ; node_size; ) {
709             hwaddr sizetmp = pow2floor(node_size);
710 
711             /* mem_start != 0 here */
712             if (ctzl(mem_start) < ctzl(sizetmp)) {
713                 sizetmp = 1ULL << ctzl(mem_start);
714             }
715 
716             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
717             node_size -= sizetmp;
718             mem_start += sizetmp;
719         }
720     }
721 
722     /* Generate ibm,dynamic-reconfiguration-memory node if required */
723     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
724         int ret;
725 
726         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
727         if (ret) {
728             return ret;
729         }
730     }
731 
732     return 0;
733 }
734 
735 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
736                          SpaprMachineState *spapr)
737 {
738     MachineState *ms = MACHINE(spapr);
739     PowerPCCPU *cpu = POWERPC_CPU(cs);
740     CPUPPCState *env = &cpu->env;
741     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
742     int index = spapr_get_vcpu_id(cpu);
743     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
744                        0xffffffff, 0xffffffff};
745     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
746         : SPAPR_TIMEBASE_FREQ;
747     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
748     uint32_t page_sizes_prop[64];
749     size_t page_sizes_prop_size;
750     unsigned int smp_threads = ms->smp.threads;
751     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
752     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
753     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
754     SpaprDrc *drc;
755     int drc_index;
756     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
757     int i;
758 
759     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
760     if (drc) {
761         drc_index = spapr_drc_index(drc);
762         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
763     }
764 
765     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
766     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
767 
768     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
769     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
770                            env->dcache_line_size)));
771     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
772                            env->dcache_line_size)));
773     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
774                            env->icache_line_size)));
775     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
776                            env->icache_line_size)));
777 
778     if (pcc->l1_dcache_size) {
779         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
780                                pcc->l1_dcache_size)));
781     } else {
782         warn_report("Unknown L1 dcache size for cpu");
783     }
784     if (pcc->l1_icache_size) {
785         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
786                                pcc->l1_icache_size)));
787     } else {
788         warn_report("Unknown L1 icache size for cpu");
789     }
790 
791     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
792     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
793     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
794     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
795     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
796     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
797 
798     if (ppc_has_spr(cpu, SPR_PURR)) {
799         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
800     }
801     if (ppc_has_spr(cpu, SPR_PURR)) {
802         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
803     }
804 
805     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
806         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
807                           segs, sizeof(segs))));
808     }
809 
810     /* Advertise VSX (vector extensions) if available
811      *   1               == VMX / Altivec available
812      *   2               == VSX available
813      *
814      * Only CPUs for which we create core types in spapr_cpu_core.c
815      * are possible, and all of those have VMX */
816     if (env->insns_flags & PPC_ALTIVEC) {
817         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
818             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
819         } else {
820             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
821         }
822     }
823 
824     /* Advertise DFP (Decimal Floating Point) if available
825      *   0 / no property == no DFP
826      *   1               == DFP available */
827     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
828         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
829     }
830 
831     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
832                                                       sizeof(page_sizes_prop));
833     if (page_sizes_prop_size) {
834         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
835                           page_sizes_prop, page_sizes_prop_size)));
836     }
837 
838     spapr_dt_pa_features(spapr, cpu, fdt, offset);
839 
840     spapr_dt_pi_features(spapr, cpu, fdt, offset);
841 
842     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
843                            cs->cpu_index / vcpus_per_socket)));
844 
845     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
846                       pft_size_prop, sizeof(pft_size_prop))));
847 
848     if (ms->numa_state->num_nodes > 1) {
849         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
850     }
851 
852     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
853 
854     if (pcc->radix_page_info) {
855         for (i = 0; i < pcc->radix_page_info->count; i++) {
856             radix_AP_encodings[i] =
857                 cpu_to_be32(pcc->radix_page_info->entries[i]);
858         }
859         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
860                           radix_AP_encodings,
861                           pcc->radix_page_info->count *
862                           sizeof(radix_AP_encodings[0]))));
863     }
864 
865     /*
866      * We set this property to let the guest know that it can use the large
867      * decrementer and its width in bits.
868      */
869     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
870         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
871                               pcc->lrg_decr_bits)));
872 }
873 
874 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
875                              int cpus_offset)
876 {
877     PowerPCCPU *cpu = POWERPC_CPU(cs);
878     int index = spapr_get_vcpu_id(cpu);
879     DeviceClass *dc = DEVICE_GET_CLASS(cs);
880     g_autofree char *nodename = NULL;
881     int offset;
882 
883     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
884         return;
885     }
886 
887     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
888     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
889     _FDT(offset);
890     spapr_dt_cpu(cs, fdt, offset, spapr);
891 }
892 
893 
894 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
895 {
896     CPUState **rev;
897     CPUState *cs;
898     int n_cpus;
899     int cpus_offset;
900     int i;
901 
902     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
903     _FDT(cpus_offset);
904     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
905     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
906 
907     /*
908      * We walk the CPUs in reverse order to ensure that CPU DT nodes
909      * created by fdt_add_subnode() end up in the right order in FDT
910      * for the guest kernel the enumerate the CPUs correctly.
911      *
912      * The CPU list cannot be traversed in reverse order, so we need
913      * to do extra work.
914      */
915     n_cpus = 0;
916     rev = NULL;
917     CPU_FOREACH(cs) {
918         rev = g_renew(CPUState *, rev, n_cpus + 1);
919         rev[n_cpus++] = cs;
920     }
921 
922     for (i = n_cpus - 1; i >= 0; i--) {
923         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
924     }
925 
926     g_free(rev);
927 }
928 
929 static int spapr_dt_rng(void *fdt)
930 {
931     int node;
932     int ret;
933 
934     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
935     if (node <= 0) {
936         return -1;
937     }
938     ret = fdt_setprop_string(fdt, node, "device_type",
939                              "ibm,platform-facilities");
940     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
941     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
942 
943     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
944     if (node <= 0) {
945         return -1;
946     }
947     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
948 
949     return ret ? -1 : 0;
950 }
951 
952 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
953 {
954     MachineState *ms = MACHINE(spapr);
955     int rtas;
956     GString *hypertas = g_string_sized_new(256);
957     GString *qemu_hypertas = g_string_sized_new(256);
958     uint32_t lrdr_capacity[] = {
959         0,
960         0,
961         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
962         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
963         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
964     };
965 
966     /* Do we have device memory? */
967     if (MACHINE(spapr)->device_memory) {
968         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
969             memory_region_size(&MACHINE(spapr)->device_memory->mr);
970 
971         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
972         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
973     }
974 
975     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
976 
977     /* hypertas */
978     add_str(hypertas, "hcall-pft");
979     add_str(hypertas, "hcall-term");
980     add_str(hypertas, "hcall-dabr");
981     add_str(hypertas, "hcall-interrupt");
982     add_str(hypertas, "hcall-tce");
983     add_str(hypertas, "hcall-vio");
984     add_str(hypertas, "hcall-splpar");
985     add_str(hypertas, "hcall-join");
986     add_str(hypertas, "hcall-bulk");
987     add_str(hypertas, "hcall-set-mode");
988     add_str(hypertas, "hcall-sprg0");
989     add_str(hypertas, "hcall-copy");
990     add_str(hypertas, "hcall-debug");
991     add_str(hypertas, "hcall-vphn");
992     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
993         add_str(hypertas, "hcall-rpt-invalidate");
994     }
995 
996     add_str(qemu_hypertas, "hcall-memop1");
997 
998     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
999         add_str(hypertas, "hcall-multi-tce");
1000     }
1001 
1002     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1003         add_str(hypertas, "hcall-hpt-resize");
1004     }
1005 
1006     add_str(hypertas, "hcall-watchdog");
1007 
1008     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1009                      hypertas->str, hypertas->len));
1010     g_string_free(hypertas, TRUE);
1011     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1012                      qemu_hypertas->str, qemu_hypertas->len));
1013     g_string_free(qemu_hypertas, TRUE);
1014 
1015     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
1016 
1017     /*
1018      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
1019      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
1020      *
1021      * The system reset requirements are driven by existing Linux and PowerVM
1022      * implementation which (contrary to PAPR) saves r3 in the error log
1023      * structure like machine check, so Linux expects to find the saved r3
1024      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
1025      * does not look at the error value).
1026      *
1027      * System reset interrupts are not subject to interlock like machine
1028      * check, so this memory area could be corrupted if the sreset is
1029      * interrupted by a machine check (or vice versa) if it was shared. To
1030      * prevent this, system reset uses per-CPU areas for the sreset save
1031      * area. A system reset that interrupts a system reset handler could
1032      * still overwrite this area, but Linux doesn't try to recover in that
1033      * case anyway.
1034      *
1035      * The extra 8 bytes is required because Linux's FWNMI error log check
1036      * is off-by-one.
1037      *
1038      * RTAS_MIN_SIZE is required for the RTAS blob itself.
1039      */
1040     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
1041                           RTAS_ERROR_LOG_MAX +
1042                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
1043                           sizeof(uint64_t)));
1044     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1045                           RTAS_ERROR_LOG_MAX));
1046     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1047                           RTAS_EVENT_SCAN_RATE));
1048 
1049     g_assert(msi_nonbroken);
1050     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1051 
1052     /*
1053      * According to PAPR, rtas ibm,os-term does not guarantee a return
1054      * back to the guest cpu.
1055      *
1056      * While an additional ibm,extended-os-term property indicates
1057      * that rtas call return will always occur. Set this property.
1058      */
1059     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1060 
1061     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1062                      lrdr_capacity, sizeof(lrdr_capacity)));
1063 
1064     spapr_dt_rtas_tokens(fdt, rtas);
1065 }
1066 
1067 /*
1068  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1069  * and the XIVE features that the guest may request and thus the valid
1070  * values for bytes 23..26 of option vector 5:
1071  */
1072 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1073                                           int chosen)
1074 {
1075     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1076 
1077     char val[2 * 4] = {
1078         23, 0x00, /* XICS / XIVE mode */
1079         24, 0x00, /* Hash/Radix, filled in below. */
1080         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1081         26, 0x40, /* Radix options: GTSE == yes. */
1082     };
1083 
1084     if (spapr->irq->xics && spapr->irq->xive) {
1085         val[1] = SPAPR_OV5_XIVE_BOTH;
1086     } else if (spapr->irq->xive) {
1087         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1088     } else {
1089         assert(spapr->irq->xics);
1090         val[1] = SPAPR_OV5_XIVE_LEGACY;
1091     }
1092 
1093     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1094                           first_ppc_cpu->compat_pvr)) {
1095         /*
1096          * If we're in a pre POWER9 compat mode then the guest should
1097          * do hash and use the legacy interrupt mode
1098          */
1099         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1100         val[3] = 0x00; /* Hash */
1101         spapr_check_mmu_mode(false);
1102     } else if (kvm_enabled()) {
1103         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1104             val[3] = 0x80; /* OV5_MMU_BOTH */
1105         } else if (kvmppc_has_cap_mmu_radix()) {
1106             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1107         } else {
1108             val[3] = 0x00; /* Hash */
1109         }
1110     } else {
1111         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1112         val[3] = 0xC0;
1113     }
1114     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1115                      val, sizeof(val)));
1116 }
1117 
1118 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1119 {
1120     MachineState *machine = MACHINE(spapr);
1121     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1122     int chosen;
1123 
1124     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1125 
1126     if (reset) {
1127         const char *boot_device = spapr->boot_device;
1128         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1129         size_t cb = 0;
1130         g_autofree char *bootlist = get_boot_devices_list(&cb);
1131 
1132         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1133             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1134                                     machine->kernel_cmdline));
1135         }
1136 
1137         if (spapr->initrd_size) {
1138             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1139                                   spapr->initrd_base));
1140             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1141                                   spapr->initrd_base + spapr->initrd_size));
1142         }
1143 
1144         if (spapr->kernel_size) {
1145             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1146                                   cpu_to_be64(spapr->kernel_size) };
1147 
1148             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1149                          &kprop, sizeof(kprop)));
1150             if (spapr->kernel_le) {
1151                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1152             }
1153         }
1154         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1155             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1156         }
1157         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1158         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1159         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1160 
1161         if (cb && bootlist) {
1162             int i;
1163 
1164             for (i = 0; i < cb; i++) {
1165                 if (bootlist[i] == '\n') {
1166                     bootlist[i] = ' ';
1167                 }
1168             }
1169             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1170         }
1171 
1172         if (boot_device && strlen(boot_device)) {
1173             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1174         }
1175 
1176         if (spapr->want_stdout_path && stdout_path) {
1177             /*
1178              * "linux,stdout-path" and "stdout" properties are
1179              * deprecated by linux kernel. New platforms should only
1180              * use the "stdout-path" property. Set the new property
1181              * and continue using older property to remain compatible
1182              * with the existing firmware.
1183              */
1184             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1185             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1186         }
1187 
1188         /*
1189          * We can deal with BAR reallocation just fine, advertise it
1190          * to the guest
1191          */
1192         if (smc->linux_pci_probe) {
1193             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1194         }
1195 
1196         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1197     }
1198 
1199     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1200 
1201     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1202 }
1203 
1204 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1205 {
1206     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1207      * KVM to work under pHyp with some guest co-operation */
1208     int hypervisor;
1209     uint8_t hypercall[16];
1210 
1211     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1212     /* indicate KVM hypercall interface */
1213     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1214     if (kvmppc_has_cap_fixup_hcalls()) {
1215         /*
1216          * Older KVM versions with older guest kernels were broken
1217          * with the magic page, don't allow the guest to map it.
1218          */
1219         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1220                                   sizeof(hypercall))) {
1221             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1222                              hypercall, sizeof(hypercall)));
1223         }
1224     }
1225 }
1226 
1227 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1228 {
1229     MachineState *machine = MACHINE(spapr);
1230     MachineClass *mc = MACHINE_GET_CLASS(machine);
1231     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1232     uint32_t root_drc_type_mask = 0;
1233     int ret;
1234     void *fdt;
1235     SpaprPhbState *phb;
1236     char *buf;
1237 
1238     fdt = g_malloc0(space);
1239     _FDT((fdt_create_empty_tree(fdt, space)));
1240 
1241     /* Root node */
1242     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1243     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1244     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1245 
1246     /* Guest UUID & Name*/
1247     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1248     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1249     if (qemu_uuid_set) {
1250         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1251     }
1252     g_free(buf);
1253 
1254     if (qemu_get_vm_name()) {
1255         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1256                                 qemu_get_vm_name()));
1257     }
1258 
1259     /* Host Model & Serial Number */
1260     if (spapr->host_model) {
1261         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1262     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1263         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1264         g_free(buf);
1265     }
1266 
1267     if (spapr->host_serial) {
1268         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1269     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1270         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1271         g_free(buf);
1272     }
1273 
1274     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1275     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1276 
1277     /* /interrupt controller */
1278     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1279 
1280     ret = spapr_dt_memory(spapr, fdt);
1281     if (ret < 0) {
1282         error_report("couldn't setup memory nodes in fdt");
1283         exit(1);
1284     }
1285 
1286     /* /vdevice */
1287     spapr_dt_vdevice(spapr->vio_bus, fdt);
1288 
1289     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1290         ret = spapr_dt_rng(fdt);
1291         if (ret < 0) {
1292             error_report("could not set up rng device in the fdt");
1293             exit(1);
1294         }
1295     }
1296 
1297     QLIST_FOREACH(phb, &spapr->phbs, list) {
1298         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1299         if (ret < 0) {
1300             error_report("couldn't setup PCI devices in fdt");
1301             exit(1);
1302         }
1303     }
1304 
1305     spapr_dt_cpus(fdt, spapr);
1306 
1307     /* ibm,drc-indexes and friends */
1308     root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1309     if (smc->dr_phb_enabled) {
1310         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1311     }
1312     if (mc->nvdimm_supported) {
1313         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1314     }
1315     if (root_drc_type_mask) {
1316         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1317     }
1318 
1319     if (mc->has_hotpluggable_cpus) {
1320         int offset = fdt_path_offset(fdt, "/cpus");
1321         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1322         if (ret < 0) {
1323             error_report("Couldn't set up CPU DR device tree properties");
1324             exit(1);
1325         }
1326     }
1327 
1328     /* /event-sources */
1329     spapr_dt_events(spapr, fdt);
1330 
1331     /* /rtas */
1332     spapr_dt_rtas(spapr, fdt);
1333 
1334     /* /chosen */
1335     spapr_dt_chosen(spapr, fdt, reset);
1336 
1337     /* /hypervisor */
1338     if (kvm_enabled()) {
1339         spapr_dt_hypervisor(spapr, fdt);
1340     }
1341 
1342     /* Build memory reserve map */
1343     if (reset) {
1344         if (spapr->kernel_size) {
1345             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1346                                   spapr->kernel_size)));
1347         }
1348         if (spapr->initrd_size) {
1349             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1350                                   spapr->initrd_size)));
1351         }
1352     }
1353 
1354     /* NVDIMM devices */
1355     if (mc->nvdimm_supported) {
1356         spapr_dt_persistent_memory(spapr, fdt);
1357     }
1358 
1359     return fdt;
1360 }
1361 
1362 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1363 {
1364     SpaprMachineState *spapr = opaque;
1365 
1366     return (addr & 0x0fffffff) + spapr->kernel_addr;
1367 }
1368 
1369 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1370                                     PowerPCCPU *cpu)
1371 {
1372     CPUPPCState *env = &cpu->env;
1373 
1374     /* The TCG path should also be holding the BQL at this point */
1375     g_assert(bql_locked());
1376 
1377     g_assert(!vhyp_cpu_in_nested(cpu));
1378 
1379     if (FIELD_EX64(env->msr, MSR, PR)) {
1380         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1381         env->gpr[3] = H_PRIVILEGE;
1382     } else {
1383         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1384     }
1385 }
1386 
1387 struct LPCRSyncState {
1388     target_ulong value;
1389     target_ulong mask;
1390 };
1391 
1392 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1393 {
1394     struct LPCRSyncState *s = arg.host_ptr;
1395     PowerPCCPU *cpu = POWERPC_CPU(cs);
1396     CPUPPCState *env = &cpu->env;
1397     target_ulong lpcr;
1398 
1399     cpu_synchronize_state(cs);
1400     lpcr = env->spr[SPR_LPCR];
1401     lpcr &= ~s->mask;
1402     lpcr |= s->value;
1403     ppc_store_lpcr(cpu, lpcr);
1404 }
1405 
1406 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1407 {
1408     CPUState *cs;
1409     struct LPCRSyncState s = {
1410         .value = value,
1411         .mask = mask
1412     };
1413     CPU_FOREACH(cs) {
1414         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1415     }
1416 }
1417 
1418 /* May be used when the machine is not running */
1419 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1420 {
1421     CPUState *cs;
1422     CPU_FOREACH(cs) {
1423         PowerPCCPU *cpu = POWERPC_CPU(cs);
1424         CPUPPCState *env = &cpu->env;
1425         target_ulong lpcr;
1426 
1427         lpcr = env->spr[SPR_LPCR];
1428         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1429         ppc_store_lpcr(cpu, lpcr);
1430     }
1431 }
1432 
1433 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1434                            target_ulong lpid, ppc_v3_pate_t *entry)
1435 {
1436     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1437     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1438 
1439     if (!spapr_cpu->in_nested) {
1440         assert(lpid == 0);
1441 
1442         /* Copy PATE1:GR into PATE0:HR */
1443         entry->dw0 = spapr->patb_entry & PATE0_HR;
1444         entry->dw1 = spapr->patb_entry;
1445         return true;
1446     } else {
1447         if (spapr_nested_api(spapr) == NESTED_API_KVM_HV) {
1448             return spapr_get_pate_nested_hv(spapr, cpu, lpid, entry);
1449         } else if (spapr_nested_api(spapr) == NESTED_API_PAPR) {
1450             return spapr_get_pate_nested_papr(spapr, cpu, lpid, entry);
1451         } else {
1452             g_assert_not_reached();
1453         }
1454     }
1455 }
1456 
1457 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1458 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1459 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1460 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1461 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1462 
1463 /*
1464  * Get the fd to access the kernel htab, re-opening it if necessary
1465  */
1466 static int get_htab_fd(SpaprMachineState *spapr)
1467 {
1468     Error *local_err = NULL;
1469 
1470     if (spapr->htab_fd >= 0) {
1471         return spapr->htab_fd;
1472     }
1473 
1474     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1475     if (spapr->htab_fd < 0) {
1476         error_report_err(local_err);
1477     }
1478 
1479     return spapr->htab_fd;
1480 }
1481 
1482 void close_htab_fd(SpaprMachineState *spapr)
1483 {
1484     if (spapr->htab_fd >= 0) {
1485         close(spapr->htab_fd);
1486     }
1487     spapr->htab_fd = -1;
1488 }
1489 
1490 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1491 {
1492     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1493 
1494     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1495 }
1496 
1497 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1498 {
1499     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1500 
1501     assert(kvm_enabled());
1502 
1503     if (!spapr->htab) {
1504         return 0;
1505     }
1506 
1507     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1508 }
1509 
1510 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1511                                                 hwaddr ptex, int n)
1512 {
1513     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1514     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1515 
1516     if (!spapr->htab) {
1517         /*
1518          * HTAB is controlled by KVM. Fetch into temporary buffer
1519          */
1520         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1521         kvmppc_read_hptes(hptes, ptex, n);
1522         return hptes;
1523     }
1524 
1525     /*
1526      * HTAB is controlled by QEMU. Just point to the internally
1527      * accessible PTEG.
1528      */
1529     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1530 }
1531 
1532 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1533                               const ppc_hash_pte64_t *hptes,
1534                               hwaddr ptex, int n)
1535 {
1536     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1537 
1538     if (!spapr->htab) {
1539         g_free((void *)hptes);
1540     }
1541 
1542     /* Nothing to do for qemu managed HPT */
1543 }
1544 
1545 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1546                       uint64_t pte0, uint64_t pte1)
1547 {
1548     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1549     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1550 
1551     if (!spapr->htab) {
1552         kvmppc_write_hpte(ptex, pte0, pte1);
1553     } else {
1554         if (pte0 & HPTE64_V_VALID) {
1555             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1556             /*
1557              * When setting valid, we write PTE1 first. This ensures
1558              * proper synchronization with the reading code in
1559              * ppc_hash64_pteg_search()
1560              */
1561             smp_wmb();
1562             stq_p(spapr->htab + offset, pte0);
1563         } else {
1564             stq_p(spapr->htab + offset, pte0);
1565             /*
1566              * When clearing it we set PTE0 first. This ensures proper
1567              * synchronization with the reading code in
1568              * ppc_hash64_pteg_search()
1569              */
1570             smp_wmb();
1571             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1572         }
1573     }
1574 }
1575 
1576 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1577                              uint64_t pte1)
1578 {
1579     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1580     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1581 
1582     if (!spapr->htab) {
1583         /* There should always be a hash table when this is called */
1584         error_report("spapr_hpte_set_c called with no hash table !");
1585         return;
1586     }
1587 
1588     /* The HW performs a non-atomic byte update */
1589     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1590 }
1591 
1592 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1593                              uint64_t pte1)
1594 {
1595     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1596     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1597 
1598     if (!spapr->htab) {
1599         /* There should always be a hash table when this is called */
1600         error_report("spapr_hpte_set_r called with no hash table !");
1601         return;
1602     }
1603 
1604     /* The HW performs a non-atomic byte update */
1605     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1606 }
1607 
1608 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1609 {
1610     int shift;
1611 
1612     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1613      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1614      * that's much more than is needed for Linux guests */
1615     shift = ctz64(pow2ceil(ramsize)) - 7;
1616     shift = MAX(shift, 18); /* Minimum architected size */
1617     shift = MIN(shift, 46); /* Maximum architected size */
1618     return shift;
1619 }
1620 
1621 void spapr_free_hpt(SpaprMachineState *spapr)
1622 {
1623     qemu_vfree(spapr->htab);
1624     spapr->htab = NULL;
1625     spapr->htab_shift = 0;
1626     close_htab_fd(spapr);
1627 }
1628 
1629 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1630 {
1631     ERRP_GUARD();
1632     long rc;
1633 
1634     /* Clean up any HPT info from a previous boot */
1635     spapr_free_hpt(spapr);
1636 
1637     rc = kvmppc_reset_htab(shift);
1638 
1639     if (rc == -EOPNOTSUPP) {
1640         error_setg(errp, "HPT not supported in nested guests");
1641         return -EOPNOTSUPP;
1642     }
1643 
1644     if (rc < 0) {
1645         /* kernel-side HPT needed, but couldn't allocate one */
1646         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1647                          shift);
1648         error_append_hint(errp, "Try smaller maxmem?\n");
1649         return -errno;
1650     } else if (rc > 0) {
1651         /* kernel-side HPT allocated */
1652         if (rc != shift) {
1653             error_setg(errp,
1654                        "Requested order %d HPT, but kernel allocated order %ld",
1655                        shift, rc);
1656             error_append_hint(errp, "Try smaller maxmem?\n");
1657             return -ENOSPC;
1658         }
1659 
1660         spapr->htab_shift = shift;
1661         spapr->htab = NULL;
1662     } else {
1663         /* kernel-side HPT not needed, allocate in userspace instead */
1664         size_t size = 1ULL << shift;
1665         int i;
1666 
1667         spapr->htab = qemu_memalign(size, size);
1668         memset(spapr->htab, 0, size);
1669         spapr->htab_shift = shift;
1670 
1671         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1672             DIRTY_HPTE(HPTE(spapr->htab, i));
1673         }
1674     }
1675     /* We're setting up a hash table, so that means we're not radix */
1676     spapr->patb_entry = 0;
1677     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1678     return 0;
1679 }
1680 
1681 void spapr_setup_hpt(SpaprMachineState *spapr)
1682 {
1683     int hpt_shift;
1684 
1685     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1686         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1687     } else {
1688         uint64_t current_ram_size;
1689 
1690         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1691         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1692     }
1693     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1694 
1695     if (kvm_enabled()) {
1696         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1697 
1698         /* Check our RMA fits in the possible VRMA */
1699         if (vrma_limit < spapr->rma_size) {
1700             error_report("Unable to create %" HWADDR_PRIu
1701                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1702                          spapr->rma_size / MiB, vrma_limit / MiB);
1703             exit(EXIT_FAILURE);
1704         }
1705     }
1706 }
1707 
1708 void spapr_check_mmu_mode(bool guest_radix)
1709 {
1710     if (guest_radix) {
1711         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1712             error_report("Guest requested unavailable MMU mode (radix).");
1713             exit(EXIT_FAILURE);
1714         }
1715     } else {
1716         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1717             && !kvmppc_has_cap_mmu_hash_v3()) {
1718             error_report("Guest requested unavailable MMU mode (hash).");
1719             exit(EXIT_FAILURE);
1720         }
1721     }
1722 }
1723 
1724 static void spapr_machine_reset(MachineState *machine, ResetType type)
1725 {
1726     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1727     PowerPCCPU *first_ppc_cpu;
1728     hwaddr fdt_addr;
1729     void *fdt;
1730     int rc;
1731 
1732     if (type != RESET_TYPE_SNAPSHOT_LOAD) {
1733         /*
1734          * Record-replay snapshot load must not consume random, this was
1735          * already replayed from initial machine reset.
1736          */
1737         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1738     }
1739 
1740     if (machine->cgs) {
1741         confidential_guest_kvm_reset(machine->cgs, &error_fatal);
1742     }
1743     spapr_caps_apply(spapr);
1744     spapr_nested_reset(spapr);
1745 
1746     first_ppc_cpu = POWERPC_CPU(first_cpu);
1747     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1748         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1749                               spapr->max_compat_pvr)) {
1750         /*
1751          * If using KVM with radix mode available, VCPUs can be started
1752          * without a HPT because KVM will start them in radix mode.
1753          * Set the GR bit in PATE so that we know there is no HPT.
1754          */
1755         spapr->patb_entry = PATE1_GR;
1756         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1757     } else {
1758         spapr_setup_hpt(spapr);
1759     }
1760 
1761     qemu_devices_reset(type);
1762 
1763     spapr_ovec_cleanup(spapr->ov5_cas);
1764     spapr->ov5_cas = spapr_ovec_new();
1765 
1766     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1767 
1768     /*
1769      * This is fixing some of the default configuration of the XIVE
1770      * devices. To be called after the reset of the machine devices.
1771      */
1772     spapr_irq_reset(spapr, &error_fatal);
1773 
1774     /*
1775      * There is no CAS under qtest. Simulate one to please the code that
1776      * depends on spapr->ov5_cas. This is especially needed to test device
1777      * unplug, so we do that before resetting the DRCs.
1778      */
1779     if (qtest_enabled()) {
1780         spapr_ovec_cleanup(spapr->ov5_cas);
1781         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1782     }
1783 
1784     spapr_nvdimm_finish_flushes();
1785 
1786     /* DRC reset may cause a device to be unplugged. This will cause troubles
1787      * if this device is used by another device (eg, a running vhost backend
1788      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1789      * situations, we reset DRCs after all devices have been reset.
1790      */
1791     spapr_drc_reset_all(spapr);
1792 
1793     spapr_clear_pending_events(spapr);
1794 
1795     /*
1796      * We place the device tree just below either the top of the RMA,
1797      * or just below 2GB, whichever is lower, so that it can be
1798      * processed with 32-bit real mode code if necessary
1799      */
1800     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1801 
1802     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1803     if (spapr->vof) {
1804         spapr_vof_reset(spapr, fdt, &error_fatal);
1805         /*
1806          * Do not pack the FDT as the client may change properties.
1807          * VOF client does not expect the FDT so we do not load it to the VM.
1808          */
1809     } else {
1810         rc = fdt_pack(fdt);
1811         /* Should only fail if we've built a corrupted tree */
1812         assert(rc == 0);
1813 
1814         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1815                                   0, fdt_addr, 0);
1816         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1817     }
1818     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1819 
1820     g_free(spapr->fdt_blob);
1821     spapr->fdt_size = fdt_totalsize(fdt);
1822     spapr->fdt_initial_size = spapr->fdt_size;
1823     spapr->fdt_blob = fdt;
1824 
1825     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1826     machine->fdt = fdt;
1827 
1828     /* Set up the entry state */
1829     first_ppc_cpu->env.gpr[5] = 0;
1830 
1831     spapr->fwnmi_system_reset_addr = -1;
1832     spapr->fwnmi_machine_check_addr = -1;
1833     spapr->fwnmi_machine_check_interlock = -1;
1834 
1835     /* Signal all vCPUs waiting on this condition */
1836     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1837 
1838     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1839 }
1840 
1841 static void spapr_create_nvram(SpaprMachineState *spapr)
1842 {
1843     DeviceState *dev = qdev_new("spapr-nvram");
1844     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1845 
1846     if (dinfo) {
1847         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1848                                 &error_fatal);
1849     }
1850 
1851     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1852 
1853     spapr->nvram = (struct SpaprNvram *)dev;
1854 }
1855 
1856 static void spapr_rtc_create(SpaprMachineState *spapr)
1857 {
1858     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1859                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1860                                        &error_fatal, NULL);
1861     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1862     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1863                               "date");
1864 }
1865 
1866 /* Returns whether we want to use VGA or not */
1867 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1868 {
1869     vga_interface_created = true;
1870     switch (vga_interface_type) {
1871     case VGA_NONE:
1872         return false;
1873     case VGA_DEVICE:
1874         return true;
1875     case VGA_STD:
1876     case VGA_VIRTIO:
1877     case VGA_CIRRUS:
1878         return pci_vga_init(pci_bus) != NULL;
1879     default:
1880         error_setg(errp,
1881                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1882         return false;
1883     }
1884 }
1885 
1886 static int spapr_pre_load(void *opaque)
1887 {
1888     int rc;
1889 
1890     rc = spapr_caps_pre_load(opaque);
1891     if (rc) {
1892         return rc;
1893     }
1894 
1895     return 0;
1896 }
1897 
1898 static int spapr_post_load(void *opaque, int version_id)
1899 {
1900     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1901     int err = 0;
1902 
1903     err = spapr_caps_post_migration(spapr);
1904     if (err) {
1905         return err;
1906     }
1907 
1908     /*
1909      * In earlier versions, there was no separate qdev for the PAPR
1910      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1911      * So when migrating from those versions, poke the incoming offset
1912      * value into the RTC device
1913      */
1914     if (version_id < 3) {
1915         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1916         if (err) {
1917             return err;
1918         }
1919     }
1920 
1921     if (kvm_enabled() && spapr->patb_entry) {
1922         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1923         bool radix = !!(spapr->patb_entry & PATE1_GR);
1924         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1925 
1926         /*
1927          * Update LPCR:HR and UPRT as they may not be set properly in
1928          * the stream
1929          */
1930         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1931                             LPCR_HR | LPCR_UPRT);
1932 
1933         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1934         if (err) {
1935             error_report("Process table config unsupported by the host");
1936             return -EINVAL;
1937         }
1938     }
1939 
1940     err = spapr_irq_post_load(spapr, version_id);
1941     if (err) {
1942         return err;
1943     }
1944 
1945     return err;
1946 }
1947 
1948 static int spapr_pre_save(void *opaque)
1949 {
1950     int rc;
1951 
1952     rc = spapr_caps_pre_save(opaque);
1953     if (rc) {
1954         return rc;
1955     }
1956 
1957     return 0;
1958 }
1959 
1960 static bool version_before_3(void *opaque, int version_id)
1961 {
1962     return version_id < 3;
1963 }
1964 
1965 static bool spapr_pending_events_needed(void *opaque)
1966 {
1967     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1968     return !QTAILQ_EMPTY(&spapr->pending_events);
1969 }
1970 
1971 static const VMStateDescription vmstate_spapr_event_entry = {
1972     .name = "spapr_event_log_entry",
1973     .version_id = 1,
1974     .minimum_version_id = 1,
1975     .fields = (const VMStateField[]) {
1976         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1977         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1978         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1979                                      NULL, extended_length),
1980         VMSTATE_END_OF_LIST()
1981     },
1982 };
1983 
1984 static const VMStateDescription vmstate_spapr_pending_events = {
1985     .name = "spapr_pending_events",
1986     .version_id = 1,
1987     .minimum_version_id = 1,
1988     .needed = spapr_pending_events_needed,
1989     .fields = (const VMStateField[]) {
1990         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1991                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1992         VMSTATE_END_OF_LIST()
1993     },
1994 };
1995 
1996 static bool spapr_ov5_cas_needed(void *opaque)
1997 {
1998     SpaprMachineState *spapr = opaque;
1999     SpaprOptionVector *ov5_mask = spapr_ovec_new();
2000     bool cas_needed;
2001 
2002     /* Prior to the introduction of SpaprOptionVector, we had two option
2003      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
2004      * Both of these options encode machine topology into the device-tree
2005      * in such a way that the now-booted OS should still be able to interact
2006      * appropriately with QEMU regardless of what options were actually
2007      * negotiatied on the source side.
2008      *
2009      * As such, we can avoid migrating the CAS-negotiated options if these
2010      * are the only options available on the current machine/platform.
2011      * Since these are the only options available for pseries-2.7 and
2012      * earlier, this allows us to maintain old->new/new->old migration
2013      * compatibility.
2014      *
2015      * For QEMU 2.8+, there are additional CAS-negotiatable options available
2016      * via default pseries-2.8 machines and explicit command-line parameters.
2017      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
2018      * of the actual CAS-negotiated values to continue working properly. For
2019      * example, availability of memory unplug depends on knowing whether
2020      * OV5_HP_EVT was negotiated via CAS.
2021      *
2022      * Thus, for any cases where the set of available CAS-negotiatable
2023      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2024      * include the CAS-negotiated options in the migration stream, unless
2025      * if they affect boot time behaviour only.
2026      */
2027     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2028     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2029     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2030 
2031     /* We need extra information if we have any bits outside the mask
2032      * defined above */
2033     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
2034 
2035     spapr_ovec_cleanup(ov5_mask);
2036 
2037     return cas_needed;
2038 }
2039 
2040 static const VMStateDescription vmstate_spapr_ov5_cas = {
2041     .name = "spapr_option_vector_ov5_cas",
2042     .version_id = 1,
2043     .minimum_version_id = 1,
2044     .needed = spapr_ov5_cas_needed,
2045     .fields = (const VMStateField[]) {
2046         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2047                                  vmstate_spapr_ovec, SpaprOptionVector),
2048         VMSTATE_END_OF_LIST()
2049     },
2050 };
2051 
2052 static bool spapr_patb_entry_needed(void *opaque)
2053 {
2054     SpaprMachineState *spapr = opaque;
2055 
2056     return !!spapr->patb_entry;
2057 }
2058 
2059 static const VMStateDescription vmstate_spapr_patb_entry = {
2060     .name = "spapr_patb_entry",
2061     .version_id = 1,
2062     .minimum_version_id = 1,
2063     .needed = spapr_patb_entry_needed,
2064     .fields = (const VMStateField[]) {
2065         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2066         VMSTATE_END_OF_LIST()
2067     },
2068 };
2069 
2070 static bool spapr_irq_map_needed(void *opaque)
2071 {
2072     SpaprMachineState *spapr = opaque;
2073 
2074     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2075 }
2076 
2077 static const VMStateDescription vmstate_spapr_irq_map = {
2078     .name = "spapr_irq_map",
2079     .version_id = 1,
2080     .minimum_version_id = 1,
2081     .needed = spapr_irq_map_needed,
2082     .fields = (const VMStateField[]) {
2083         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2084         VMSTATE_END_OF_LIST()
2085     },
2086 };
2087 
2088 static bool spapr_dtb_needed(void *opaque)
2089 {
2090     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2091 
2092     return smc->update_dt_enabled;
2093 }
2094 
2095 static int spapr_dtb_pre_load(void *opaque)
2096 {
2097     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2098 
2099     g_free(spapr->fdt_blob);
2100     spapr->fdt_blob = NULL;
2101     spapr->fdt_size = 0;
2102 
2103     return 0;
2104 }
2105 
2106 static const VMStateDescription vmstate_spapr_dtb = {
2107     .name = "spapr_dtb",
2108     .version_id = 1,
2109     .minimum_version_id = 1,
2110     .needed = spapr_dtb_needed,
2111     .pre_load = spapr_dtb_pre_load,
2112     .fields = (const VMStateField[]) {
2113         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2114         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2115         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2116                                      fdt_size),
2117         VMSTATE_END_OF_LIST()
2118     },
2119 };
2120 
2121 static bool spapr_fwnmi_needed(void *opaque)
2122 {
2123     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2124 
2125     return spapr->fwnmi_machine_check_addr != -1;
2126 }
2127 
2128 static int spapr_fwnmi_pre_save(void *opaque)
2129 {
2130     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2131 
2132     /*
2133      * Check if machine check handling is in progress and print a
2134      * warning message.
2135      */
2136     if (spapr->fwnmi_machine_check_interlock != -1) {
2137         warn_report("A machine check is being handled during migration. The"
2138                 "handler may run and log hardware error on the destination");
2139     }
2140 
2141     return 0;
2142 }
2143 
2144 static const VMStateDescription vmstate_spapr_fwnmi = {
2145     .name = "spapr_fwnmi",
2146     .version_id = 1,
2147     .minimum_version_id = 1,
2148     .needed = spapr_fwnmi_needed,
2149     .pre_save = spapr_fwnmi_pre_save,
2150     .fields = (const VMStateField[]) {
2151         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2152         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2153         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2154         VMSTATE_END_OF_LIST()
2155     },
2156 };
2157 
2158 static const VMStateDescription vmstate_spapr = {
2159     .name = "spapr",
2160     .version_id = 3,
2161     .minimum_version_id = 1,
2162     .pre_load = spapr_pre_load,
2163     .post_load = spapr_post_load,
2164     .pre_save = spapr_pre_save,
2165     .fields = (const VMStateField[]) {
2166         /* used to be @next_irq */
2167         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2168 
2169         /* RTC offset */
2170         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2171 
2172         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2173         VMSTATE_END_OF_LIST()
2174     },
2175     .subsections = (const VMStateDescription * const []) {
2176         &vmstate_spapr_ov5_cas,
2177         &vmstate_spapr_patb_entry,
2178         &vmstate_spapr_pending_events,
2179         &vmstate_spapr_cap_htm,
2180         &vmstate_spapr_cap_vsx,
2181         &vmstate_spapr_cap_dfp,
2182         &vmstate_spapr_cap_cfpc,
2183         &vmstate_spapr_cap_sbbc,
2184         &vmstate_spapr_cap_ibs,
2185         &vmstate_spapr_cap_hpt_maxpagesize,
2186         &vmstate_spapr_irq_map,
2187         &vmstate_spapr_cap_nested_kvm_hv,
2188         &vmstate_spapr_dtb,
2189         &vmstate_spapr_cap_large_decr,
2190         &vmstate_spapr_cap_ccf_assist,
2191         &vmstate_spapr_cap_fwnmi,
2192         &vmstate_spapr_fwnmi,
2193         &vmstate_spapr_cap_rpt_invalidate,
2194         &vmstate_spapr_cap_ail_mode_3,
2195         &vmstate_spapr_cap_nested_papr,
2196         NULL
2197     }
2198 };
2199 
2200 static int htab_save_setup(QEMUFile *f, void *opaque, Error **errp)
2201 {
2202     SpaprMachineState *spapr = opaque;
2203 
2204     /* "Iteration" header */
2205     if (!spapr->htab_shift) {
2206         qemu_put_be32(f, -1);
2207     } else {
2208         qemu_put_be32(f, spapr->htab_shift);
2209     }
2210 
2211     if (spapr->htab) {
2212         spapr->htab_save_index = 0;
2213         spapr->htab_first_pass = true;
2214     } else {
2215         if (spapr->htab_shift) {
2216             assert(kvm_enabled());
2217         }
2218     }
2219 
2220 
2221     return 0;
2222 }
2223 
2224 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2225                             int chunkstart, int n_valid, int n_invalid)
2226 {
2227     qemu_put_be32(f, chunkstart);
2228     qemu_put_be16(f, n_valid);
2229     qemu_put_be16(f, n_invalid);
2230     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2231                     HASH_PTE_SIZE_64 * n_valid);
2232 }
2233 
2234 static void htab_save_end_marker(QEMUFile *f)
2235 {
2236     qemu_put_be32(f, 0);
2237     qemu_put_be16(f, 0);
2238     qemu_put_be16(f, 0);
2239 }
2240 
2241 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2242                                  int64_t max_ns)
2243 {
2244     bool has_timeout = max_ns != -1;
2245     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2246     int index = spapr->htab_save_index;
2247     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2248 
2249     assert(spapr->htab_first_pass);
2250 
2251     do {
2252         int chunkstart;
2253 
2254         /* Consume invalid HPTEs */
2255         while ((index < htabslots)
2256                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2257             CLEAN_HPTE(HPTE(spapr->htab, index));
2258             index++;
2259         }
2260 
2261         /* Consume valid HPTEs */
2262         chunkstart = index;
2263         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2264                && HPTE_VALID(HPTE(spapr->htab, index))) {
2265             CLEAN_HPTE(HPTE(spapr->htab, index));
2266             index++;
2267         }
2268 
2269         if (index > chunkstart) {
2270             int n_valid = index - chunkstart;
2271 
2272             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2273 
2274             if (has_timeout &&
2275                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2276                 break;
2277             }
2278         }
2279     } while ((index < htabslots) && !migration_rate_exceeded(f));
2280 
2281     if (index >= htabslots) {
2282         assert(index == htabslots);
2283         index = 0;
2284         spapr->htab_first_pass = false;
2285     }
2286     spapr->htab_save_index = index;
2287 }
2288 
2289 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2290                                 int64_t max_ns)
2291 {
2292     bool final = max_ns < 0;
2293     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2294     int examined = 0, sent = 0;
2295     int index = spapr->htab_save_index;
2296     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2297 
2298     assert(!spapr->htab_first_pass);
2299 
2300     do {
2301         int chunkstart, invalidstart;
2302 
2303         /* Consume non-dirty HPTEs */
2304         while ((index < htabslots)
2305                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2306             index++;
2307             examined++;
2308         }
2309 
2310         chunkstart = index;
2311         /* Consume valid dirty HPTEs */
2312         while ((index < htabslots) && (index - chunkstart < 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         invalidstart = index;
2321         /* Consume invalid dirty HPTEs */
2322         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2323                && HPTE_DIRTY(HPTE(spapr->htab, index))
2324                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2325             CLEAN_HPTE(HPTE(spapr->htab, index));
2326             index++;
2327             examined++;
2328         }
2329 
2330         if (index > chunkstart) {
2331             int n_valid = invalidstart - chunkstart;
2332             int n_invalid = index - invalidstart;
2333 
2334             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2335             sent += index - chunkstart;
2336 
2337             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2338                 break;
2339             }
2340         }
2341 
2342         if (examined >= htabslots) {
2343             break;
2344         }
2345 
2346         if (index >= htabslots) {
2347             assert(index == htabslots);
2348             index = 0;
2349         }
2350     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2351 
2352     if (index >= htabslots) {
2353         assert(index == htabslots);
2354         index = 0;
2355     }
2356 
2357     spapr->htab_save_index = index;
2358 
2359     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2360 }
2361 
2362 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2363 #define MAX_KVM_BUF_SIZE    2048
2364 
2365 static int htab_save_iterate(QEMUFile *f, void *opaque)
2366 {
2367     SpaprMachineState *spapr = opaque;
2368     int fd;
2369     int rc = 0;
2370 
2371     /* Iteration header */
2372     if (!spapr->htab_shift) {
2373         qemu_put_be32(f, -1);
2374         return 1;
2375     } else {
2376         qemu_put_be32(f, 0);
2377     }
2378 
2379     if (!spapr->htab) {
2380         assert(kvm_enabled());
2381 
2382         fd = get_htab_fd(spapr);
2383         if (fd < 0) {
2384             return fd;
2385         }
2386 
2387         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2388         if (rc < 0) {
2389             return rc;
2390         }
2391     } else  if (spapr->htab_first_pass) {
2392         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2393     } else {
2394         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2395     }
2396 
2397     htab_save_end_marker(f);
2398 
2399     return rc;
2400 }
2401 
2402 static int htab_save_complete(QEMUFile *f, void *opaque)
2403 {
2404     SpaprMachineState *spapr = opaque;
2405     int fd;
2406 
2407     /* Iteration header */
2408     if (!spapr->htab_shift) {
2409         qemu_put_be32(f, -1);
2410         return 0;
2411     } else {
2412         qemu_put_be32(f, 0);
2413     }
2414 
2415     if (!spapr->htab) {
2416         int rc;
2417 
2418         assert(kvm_enabled());
2419 
2420         fd = get_htab_fd(spapr);
2421         if (fd < 0) {
2422             return fd;
2423         }
2424 
2425         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2426         if (rc < 0) {
2427             return rc;
2428         }
2429     } else {
2430         if (spapr->htab_first_pass) {
2431             htab_save_first_pass(f, spapr, -1);
2432         }
2433         htab_save_later_pass(f, spapr, -1);
2434     }
2435 
2436     /* End marker */
2437     htab_save_end_marker(f);
2438 
2439     return 0;
2440 }
2441 
2442 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2443 {
2444     SpaprMachineState *spapr = opaque;
2445     uint32_t section_hdr;
2446     int fd = -1;
2447     Error *local_err = NULL;
2448 
2449     if (version_id < 1 || version_id > 1) {
2450         error_report("htab_load() bad version");
2451         return -EINVAL;
2452     }
2453 
2454     section_hdr = qemu_get_be32(f);
2455 
2456     if (section_hdr == -1) {
2457         spapr_free_hpt(spapr);
2458         return 0;
2459     }
2460 
2461     if (section_hdr) {
2462         int ret;
2463 
2464         /* First section gives the htab size */
2465         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2466         if (ret < 0) {
2467             error_report_err(local_err);
2468             return ret;
2469         }
2470         return 0;
2471     }
2472 
2473     if (!spapr->htab) {
2474         assert(kvm_enabled());
2475 
2476         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2477         if (fd < 0) {
2478             error_report_err(local_err);
2479             return fd;
2480         }
2481     }
2482 
2483     while (true) {
2484         uint32_t index;
2485         uint16_t n_valid, n_invalid;
2486 
2487         index = qemu_get_be32(f);
2488         n_valid = qemu_get_be16(f);
2489         n_invalid = qemu_get_be16(f);
2490 
2491         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2492             /* End of Stream */
2493             break;
2494         }
2495 
2496         if ((index + n_valid + n_invalid) >
2497             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2498             /* Bad index in stream */
2499             error_report(
2500                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2501                 index, n_valid, n_invalid, spapr->htab_shift);
2502             return -EINVAL;
2503         }
2504 
2505         if (spapr->htab) {
2506             if (n_valid) {
2507                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2508                                 HASH_PTE_SIZE_64 * n_valid);
2509             }
2510             if (n_invalid) {
2511                 memset(HPTE(spapr->htab, index + n_valid), 0,
2512                        HASH_PTE_SIZE_64 * n_invalid);
2513             }
2514         } else {
2515             int rc;
2516 
2517             assert(fd >= 0);
2518 
2519             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2520                                         &local_err);
2521             if (rc < 0) {
2522                 error_report_err(local_err);
2523                 return rc;
2524             }
2525         }
2526     }
2527 
2528     if (!spapr->htab) {
2529         assert(fd >= 0);
2530         close(fd);
2531     }
2532 
2533     return 0;
2534 }
2535 
2536 static void htab_save_cleanup(void *opaque)
2537 {
2538     SpaprMachineState *spapr = opaque;
2539 
2540     close_htab_fd(spapr);
2541 }
2542 
2543 static SaveVMHandlers savevm_htab_handlers = {
2544     .save_setup = htab_save_setup,
2545     .save_live_iterate = htab_save_iterate,
2546     .save_live_complete_precopy = htab_save_complete,
2547     .save_cleanup = htab_save_cleanup,
2548     .load_state = htab_load,
2549 };
2550 
2551 static void spapr_boot_set(void *opaque, const char *boot_device,
2552                            Error **errp)
2553 {
2554     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2555 
2556     g_free(spapr->boot_device);
2557     spapr->boot_device = g_strdup(boot_device);
2558 }
2559 
2560 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2561 {
2562     MachineState *machine = MACHINE(spapr);
2563     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2564     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2565     int i;
2566 
2567     g_assert(!nr_lmbs || machine->device_memory);
2568     for (i = 0; i < nr_lmbs; i++) {
2569         uint64_t addr;
2570 
2571         addr = i * lmb_size + machine->device_memory->base;
2572         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2573                                addr / lmb_size);
2574     }
2575 }
2576 
2577 /*
2578  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2579  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2580  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2581  */
2582 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2583 {
2584     int i;
2585 
2586     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2587         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2588                    " is not aligned to %" PRIu64 " MiB",
2589                    machine->ram_size,
2590                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2591         return;
2592     }
2593 
2594     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2595         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2596                    " is not aligned to %" PRIu64 " MiB",
2597                    machine->ram_size,
2598                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2599         return;
2600     }
2601 
2602     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2603         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2604             error_setg(errp,
2605                        "Node %d memory size 0x%" PRIx64
2606                        " is not aligned to %" PRIu64 " MiB",
2607                        i, machine->numa_state->nodes[i].node_mem,
2608                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2609             return;
2610         }
2611     }
2612 }
2613 
2614 /* find cpu slot in machine->possible_cpus by core_id */
2615 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2616 {
2617     int index = id / ms->smp.threads;
2618 
2619     if (index >= ms->possible_cpus->len) {
2620         return NULL;
2621     }
2622     if (idx) {
2623         *idx = index;
2624     }
2625     return &ms->possible_cpus->cpus[index];
2626 }
2627 
2628 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2629 {
2630     MachineState *ms = MACHINE(spapr);
2631     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2632     Error *local_err = NULL;
2633     bool vsmt_user = !!spapr->vsmt;
2634     int kvm_smt = kvmppc_smt_threads();
2635     int ret;
2636     unsigned int smp_threads = ms->smp.threads;
2637 
2638     if (tcg_enabled()) {
2639         if (smp_threads > 1 &&
2640             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2641                                    spapr->max_compat_pvr)) {
2642             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2643             return;
2644         }
2645 
2646         if (smp_threads > 8) {
2647             error_setg(errp, "TCG cannot support more than 8 threads/core "
2648                        "on a pseries machine");
2649             return;
2650         }
2651     }
2652     if (!is_power_of_2(smp_threads)) {
2653         error_setg(errp, "Cannot support %d threads/core on a pseries "
2654                    "machine because it must be a power of 2", smp_threads);
2655         return;
2656     }
2657 
2658     /* Determine the VSMT mode to use: */
2659     if (vsmt_user) {
2660         if (spapr->vsmt < smp_threads) {
2661             error_setg(errp, "Cannot support VSMT mode %d"
2662                        " because it must be >= threads/core (%d)",
2663                        spapr->vsmt, smp_threads);
2664             return;
2665         }
2666         /* In this case, spapr->vsmt has been set by the command line */
2667     } else if (!smc->smp_threads_vsmt) {
2668         /*
2669          * Default VSMT value is tricky, because we need it to be as
2670          * consistent as possible (for migration), but this requires
2671          * changing it for at least some existing cases.  We pick 8 as
2672          * the value that we'd get with KVM on POWER8, the
2673          * overwhelmingly common case in production systems.
2674          */
2675         spapr->vsmt = MAX(8, smp_threads);
2676     } else {
2677         spapr->vsmt = smp_threads;
2678     }
2679 
2680     /* KVM: If necessary, set the SMT mode: */
2681     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2682         ret = kvmppc_set_smt_threads(spapr->vsmt);
2683         if (ret) {
2684             /* Looks like KVM isn't able to change VSMT mode */
2685             error_setg(&local_err,
2686                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2687                        spapr->vsmt, ret);
2688             /* We can live with that if the default one is big enough
2689              * for the number of threads, and a submultiple of the one
2690              * we want.  In this case we'll waste some vcpu ids, but
2691              * behaviour will be correct */
2692             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2693                 warn_report_err(local_err);
2694             } else {
2695                 if (!vsmt_user) {
2696                     error_append_hint(&local_err,
2697                                       "On PPC, a VM with %d threads/core"
2698                                       " on a host with %d threads/core"
2699                                       " requires the use of VSMT mode %d.\n",
2700                                       smp_threads, kvm_smt, spapr->vsmt);
2701                 }
2702                 kvmppc_error_append_smt_possible_hint(&local_err);
2703                 error_propagate(errp, local_err);
2704             }
2705         }
2706     }
2707     /* else TCG: nothing to do currently */
2708 }
2709 
2710 static void spapr_init_cpus(SpaprMachineState *spapr)
2711 {
2712     MachineState *machine = MACHINE(spapr);
2713     MachineClass *mc = MACHINE_GET_CLASS(machine);
2714     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2715     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2716     const CPUArchIdList *possible_cpus;
2717     unsigned int smp_cpus = machine->smp.cpus;
2718     unsigned int smp_threads = machine->smp.threads;
2719     unsigned int max_cpus = machine->smp.max_cpus;
2720     int boot_cores_nr = smp_cpus / smp_threads;
2721     int i;
2722 
2723     possible_cpus = mc->possible_cpu_arch_ids(machine);
2724     if (mc->has_hotpluggable_cpus) {
2725         if (smp_cpus % smp_threads) {
2726             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2727                          smp_cpus, smp_threads);
2728             exit(1);
2729         }
2730         if (max_cpus % smp_threads) {
2731             error_report("max_cpus (%u) must be multiple of threads (%u)",
2732                          max_cpus, smp_threads);
2733             exit(1);
2734         }
2735     } else {
2736         if (max_cpus != smp_cpus) {
2737             error_report("This machine version does not support CPU hotplug");
2738             exit(1);
2739         }
2740         boot_cores_nr = possible_cpus->len;
2741     }
2742 
2743     if (smc->pre_2_10_has_unused_icps) {
2744         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2745             /* Dummy entries get deregistered when real ICPState objects
2746              * are registered during CPU core hotplug.
2747              */
2748             pre_2_10_vmstate_register_dummy_icp(i);
2749         }
2750     }
2751 
2752     for (i = 0; i < possible_cpus->len; i++) {
2753         int core_id = i * smp_threads;
2754 
2755         if (mc->has_hotpluggable_cpus) {
2756             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2757                                    spapr_vcpu_id(spapr, core_id));
2758         }
2759 
2760         if (i < boot_cores_nr) {
2761             Object *core  = object_new(type);
2762             int nr_threads = smp_threads;
2763 
2764             /* Handle the partially filled core for older machine types */
2765             if ((i + 1) * smp_threads >= smp_cpus) {
2766                 nr_threads = smp_cpus - i * smp_threads;
2767             }
2768 
2769             object_property_set_int(core, "nr-threads", nr_threads,
2770                                     &error_fatal);
2771             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2772                                     &error_fatal);
2773             qdev_realize(DEVICE(core), NULL, &error_fatal);
2774 
2775             object_unref(core);
2776         }
2777     }
2778 }
2779 
2780 static PCIHostState *spapr_create_default_phb(void)
2781 {
2782     DeviceState *dev;
2783 
2784     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2785     qdev_prop_set_uint32(dev, "index", 0);
2786     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2787 
2788     return PCI_HOST_BRIDGE(dev);
2789 }
2790 
2791 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2792 {
2793     MachineState *machine = MACHINE(spapr);
2794     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2795     hwaddr rma_size = machine->ram_size;
2796     hwaddr node0_size = spapr_node0_size(machine);
2797 
2798     /* RMA has to fit in the first NUMA node */
2799     rma_size = MIN(rma_size, node0_size);
2800 
2801     /*
2802      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2803      * never exceed that
2804      */
2805     rma_size = MIN(rma_size, 1 * TiB);
2806 
2807     /*
2808      * Clamp the RMA size based on machine type.  This is for
2809      * migration compatibility with older qemu versions, which limited
2810      * the RMA size for complicated and mostly bad reasons.
2811      */
2812     if (smc->rma_limit) {
2813         rma_size = MIN(rma_size, smc->rma_limit);
2814     }
2815 
2816     if (rma_size < MIN_RMA_SLOF) {
2817         error_setg(errp,
2818                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2819                    "ldMiB guest RMA (Real Mode Area memory)",
2820                    MIN_RMA_SLOF / MiB);
2821         return 0;
2822     }
2823 
2824     return rma_size;
2825 }
2826 
2827 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2828 {
2829     MachineState *machine = MACHINE(spapr);
2830     int i;
2831 
2832     for (i = 0; i < machine->ram_slots; i++) {
2833         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2834     }
2835 }
2836 
2837 /* pSeries LPAR / sPAPR hardware init */
2838 static void spapr_machine_init(MachineState *machine)
2839 {
2840     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2841     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2842     MachineClass *mc = MACHINE_GET_CLASS(machine);
2843     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2844     const char *bios_name = machine->firmware ?: bios_default;
2845     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2846     const char *kernel_filename = machine->kernel_filename;
2847     const char *initrd_filename = machine->initrd_filename;
2848     PCIHostState *phb;
2849     bool has_vga;
2850     int i;
2851     MemoryRegion *sysmem = get_system_memory();
2852     long load_limit, fw_size;
2853     Error *resize_hpt_err = NULL;
2854     NICInfo *nd;
2855 
2856     if (!filename) {
2857         error_report("Could not find LPAR firmware '%s'", bios_name);
2858         exit(1);
2859     }
2860     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2861     if (fw_size <= 0) {
2862         error_report("Could not load LPAR firmware '%s'", filename);
2863         exit(1);
2864     }
2865 
2866     /*
2867      * if Secure VM (PEF) support is configured, then initialize it
2868      */
2869     if (machine->cgs) {
2870         confidential_guest_kvm_init(machine->cgs, &error_fatal);
2871     }
2872 
2873     msi_nonbroken = true;
2874 
2875     QLIST_INIT(&spapr->phbs);
2876     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2877 
2878     /* Determine capabilities to run with */
2879     spapr_caps_init(spapr);
2880 
2881     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2882     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2883         /*
2884          * If the user explicitly requested a mode we should either
2885          * supply it, or fail completely (which we do below).  But if
2886          * it's not set explicitly, we reset our mode to something
2887          * that works
2888          */
2889         if (resize_hpt_err) {
2890             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2891             error_free(resize_hpt_err);
2892             resize_hpt_err = NULL;
2893         } else {
2894             spapr->resize_hpt = smc->resize_hpt_default;
2895         }
2896     }
2897 
2898     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2899 
2900     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2901         /*
2902          * User requested HPT resize, but this host can't supply it.  Bail out
2903          */
2904         error_report_err(resize_hpt_err);
2905         exit(1);
2906     }
2907     error_free(resize_hpt_err);
2908 
2909     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2910 
2911     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2912     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2913 
2914     /*
2915      * VSMT must be set in order to be able to compute VCPU ids, ie to
2916      * call spapr_max_server_number() or spapr_vcpu_id().
2917      */
2918     spapr_set_vsmt_mode(spapr, &error_fatal);
2919 
2920     /* Set up Interrupt Controller before we create the VCPUs */
2921     spapr_irq_init(spapr, &error_fatal);
2922 
2923     /* Set up containers for ibm,client-architecture-support negotiated options
2924      */
2925     spapr->ov5 = spapr_ovec_new();
2926     spapr->ov5_cas = spapr_ovec_new();
2927 
2928     spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2929     spapr_validate_node_memory(machine, &error_fatal);
2930 
2931     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2932 
2933     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2934     if (!smc->pre_6_2_numa_affinity) {
2935         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2936     }
2937 
2938     /* advertise support for dedicated HP event source to guests */
2939     if (spapr->use_hotplug_event_source) {
2940         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2941     }
2942 
2943     /* advertise support for HPT resizing */
2944     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2945         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2946     }
2947 
2948     /* advertise support for ibm,dyamic-memory-v2 */
2949     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2950 
2951     /* advertise XIVE on POWER9 machines */
2952     if (spapr->irq->xive) {
2953         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2954     }
2955 
2956     /* init CPUs */
2957     spapr_init_cpus(spapr);
2958 
2959     /* Init numa_assoc_array */
2960     spapr_numa_associativity_init(spapr, machine);
2961 
2962     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2963         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2964                               spapr->max_compat_pvr)) {
2965         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2966         /* KVM and TCG always allow GTSE with radix... */
2967         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2968     }
2969     /* ... but not with hash (currently). */
2970 
2971     if (kvm_enabled()) {
2972         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2973         kvmppc_enable_logical_ci_hcalls();
2974         kvmppc_enable_set_mode_hcall();
2975 
2976         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2977         kvmppc_enable_clear_ref_mod_hcalls();
2978 
2979         /* Enable H_PAGE_INIT */
2980         kvmppc_enable_h_page_init();
2981     }
2982 
2983     /* map RAM */
2984     memory_region_add_subregion(sysmem, 0, machine->ram);
2985 
2986     /* initialize hotplug memory address space */
2987     if (machine->ram_size < machine->maxram_size) {
2988         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2989         hwaddr device_mem_base;
2990 
2991         /*
2992          * Limit the number of hotpluggable memory slots to half the number
2993          * slots that KVM supports, leaving the other half for PCI and other
2994          * devices. However ensure that number of slots doesn't drop below 32.
2995          */
2996         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2997                            SPAPR_MAX_RAM_SLOTS;
2998 
2999         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
3000             max_memslots = SPAPR_MAX_RAM_SLOTS;
3001         }
3002         if (machine->ram_slots > max_memslots) {
3003             error_report("Specified number of memory slots %"
3004                          PRIu64" exceeds max supported %d",
3005                          machine->ram_slots, max_memslots);
3006             exit(1);
3007         }
3008 
3009         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
3010         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
3011     }
3012 
3013     spapr_create_lmb_dr_connectors(spapr);
3014 
3015     if (mc->nvdimm_supported) {
3016         spapr_create_nvdimm_dr_connectors(spapr);
3017     }
3018 
3019     /* Set up RTAS event infrastructure */
3020     spapr_events_init(spapr);
3021 
3022     /* Set up the RTC RTAS interfaces */
3023     spapr_rtc_create(spapr);
3024 
3025     /* Set up VIO bus */
3026     spapr->vio_bus = spapr_vio_bus_init();
3027 
3028     for (i = 0; serial_hd(i); i++) {
3029         spapr_vty_create(spapr->vio_bus, serial_hd(i));
3030     }
3031 
3032     /* We always have at least the nvram device on VIO */
3033     spapr_create_nvram(spapr);
3034 
3035     /*
3036      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3037      * connectors (described in root DT node's "ibm,drc-types" property)
3038      * are pre-initialized here. additional child connectors (such as
3039      * connectors for a PHBs PCI slots) are added as needed during their
3040      * parent's realization.
3041      */
3042     if (smc->dr_phb_enabled) {
3043         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3044             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3045         }
3046     }
3047 
3048     /* Set up PCI */
3049     spapr_pci_rtas_init();
3050 
3051     phb = spapr_create_default_phb();
3052 
3053     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3054         spapr_vlan_create(spapr->vio_bus, nd);
3055     }
3056 
3057     pci_init_nic_devices(phb->bus, NULL);
3058 
3059     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3060         spapr_vscsi_create(spapr->vio_bus);
3061     }
3062 
3063     /* Graphics */
3064     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3065     if (has_vga) {
3066         spapr->want_stdout_path = !machine->enable_graphics;
3067         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3068     } else {
3069         spapr->want_stdout_path = true;
3070     }
3071 
3072     if (machine->usb) {
3073         if (smc->use_ohci_by_default) {
3074             pci_create_simple(phb->bus, -1, "pci-ohci");
3075         } else {
3076             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3077         }
3078 
3079         if (has_vga) {
3080             USBBus *usb_bus;
3081 
3082             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3083                                                               &error_abort));
3084             usb_create_simple(usb_bus, "usb-kbd");
3085             usb_create_simple(usb_bus, "usb-mouse");
3086         }
3087     }
3088 
3089     if (kernel_filename) {
3090         uint64_t loaded_addr = 0;
3091 
3092         spapr->kernel_size = load_elf(kernel_filename, NULL,
3093                                       translate_kernel_address, spapr,
3094                                       NULL, &loaded_addr, NULL, NULL, 1,
3095                                       PPC_ELF_MACHINE, 0, 0);
3096         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3097             spapr->kernel_size = load_elf(kernel_filename, NULL,
3098                                           translate_kernel_address, spapr,
3099                                           NULL, &loaded_addr, NULL, NULL, 0,
3100                                           PPC_ELF_MACHINE, 0, 0);
3101             spapr->kernel_le = spapr->kernel_size > 0;
3102         }
3103         if (spapr->kernel_size < 0) {
3104             error_report("error loading %s: %s", kernel_filename,
3105                          load_elf_strerror(spapr->kernel_size));
3106             exit(1);
3107         }
3108 
3109         if (spapr->kernel_addr != loaded_addr) {
3110             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3111                         " to 0x%"PRIx64,
3112                         spapr->kernel_addr, loaded_addr);
3113             spapr->kernel_addr = loaded_addr;
3114         }
3115 
3116         /* load initrd */
3117         if (initrd_filename) {
3118             /* Try to locate the initrd in the gap between the kernel
3119              * and the firmware. Add a bit of space just in case
3120              */
3121             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3122                                   + 0x1ffff) & ~0xffff;
3123             spapr->initrd_size = load_image_targphys(initrd_filename,
3124                                                      spapr->initrd_base,
3125                                                      load_limit
3126                                                      - spapr->initrd_base);
3127             if (spapr->initrd_size < 0) {
3128                 error_report("could not load initial ram disk '%s'",
3129                              initrd_filename);
3130                 exit(1);
3131             }
3132         }
3133     }
3134 
3135     /* FIXME: Should register things through the MachineState's qdev
3136      * interface, this is a legacy from the sPAPREnvironment structure
3137      * which predated MachineState but had a similar function */
3138     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3139     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3140                          &savevm_htab_handlers, spapr);
3141 
3142     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3143 
3144     qemu_register_boot_set(spapr_boot_set, spapr);
3145 
3146     /*
3147      * Nothing needs to be done to resume a suspended guest because
3148      * suspending does not change the machine state, so no need for
3149      * a ->wakeup method.
3150      */
3151     qemu_register_wakeup_support();
3152 
3153     if (kvm_enabled()) {
3154         /* to stop and start vmclock */
3155         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3156                                          &spapr->tb);
3157 
3158         kvmppc_spapr_enable_inkernel_multitce();
3159     }
3160 
3161     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3162     if (spapr->vof) {
3163         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3164         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3165     }
3166 
3167     spapr_watchdog_init(spapr);
3168 }
3169 
3170 #define DEFAULT_KVM_TYPE "auto"
3171 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3172 {
3173     /*
3174      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3175      * accommodate the 'HV' and 'PV' formats that exists in the
3176      * wild. The 'auto' mode is being introduced already as
3177      * lower-case, thus we don't need to bother checking for
3178      * "AUTO".
3179      */
3180     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3181         return 0;
3182     }
3183 
3184     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3185         return 1;
3186     }
3187 
3188     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3189         return 2;
3190     }
3191 
3192     error_report("Unknown kvm-type specified '%s'", vm_type);
3193     return -1;
3194 }
3195 
3196 /*
3197  * Implementation of an interface to adjust firmware path
3198  * for the bootindex property handling.
3199  */
3200 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3201                                    DeviceState *dev)
3202 {
3203 #define CAST(type, obj, name) \
3204     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3205     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3206     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3207     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3208     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3209 
3210     if (d && bus) {
3211         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3212         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3213         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3214 
3215         if (spapr) {
3216             /*
3217              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3218              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3219              * 0x8000 | (target << 8) | (bus << 5) | lun
3220              * (see the "Logical unit addressing format" table in SAM5)
3221              */
3222             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3223             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3224                                    (uint64_t)id << 48);
3225         } else if (virtio) {
3226             /*
3227              * We use SRP luns of the form 01000000 | (target << 8) | lun
3228              * in the top 32 bits of the 64-bit LUN
3229              * Note: the quote above is from SLOF and it is wrong,
3230              * the actual binding is:
3231              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3232              */
3233             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3234             if (d->lun >= 256) {
3235                 /* Use the LUN "flat space addressing method" */
3236                 id |= 0x4000;
3237             }
3238             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3239                                    (uint64_t)id << 32);
3240         } else if (usb) {
3241             /*
3242              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3243              * in the top 32 bits of the 64-bit LUN
3244              */
3245             unsigned usb_port = atoi(usb->port->path);
3246             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3247             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3248                                    (uint64_t)id << 32);
3249         }
3250     }
3251 
3252     /*
3253      * SLOF probes the USB devices, and if it recognizes that the device is a
3254      * storage device, it changes its name to "storage" instead of "usb-host",
3255      * and additionally adds a child node for the SCSI LUN, so the correct
3256      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3257      */
3258     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3259         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3260         if (usb_device_is_scsi_storage(usbdev)) {
3261             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3262         }
3263     }
3264 
3265     if (phb) {
3266         /* Replace "pci" with "pci@800000020000000" */
3267         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3268     }
3269 
3270     if (vsc) {
3271         /* Same logic as virtio above */
3272         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3273         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3274     }
3275 
3276     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3277         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3278         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3279         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3280     }
3281 
3282     if (pcidev) {
3283         return spapr_pci_fw_dev_name(pcidev);
3284     }
3285 
3286     return NULL;
3287 }
3288 
3289 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3290 {
3291     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3292 
3293     return g_strdup(spapr->kvm_type);
3294 }
3295 
3296 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3297 {
3298     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3299 
3300     g_free(spapr->kvm_type);
3301     spapr->kvm_type = g_strdup(value);
3302 }
3303 
3304 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3305 {
3306     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3307 
3308     return spapr->use_hotplug_event_source;
3309 }
3310 
3311 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3312                                             Error **errp)
3313 {
3314     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3315 
3316     spapr->use_hotplug_event_source = value;
3317 }
3318 
3319 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3320 {
3321     return true;
3322 }
3323 
3324 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3325 {
3326     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3327 
3328     switch (spapr->resize_hpt) {
3329     case SPAPR_RESIZE_HPT_DEFAULT:
3330         return g_strdup("default");
3331     case SPAPR_RESIZE_HPT_DISABLED:
3332         return g_strdup("disabled");
3333     case SPAPR_RESIZE_HPT_ENABLED:
3334         return g_strdup("enabled");
3335     case SPAPR_RESIZE_HPT_REQUIRED:
3336         return g_strdup("required");
3337     }
3338     g_assert_not_reached();
3339 }
3340 
3341 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3342 {
3343     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3344 
3345     if (strcmp(value, "default") == 0) {
3346         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3347     } else if (strcmp(value, "disabled") == 0) {
3348         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3349     } else if (strcmp(value, "enabled") == 0) {
3350         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3351     } else if (strcmp(value, "required") == 0) {
3352         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3353     } else {
3354         error_setg(errp, "Bad value for \"resize-hpt\" property");
3355     }
3356 }
3357 
3358 static bool spapr_get_vof(Object *obj, Error **errp)
3359 {
3360     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3361 
3362     return spapr->vof != NULL;
3363 }
3364 
3365 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3366 {
3367     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3368 
3369     if (spapr->vof) {
3370         vof_cleanup(spapr->vof);
3371         g_free(spapr->vof);
3372         spapr->vof = NULL;
3373     }
3374     if (!value) {
3375         return;
3376     }
3377     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3378 }
3379 
3380 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3381 {
3382     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3383 
3384     if (spapr->irq == &spapr_irq_xics_legacy) {
3385         return g_strdup("legacy");
3386     } else if (spapr->irq == &spapr_irq_xics) {
3387         return g_strdup("xics");
3388     } else if (spapr->irq == &spapr_irq_xive) {
3389         return g_strdup("xive");
3390     } else if (spapr->irq == &spapr_irq_dual) {
3391         return g_strdup("dual");
3392     }
3393     g_assert_not_reached();
3394 }
3395 
3396 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3397 {
3398     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3399 
3400     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3401         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3402         return;
3403     }
3404 
3405     /* The legacy IRQ backend can not be set */
3406     if (strcmp(value, "xics") == 0) {
3407         spapr->irq = &spapr_irq_xics;
3408     } else if (strcmp(value, "xive") == 0) {
3409         spapr->irq = &spapr_irq_xive;
3410     } else if (strcmp(value, "dual") == 0) {
3411         spapr->irq = &spapr_irq_dual;
3412     } else {
3413         error_setg(errp, "Bad value for \"ic-mode\" property");
3414     }
3415 }
3416 
3417 static char *spapr_get_host_model(Object *obj, Error **errp)
3418 {
3419     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3420 
3421     return g_strdup(spapr->host_model);
3422 }
3423 
3424 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3425 {
3426     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3427 
3428     g_free(spapr->host_model);
3429     spapr->host_model = g_strdup(value);
3430 }
3431 
3432 static char *spapr_get_host_serial(Object *obj, Error **errp)
3433 {
3434     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3435 
3436     return g_strdup(spapr->host_serial);
3437 }
3438 
3439 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3440 {
3441     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3442 
3443     g_free(spapr->host_serial);
3444     spapr->host_serial = g_strdup(value);
3445 }
3446 
3447 static void spapr_instance_init(Object *obj)
3448 {
3449     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3450     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3451     MachineState *ms = MACHINE(spapr);
3452     MachineClass *mc = MACHINE_GET_CLASS(ms);
3453 
3454     /*
3455      * NVDIMM support went live in 5.1 without considering that, in
3456      * other archs, the user needs to enable NVDIMM support with the
3457      * 'nvdimm' machine option and the default behavior is NVDIMM
3458      * support disabled. It is too late to roll back to the standard
3459      * behavior without breaking 5.1 guests.
3460      */
3461     if (mc->nvdimm_supported) {
3462         ms->nvdimms_state->is_enabled = true;
3463     }
3464 
3465     spapr->htab_fd = -1;
3466     spapr->use_hotplug_event_source = true;
3467     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3468     object_property_add_str(obj, "kvm-type",
3469                             spapr_get_kvm_type, spapr_set_kvm_type);
3470     object_property_set_description(obj, "kvm-type",
3471                                     "Specifies the KVM virtualization mode (auto,"
3472                                     " hv, pr). Defaults to 'auto'. This mode will use"
3473                                     " any available KVM module loaded in the host,"
3474                                     " where kvm_hv takes precedence if both kvm_hv and"
3475                                     " kvm_pr are loaded.");
3476     object_property_add_bool(obj, "modern-hotplug-events",
3477                             spapr_get_modern_hotplug_events,
3478                             spapr_set_modern_hotplug_events);
3479     object_property_set_description(obj, "modern-hotplug-events",
3480                                     "Use dedicated hotplug event mechanism in"
3481                                     " place of standard EPOW events when possible"
3482                                     " (required for memory hot-unplug support)");
3483     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3484                             "Maximum permitted CPU compatibility mode");
3485 
3486     object_property_add_str(obj, "resize-hpt",
3487                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3488     object_property_set_description(obj, "resize-hpt",
3489                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3490     object_property_add_uint32_ptr(obj, "vsmt",
3491                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3492     object_property_set_description(obj, "vsmt",
3493                                     "Virtual SMT: KVM behaves as if this were"
3494                                     " the host's SMT mode");
3495 
3496     object_property_add_bool(obj, "vfio-no-msix-emulation",
3497                              spapr_get_msix_emulation, NULL);
3498 
3499     object_property_add_uint64_ptr(obj, "kernel-addr",
3500                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3501     object_property_set_description(obj, "kernel-addr",
3502                                     stringify(KERNEL_LOAD_ADDR)
3503                                     " for -kernel is the default");
3504     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3505 
3506     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3507     object_property_set_description(obj, "x-vof",
3508                                     "Enable Virtual Open Firmware (experimental)");
3509 
3510     /* The machine class defines the default interrupt controller mode */
3511     spapr->irq = smc->irq;
3512     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3513                             spapr_set_ic_mode);
3514     object_property_set_description(obj, "ic-mode",
3515                  "Specifies the interrupt controller mode (xics, xive, dual)");
3516 
3517     object_property_add_str(obj, "host-model",
3518         spapr_get_host_model, spapr_set_host_model);
3519     object_property_set_description(obj, "host-model",
3520         "Host model to advertise in guest device tree");
3521     object_property_add_str(obj, "host-serial",
3522         spapr_get_host_serial, spapr_set_host_serial);
3523     object_property_set_description(obj, "host-serial",
3524         "Host serial number to advertise in guest device tree");
3525 }
3526 
3527 static void spapr_machine_finalizefn(Object *obj)
3528 {
3529     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3530 
3531     g_free(spapr->kvm_type);
3532 }
3533 
3534 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3535 {
3536     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3537     CPUPPCState *env = cpu_env(cs);
3538 
3539     cpu_synchronize_state(cs);
3540     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3541     if (spapr->fwnmi_system_reset_addr != -1) {
3542         uint64_t rtas_addr, addr;
3543 
3544         /* get rtas addr from fdt */
3545         rtas_addr = spapr_get_rtas_addr();
3546         if (!rtas_addr) {
3547             qemu_system_guest_panicked(NULL);
3548             return;
3549         }
3550 
3551         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3552         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3553         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3554         env->gpr[3] = addr;
3555     }
3556     ppc_cpu_do_system_reset(cs);
3557     if (spapr->fwnmi_system_reset_addr != -1) {
3558         env->nip = spapr->fwnmi_system_reset_addr;
3559     }
3560 }
3561 
3562 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3563 {
3564     CPUState *cs;
3565 
3566     CPU_FOREACH(cs) {
3567         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3568     }
3569 }
3570 
3571 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3572                           void *fdt, int *fdt_start_offset, Error **errp)
3573 {
3574     uint64_t addr;
3575     uint32_t node;
3576 
3577     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3578     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3579                                     &error_abort);
3580     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3581                                              SPAPR_MEMORY_BLOCK_SIZE);
3582     return 0;
3583 }
3584 
3585 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3586                            bool dedicated_hp_event_source)
3587 {
3588     SpaprDrc *drc;
3589     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3590     int i;
3591     uint64_t addr = addr_start;
3592     bool hotplugged = spapr_drc_hotplugged(dev);
3593 
3594     for (i = 0; i < nr_lmbs; i++) {
3595         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3596                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3597         g_assert(drc);
3598 
3599         /*
3600          * memory_device_get_free_addr() provided a range of free addresses
3601          * that doesn't overlap with any existing mapping at pre-plug. The
3602          * corresponding LMB DRCs are thus assumed to be all attachable.
3603          */
3604         spapr_drc_attach(drc, dev);
3605         if (!hotplugged) {
3606             spapr_drc_reset(drc);
3607         }
3608         addr += SPAPR_MEMORY_BLOCK_SIZE;
3609     }
3610     /* send hotplug notification to the
3611      * guest only in case of hotplugged memory
3612      */
3613     if (hotplugged) {
3614         if (dedicated_hp_event_source) {
3615             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3616                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3617             g_assert(drc);
3618             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3619                                                    nr_lmbs,
3620                                                    spapr_drc_index(drc));
3621         } else {
3622             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3623                                            nr_lmbs);
3624         }
3625     }
3626 }
3627 
3628 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3629 {
3630     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3631     PCDIMMDevice *dimm = PC_DIMM(dev);
3632     uint64_t size, addr;
3633     int64_t slot;
3634     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3635 
3636     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3637 
3638     pc_dimm_plug(dimm, MACHINE(ms));
3639 
3640     if (!is_nvdimm) {
3641         addr = object_property_get_uint(OBJECT(dimm),
3642                                         PC_DIMM_ADDR_PROP, &error_abort);
3643         spapr_add_lmbs(dev, addr, size,
3644                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3645     } else {
3646         slot = object_property_get_int(OBJECT(dimm),
3647                                        PC_DIMM_SLOT_PROP, &error_abort);
3648         /* We should have valid slot number at this point */
3649         g_assert(slot >= 0);
3650         spapr_add_nvdimm(dev, slot);
3651     }
3652 }
3653 
3654 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3655                                   Error **errp)
3656 {
3657     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3658     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3659     PCDIMMDevice *dimm = PC_DIMM(dev);
3660     Error *local_err = NULL;
3661     uint64_t size;
3662     Object *memdev;
3663     hwaddr pagesize;
3664 
3665     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3666     if (local_err) {
3667         error_propagate(errp, local_err);
3668         return;
3669     }
3670 
3671     if (is_nvdimm) {
3672         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3673             return;
3674         }
3675     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3676         error_setg(errp, "Hotplugged memory size must be a multiple of "
3677                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3678         return;
3679     }
3680 
3681     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3682                                       &error_abort);
3683     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3684     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3685         return;
3686     }
3687 
3688     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), errp);
3689 }
3690 
3691 struct SpaprDimmState {
3692     PCDIMMDevice *dimm;
3693     uint32_t nr_lmbs;
3694     QTAILQ_ENTRY(SpaprDimmState) next;
3695 };
3696 
3697 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3698                                                        PCDIMMDevice *dimm)
3699 {
3700     SpaprDimmState *dimm_state = NULL;
3701 
3702     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3703         if (dimm_state->dimm == dimm) {
3704             break;
3705         }
3706     }
3707     return dimm_state;
3708 }
3709 
3710 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3711                                                       uint32_t nr_lmbs,
3712                                                       PCDIMMDevice *dimm)
3713 {
3714     SpaprDimmState *ds = NULL;
3715 
3716     /*
3717      * If this request is for a DIMM whose removal had failed earlier
3718      * (due to guest's refusal to remove the LMBs), we would have this
3719      * dimm already in the pending_dimm_unplugs list. In that
3720      * case don't add again.
3721      */
3722     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3723     if (!ds) {
3724         ds = g_new0(SpaprDimmState, 1);
3725         ds->nr_lmbs = nr_lmbs;
3726         ds->dimm = dimm;
3727         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3728     }
3729     return ds;
3730 }
3731 
3732 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3733                                               SpaprDimmState *dimm_state)
3734 {
3735     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3736     g_free(dimm_state);
3737 }
3738 
3739 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3740                                                         PCDIMMDevice *dimm)
3741 {
3742     SpaprDrc *drc;
3743     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3744                                                   &error_abort);
3745     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3746     uint32_t avail_lmbs = 0;
3747     uint64_t addr_start, addr;
3748     int i;
3749 
3750     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3751                                           &error_abort);
3752 
3753     addr = addr_start;
3754     for (i = 0; i < nr_lmbs; i++) {
3755         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3756                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3757         g_assert(drc);
3758         if (drc->dev) {
3759             avail_lmbs++;
3760         }
3761         addr += SPAPR_MEMORY_BLOCK_SIZE;
3762     }
3763 
3764     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3765 }
3766 
3767 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3768 {
3769     SpaprDimmState *ds;
3770     PCDIMMDevice *dimm;
3771     SpaprDrc *drc;
3772     uint32_t nr_lmbs;
3773     uint64_t size, addr_start, addr;
3774     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.
3811      */
3812     qapi_event_send_device_unplug_guest_error(dev->id,
3813                                               dev->canonical_path);
3814 }
3815 
3816 /* Callback to be called during DRC release. */
3817 void spapr_lmb_release(DeviceState *dev)
3818 {
3819     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3820     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3821     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3822 
3823     /* This information will get lost if a migration occurs
3824      * during the unplug process. In this case recover it. */
3825     if (ds == NULL) {
3826         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3827         g_assert(ds);
3828         /* The DRC being examined by the caller at least must be counted */
3829         g_assert(ds->nr_lmbs);
3830     }
3831 
3832     if (--ds->nr_lmbs) {
3833         return;
3834     }
3835 
3836     /*
3837      * Now that all the LMBs have been removed by the guest, call the
3838      * unplug handler chain. This can never fail.
3839      */
3840     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3841     object_unparent(OBJECT(dev));
3842 }
3843 
3844 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3845 {
3846     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3847     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3848 
3849     /* We really shouldn't get this far without anything to unplug */
3850     g_assert(ds);
3851 
3852     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3853     qdev_unrealize(dev);
3854     spapr_pending_dimm_unplugs_remove(spapr, ds);
3855 }
3856 
3857 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3858                                         DeviceState *dev, Error **errp)
3859 {
3860     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3861     PCDIMMDevice *dimm = PC_DIMM(dev);
3862     uint32_t nr_lmbs;
3863     uint64_t size, addr_start, addr;
3864     int i;
3865     SpaprDrc *drc;
3866 
3867     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3868         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3869         return;
3870     }
3871 
3872     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3873     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3874 
3875     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3876                                           &error_abort);
3877 
3878     /*
3879      * An existing pending dimm state for this DIMM means that there is an
3880      * unplug operation in progress, waiting for the spapr_lmb_release
3881      * callback to complete the job (BQL can't cover that far). In this case,
3882      * bail out to avoid detaching DRCs that were already released.
3883      */
3884     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3885         error_setg(errp, "Memory unplug already in progress for device %s",
3886                    dev->id);
3887         return;
3888     }
3889 
3890     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3891 
3892     addr = addr_start;
3893     for (i = 0; i < nr_lmbs; i++) {
3894         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3895                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3896         g_assert(drc);
3897 
3898         spapr_drc_unplug_request(drc);
3899         addr += SPAPR_MEMORY_BLOCK_SIZE;
3900     }
3901 
3902     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3903                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3904     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3905                                               nr_lmbs, spapr_drc_index(drc));
3906 }
3907 
3908 /* Callback to be called during DRC release. */
3909 void spapr_core_release(DeviceState *dev)
3910 {
3911     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3912 
3913     /* Call the unplug handler chain. This can never fail. */
3914     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3915     object_unparent(OBJECT(dev));
3916 }
3917 
3918 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3919 {
3920     MachineState *ms = MACHINE(hotplug_dev);
3921     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3922     CPUCore *cc = CPU_CORE(dev);
3923     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3924 
3925     if (smc->pre_2_10_has_unused_icps) {
3926         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3927         int i;
3928 
3929         for (i = 0; i < cc->nr_threads; i++) {
3930             CPUState *cs = CPU(sc->threads[i]);
3931 
3932             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3933         }
3934     }
3935 
3936     assert(core_slot);
3937     core_slot->cpu = NULL;
3938     qdev_unrealize(dev);
3939 }
3940 
3941 static
3942 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3943                                Error **errp)
3944 {
3945     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3946     int index;
3947     SpaprDrc *drc;
3948     CPUCore *cc = CPU_CORE(dev);
3949 
3950     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3951         error_setg(errp, "Unable to find CPU core with core-id: %d",
3952                    cc->core_id);
3953         return;
3954     }
3955     if (index == 0) {
3956         error_setg(errp, "Boot CPU core may not be unplugged");
3957         return;
3958     }
3959 
3960     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3961                           spapr_vcpu_id(spapr, cc->core_id));
3962     g_assert(drc);
3963 
3964     if (!spapr_drc_unplug_requested(drc)) {
3965         spapr_drc_unplug_request(drc);
3966     }
3967 
3968     /*
3969      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3970      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3971      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3972      * attempt (e.g. the kernel will refuse to remove the last online
3973      * CPU), we will never attempt it again because unplug_requested
3974      * will still be 'true' in that case.
3975      */
3976     spapr_hotplug_req_remove_by_index(drc);
3977 }
3978 
3979 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3980                            void *fdt, int *fdt_start_offset, Error **errp)
3981 {
3982     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3983     CPUState *cs = CPU(core->threads[0]);
3984     PowerPCCPU *cpu = POWERPC_CPU(cs);
3985     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3986     int id = spapr_get_vcpu_id(cpu);
3987     g_autofree char *nodename = NULL;
3988     int offset;
3989 
3990     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3991     offset = fdt_add_subnode(fdt, 0, nodename);
3992 
3993     spapr_dt_cpu(cs, fdt, offset, spapr);
3994 
3995     /*
3996      * spapr_dt_cpu() does not fill the 'name' property in the
3997      * CPU node. The function is called during boot process, before
3998      * and after CAS, and overwriting the 'name' property written
3999      * by SLOF is not allowed.
4000      *
4001      * Write it manually after spapr_dt_cpu(). This makes the hotplug
4002      * CPUs more compatible with the coldplugged ones, which have
4003      * the 'name' property. Linux Kernel also relies on this
4004      * property to identify CPU nodes.
4005      */
4006     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
4007 
4008     *fdt_start_offset = offset;
4009     return 0;
4010 }
4011 
4012 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4013 {
4014     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4015     MachineClass *mc = MACHINE_GET_CLASS(spapr);
4016     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4017     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
4018     CPUCore *cc = CPU_CORE(dev);
4019     SpaprDrc *drc;
4020     CPUArchId *core_slot;
4021     int index;
4022     bool hotplugged = spapr_drc_hotplugged(dev);
4023     int i;
4024 
4025     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4026     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4027 
4028     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4029                           spapr_vcpu_id(spapr, cc->core_id));
4030 
4031     g_assert(drc || !mc->has_hotpluggable_cpus);
4032 
4033     if (drc) {
4034         /*
4035          * spapr_core_pre_plug() already buys us this is a brand new
4036          * core being plugged into a free slot. Nothing should already
4037          * be attached to the corresponding DRC.
4038          */
4039         spapr_drc_attach(drc, dev);
4040 
4041         if (hotplugged) {
4042             /*
4043              * Send hotplug notification interrupt to the guest only
4044              * in case of hotplugged CPUs.
4045              */
4046             spapr_hotplug_req_add_by_index(drc);
4047         } else {
4048             spapr_drc_reset(drc);
4049         }
4050     }
4051 
4052     core_slot->cpu = CPU(dev);
4053 
4054     /*
4055      * Set compatibility mode to match the boot CPU, which was either set
4056      * by the machine reset code or by CAS. This really shouldn't fail at
4057      * this point.
4058      */
4059     if (hotplugged) {
4060         for (i = 0; i < cc->nr_threads; i++) {
4061             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4062                            &error_abort);
4063         }
4064     }
4065 
4066     if (smc->pre_2_10_has_unused_icps) {
4067         for (i = 0; i < cc->nr_threads; i++) {
4068             CPUState *cs = CPU(core->threads[i]);
4069             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4070         }
4071     }
4072 }
4073 
4074 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4075                                 Error **errp)
4076 {
4077     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4078     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4079     CPUCore *cc = CPU_CORE(dev);
4080     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4081     const char *type = object_get_typename(OBJECT(dev));
4082     CPUArchId *core_slot;
4083     int index;
4084     unsigned int smp_threads = machine->smp.threads;
4085 
4086     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4087         error_setg(errp, "CPU hotplug not supported for this machine");
4088         return;
4089     }
4090 
4091     if (strcmp(base_core_type, type)) {
4092         error_setg(errp, "CPU core type should be %s", base_core_type);
4093         return;
4094     }
4095 
4096     if (cc->core_id % smp_threads) {
4097         error_setg(errp, "invalid core id %d", cc->core_id);
4098         return;
4099     }
4100 
4101     /*
4102      * In general we should have homogeneous threads-per-core, but old
4103      * (pre hotplug support) machine types allow the last core to have
4104      * reduced threads as a compatibility hack for when we allowed
4105      * total vcpus not a multiple of threads-per-core.
4106      */
4107     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4108         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4109                    smp_threads);
4110         return;
4111     }
4112 
4113     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4114     if (!core_slot) {
4115         error_setg(errp, "core id %d out of range", cc->core_id);
4116         return;
4117     }
4118 
4119     if (core_slot->cpu) {
4120         error_setg(errp, "core %d already populated", cc->core_id);
4121         return;
4122     }
4123 
4124     numa_cpu_pre_plug(core_slot, dev, errp);
4125 }
4126 
4127 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4128                           void *fdt, int *fdt_start_offset, Error **errp)
4129 {
4130     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4131     int intc_phandle;
4132 
4133     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4134     if (intc_phandle <= 0) {
4135         return -1;
4136     }
4137 
4138     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4139         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4140         return -1;
4141     }
4142 
4143     /* generally SLOF creates these, for hotplug it's up to QEMU */
4144     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4145 
4146     return 0;
4147 }
4148 
4149 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4150                                Error **errp)
4151 {
4152     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4153     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4154     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4155     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4156     SpaprDrc *drc;
4157 
4158     if (dev->hotplugged && !smc->dr_phb_enabled) {
4159         error_setg(errp, "PHB hotplug not supported for this machine");
4160         return false;
4161     }
4162 
4163     if (sphb->index == (uint32_t)-1) {
4164         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4165         return false;
4166     }
4167 
4168     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4169     if (drc && drc->dev) {
4170         error_setg(errp, "PHB %d already attached", sphb->index);
4171         return false;
4172     }
4173 
4174     /*
4175      * This will check that sphb->index doesn't exceed the maximum number of
4176      * PHBs for the current machine type.
4177      */
4178     return
4179         smc->phb_placement(spapr, sphb->index,
4180                            &sphb->buid, &sphb->io_win_addr,
4181                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4182                            windows_supported, sphb->dma_liobn,
4183                            errp);
4184 }
4185 
4186 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4187 {
4188     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4189     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4190     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4191     SpaprDrc *drc;
4192     bool hotplugged = spapr_drc_hotplugged(dev);
4193 
4194     if (!smc->dr_phb_enabled) {
4195         return;
4196     }
4197 
4198     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4199     /* hotplug hooks should check it's enabled before getting this far */
4200     assert(drc);
4201 
4202     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4203     spapr_drc_attach(drc, dev);
4204 
4205     if (hotplugged) {
4206         spapr_hotplug_req_add_by_index(drc);
4207     } else {
4208         spapr_drc_reset(drc);
4209     }
4210 }
4211 
4212 void spapr_phb_release(DeviceState *dev)
4213 {
4214     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4215 
4216     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4217     object_unparent(OBJECT(dev));
4218 }
4219 
4220 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4221 {
4222     qdev_unrealize(dev);
4223 }
4224 
4225 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4226                                      DeviceState *dev, Error **errp)
4227 {
4228     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4229     SpaprDrc *drc;
4230 
4231     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4232     assert(drc);
4233 
4234     if (!spapr_drc_unplug_requested(drc)) {
4235         spapr_drc_unplug_request(drc);
4236         spapr_hotplug_req_remove_by_index(drc);
4237     } else {
4238         error_setg(errp,
4239                    "PCI Host Bridge unplug already in progress for device %s",
4240                    dev->id);
4241     }
4242 }
4243 
4244 static
4245 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4246                               Error **errp)
4247 {
4248     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4249 
4250     if (spapr->tpm_proxy != NULL) {
4251         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4252         return false;
4253     }
4254 
4255     return true;
4256 }
4257 
4258 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4259 {
4260     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4261     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4262 
4263     /* Already checked in spapr_tpm_proxy_pre_plug() */
4264     g_assert(spapr->tpm_proxy == NULL);
4265 
4266     spapr->tpm_proxy = tpm_proxy;
4267 }
4268 
4269 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4270 {
4271     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4272 
4273     qdev_unrealize(dev);
4274     object_unparent(OBJECT(dev));
4275     spapr->tpm_proxy = NULL;
4276 }
4277 
4278 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4279                                       DeviceState *dev, Error **errp)
4280 {
4281     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4282         spapr_memory_plug(hotplug_dev, dev);
4283     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4284         spapr_core_plug(hotplug_dev, dev);
4285     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4286         spapr_phb_plug(hotplug_dev, dev);
4287     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4288         spapr_tpm_proxy_plug(hotplug_dev, dev);
4289     }
4290 }
4291 
4292 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4293                                         DeviceState *dev, Error **errp)
4294 {
4295     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4296         spapr_memory_unplug(hotplug_dev, dev);
4297     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4298         spapr_core_unplug(hotplug_dev, dev);
4299     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4300         spapr_phb_unplug(hotplug_dev, dev);
4301     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4302         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4303     }
4304 }
4305 
4306 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4307 {
4308     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4309         /*
4310          * CAS will process all pending unplug requests.
4311          *
4312          * HACK: a guest could theoretically have cleared all bits in OV5,
4313          * but none of the guests we care for do.
4314          */
4315         spapr_ovec_empty(spapr->ov5_cas);
4316 }
4317 
4318 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4319                                                 DeviceState *dev, Error **errp)
4320 {
4321     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4322     MachineClass *mc = MACHINE_GET_CLASS(sms);
4323     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4324 
4325     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4326         if (spapr_memory_hot_unplug_supported(sms)) {
4327             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4328         } else {
4329             error_setg(errp, "Memory hot unplug not supported for this guest");
4330         }
4331     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4332         if (!mc->has_hotpluggable_cpus) {
4333             error_setg(errp, "CPU hot unplug not supported on this machine");
4334             return;
4335         }
4336         spapr_core_unplug_request(hotplug_dev, dev, errp);
4337     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4338         if (!smc->dr_phb_enabled) {
4339             error_setg(errp, "PHB hot unplug not supported on this machine");
4340             return;
4341         }
4342         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4343     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4344         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4345     }
4346 }
4347 
4348 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4349                                           DeviceState *dev, Error **errp)
4350 {
4351     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4352         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4353     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4354         spapr_core_pre_plug(hotplug_dev, dev, errp);
4355     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4356         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4357     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4358         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4359     }
4360 }
4361 
4362 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4363                                                  DeviceState *dev)
4364 {
4365     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4366         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4367         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4368         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4369         return HOTPLUG_HANDLER(machine);
4370     }
4371     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4372         PCIDevice *pcidev = PCI_DEVICE(dev);
4373         PCIBus *root = pci_device_root_bus(pcidev);
4374         SpaprPhbState *phb =
4375             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4376                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4377 
4378         if (phb) {
4379             return HOTPLUG_HANDLER(phb);
4380         }
4381     }
4382     return NULL;
4383 }
4384 
4385 static CpuInstanceProperties
4386 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4387 {
4388     CPUArchId *core_slot;
4389     MachineClass *mc = MACHINE_GET_CLASS(machine);
4390 
4391     /* make sure possible_cpu are initialized */
4392     mc->possible_cpu_arch_ids(machine);
4393     /* get CPU core slot containing thread that matches cpu_index */
4394     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4395     assert(core_slot);
4396     return core_slot->props;
4397 }
4398 
4399 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4400 {
4401     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4402 }
4403 
4404 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4405 {
4406     int i;
4407     unsigned int smp_threads = machine->smp.threads;
4408     unsigned int smp_cpus = machine->smp.cpus;
4409     const char *core_type;
4410     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4411     MachineClass *mc = MACHINE_GET_CLASS(machine);
4412 
4413     if (!mc->has_hotpluggable_cpus) {
4414         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4415     }
4416     if (machine->possible_cpus) {
4417         assert(machine->possible_cpus->len == spapr_max_cores);
4418         return machine->possible_cpus;
4419     }
4420 
4421     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4422     if (!core_type) {
4423         error_report("Unable to find sPAPR CPU Core definition");
4424         exit(1);
4425     }
4426 
4427     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4428                              sizeof(CPUArchId) * spapr_max_cores);
4429     machine->possible_cpus->len = spapr_max_cores;
4430     for (i = 0; i < machine->possible_cpus->len; i++) {
4431         int core_id = i * smp_threads;
4432 
4433         machine->possible_cpus->cpus[i].type = core_type;
4434         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4435         machine->possible_cpus->cpus[i].arch_id = core_id;
4436         machine->possible_cpus->cpus[i].props.has_core_id = true;
4437         machine->possible_cpus->cpus[i].props.core_id = core_id;
4438     }
4439     return machine->possible_cpus;
4440 }
4441 
4442 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4443                                 uint64_t *buid, hwaddr *pio,
4444                                 hwaddr *mmio32, hwaddr *mmio64,
4445                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4446 {
4447     /*
4448      * New-style PHB window placement.
4449      *
4450      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4451      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4452      * windows.
4453      *
4454      * Some guest kernels can't work with MMIO windows above 1<<46
4455      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4456      *
4457      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4458      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4459      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4460      * 1TiB 64-bit MMIO windows for each PHB.
4461      */
4462     const uint64_t base_buid = 0x800000020000000ULL;
4463     int i;
4464 
4465     /* Sanity check natural alignments */
4466     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4467     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4468     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4469     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4470     /* Sanity check bounds */
4471     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4472                       SPAPR_PCI_MEM32_WIN_SIZE);
4473     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4474                       SPAPR_PCI_MEM64_WIN_SIZE);
4475 
4476     if (index >= SPAPR_MAX_PHBS) {
4477         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4478                    SPAPR_MAX_PHBS - 1);
4479         return false;
4480     }
4481 
4482     *buid = base_buid + index;
4483     for (i = 0; i < n_dma; ++i) {
4484         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4485     }
4486 
4487     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4488     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4489     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4490     return true;
4491 }
4492 
4493 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4494 {
4495     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4496 
4497     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4498 }
4499 
4500 static void spapr_ics_resend(XICSFabric *dev)
4501 {
4502     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4503 
4504     ics_resend(spapr->ics);
4505 }
4506 
4507 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4508 {
4509     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4510 
4511     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4512 }
4513 
4514 static void spapr_pic_print_info(InterruptStatsProvider *obj, GString *buf)
4515 {
4516     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4517 
4518     spapr_irq_print_info(spapr, buf);
4519     g_string_append_printf(buf, "irqchip: %s\n",
4520                            kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4521 }
4522 
4523 /*
4524  * This is a XIVE only operation
4525  */
4526 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4527                            uint8_t nvt_blk, uint32_t nvt_idx,
4528                            bool cam_ignore, uint8_t priority,
4529                            uint32_t logic_serv, XiveTCTXMatch *match)
4530 {
4531     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4532     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4533     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4534     int count;
4535 
4536     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4537                            priority, logic_serv, match);
4538     if (count < 0) {
4539         return count;
4540     }
4541 
4542     /*
4543      * When we implement the save and restore of the thread interrupt
4544      * contexts in the enter/exit CPU handlers of the machine and the
4545      * escalations in QEMU, we should be able to handle non dispatched
4546      * vCPUs.
4547      *
4548      * Until this is done, the sPAPR machine should find at least one
4549      * matching context always.
4550      */
4551     if (count == 0) {
4552         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4553                       nvt_blk, nvt_idx);
4554     }
4555 
4556     return count;
4557 }
4558 
4559 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4560 {
4561     return cpu->vcpu_id;
4562 }
4563 
4564 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4565 {
4566     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4567     MachineState *ms = MACHINE(spapr);
4568     int vcpu_id;
4569 
4570     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4571 
4572     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4573         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4574         error_append_hint(errp, "Adjust the number of cpus to %d "
4575                           "or try to raise the number of threads per core\n",
4576                           vcpu_id * ms->smp.threads / spapr->vsmt);
4577         return false;
4578     }
4579 
4580     cpu->vcpu_id = vcpu_id;
4581     return true;
4582 }
4583 
4584 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4585 {
4586     CPUState *cs;
4587 
4588     CPU_FOREACH(cs) {
4589         PowerPCCPU *cpu = POWERPC_CPU(cs);
4590 
4591         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4592             return cpu;
4593         }
4594     }
4595 
4596     return NULL;
4597 }
4598 
4599 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4600 {
4601     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4602 
4603     return spapr_cpu->in_nested;
4604 }
4605 
4606 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4607 {
4608     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4609 
4610     /* These are only called by TCG, KVM maintains dispatch state */
4611 
4612     spapr_cpu->prod = false;
4613     if (spapr_cpu->vpa_addr) {
4614         CPUState *cs = CPU(cpu);
4615         uint32_t dispatch;
4616 
4617         dispatch = ldl_be_phys(cs->as,
4618                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4619         dispatch++;
4620         if ((dispatch & 1) != 0) {
4621             qemu_log_mask(LOG_GUEST_ERROR,
4622                           "VPA: incorrect dispatch counter value for "
4623                           "dispatched partition %u, correcting.\n", dispatch);
4624             dispatch++;
4625         }
4626         stl_be_phys(cs->as,
4627                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4628     }
4629 }
4630 
4631 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4632 {
4633     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4634 
4635     if (spapr_cpu->vpa_addr) {
4636         CPUState *cs = CPU(cpu);
4637         uint32_t dispatch;
4638 
4639         dispatch = ldl_be_phys(cs->as,
4640                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4641         dispatch++;
4642         if ((dispatch & 1) != 1) {
4643             qemu_log_mask(LOG_GUEST_ERROR,
4644                           "VPA: incorrect dispatch counter value for "
4645                           "preempted partition %u, correcting.\n", dispatch);
4646             dispatch++;
4647         }
4648         stl_be_phys(cs->as,
4649                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4650     }
4651 }
4652 
4653 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4654 {
4655     MachineClass *mc = MACHINE_CLASS(oc);
4656     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4657     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4658     NMIClass *nc = NMI_CLASS(oc);
4659     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4660     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4661     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4662     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4663     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4664     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4665 
4666     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4667     mc->ignore_boot_device_suffixes = true;
4668 
4669     /*
4670      * We set up the default / latest behaviour here.  The class_init
4671      * functions for the specific versioned machine types can override
4672      * these details for backwards compatibility
4673      */
4674     mc->init = spapr_machine_init;
4675     mc->reset = spapr_machine_reset;
4676     mc->block_default_type = IF_SCSI;
4677 
4678     /*
4679      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4680      * In TCG the limit is restricted by the range of CPU IPIs available.
4681      */
4682     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4683 
4684     mc->no_parallel = 1;
4685     mc->default_boot_order = "";
4686     mc->default_ram_size = 512 * MiB;
4687     mc->default_ram_id = "ppc_spapr.ram";
4688     mc->default_display = "std";
4689     mc->kvm_type = spapr_kvm_type;
4690     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4691     mc->pci_allow_0_address = true;
4692     assert(!mc->get_hotplug_handler);
4693     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4694     hc->pre_plug = spapr_machine_device_pre_plug;
4695     hc->plug = spapr_machine_device_plug;
4696     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4697     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4698     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4699     hc->unplug_request = spapr_machine_device_unplug_request;
4700     hc->unplug = spapr_machine_device_unplug;
4701 
4702     smc->update_dt_enabled = true;
4703     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4704     mc->has_hotpluggable_cpus = true;
4705     mc->nvdimm_supported = true;
4706     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4707     fwc->get_dev_path = spapr_get_fw_dev_path;
4708     nc->nmi_monitor_handler = spapr_nmi;
4709     smc->phb_placement = spapr_phb_placement;
4710     vhc->cpu_in_nested = spapr_cpu_in_nested;
4711     vhc->deliver_hv_excp = spapr_exit_nested;
4712     vhc->hypercall = emulate_spapr_hypercall;
4713     vhc->hpt_mask = spapr_hpt_mask;
4714     vhc->map_hptes = spapr_map_hptes;
4715     vhc->unmap_hptes = spapr_unmap_hptes;
4716     vhc->hpte_set_c = spapr_hpte_set_c;
4717     vhc->hpte_set_r = spapr_hpte_set_r;
4718     vhc->get_pate = spapr_get_pate;
4719     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4720     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4721     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4722     xic->ics_get = spapr_ics_get;
4723     xic->ics_resend = spapr_ics_resend;
4724     xic->icp_get = spapr_icp_get;
4725     ispc->print_info = spapr_pic_print_info;
4726     /* Force NUMA node memory size to be a multiple of
4727      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4728      * in which LMBs are represented and hot-added
4729      */
4730     mc->numa_mem_align_shift = 28;
4731     mc->auto_enable_numa = true;
4732 
4733     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4734     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4735     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4736     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4737     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4738     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4739     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4740     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4741     smc->default_caps.caps[SPAPR_CAP_NESTED_PAPR] = SPAPR_CAP_OFF;
4742     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4743     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4744     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4745     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4746 
4747     /*
4748      * This cap specifies whether the AIL 3 mode for
4749      * H_SET_RESOURCE is supported. The default is modified
4750      * by default_caps_with_cpu().
4751      */
4752     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4753     spapr_caps_add_properties(smc);
4754     smc->irq = &spapr_irq_dual;
4755     smc->dr_phb_enabled = true;
4756     smc->linux_pci_probe = true;
4757     smc->smp_threads_vsmt = true;
4758     smc->nr_xirqs = SPAPR_NR_XIRQS;
4759     xfc->match_nvt = spapr_match_nvt;
4760     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4761     vmc->quiesce = spapr_vof_quiesce;
4762     vmc->setprop = spapr_vof_setprop;
4763 }
4764 
4765 static const TypeInfo spapr_machine_info = {
4766     .name          = TYPE_SPAPR_MACHINE,
4767     .parent        = TYPE_MACHINE,
4768     .abstract      = true,
4769     .instance_size = sizeof(SpaprMachineState),
4770     .instance_init = spapr_instance_init,
4771     .instance_finalize = spapr_machine_finalizefn,
4772     .class_size    = sizeof(SpaprMachineClass),
4773     .class_init    = spapr_machine_class_init,
4774     .interfaces = (InterfaceInfo[]) {
4775         { TYPE_FW_PATH_PROVIDER },
4776         { TYPE_NMI },
4777         { TYPE_HOTPLUG_HANDLER },
4778         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4779         { TYPE_XICS_FABRIC },
4780         { TYPE_INTERRUPT_STATS_PROVIDER },
4781         { TYPE_XIVE_FABRIC },
4782         { TYPE_VOF_MACHINE_IF },
4783         { }
4784     },
4785 };
4786 
4787 static void spapr_machine_latest_class_options(MachineClass *mc)
4788 {
4789     mc->alias = "pseries";
4790     mc->is_default = true;
4791 }
4792 
4793 #define DEFINE_SPAPR_MACHINE_IMPL(latest, ...)                       \
4794     static void MACHINE_VER_SYM(class_init, spapr, __VA_ARGS__)(     \
4795         ObjectClass *oc,                                             \
4796         void *data)                                                  \
4797     {                                                                \
4798         MachineClass *mc = MACHINE_CLASS(oc);                        \
4799         MACHINE_VER_SYM(class_options, spapr, __VA_ARGS__)(mc);      \
4800         MACHINE_VER_DEPRECATION(__VA_ARGS__);                        \
4801         if (latest) {                                                \
4802             spapr_machine_latest_class_options(mc);                  \
4803         }                                                            \
4804     }                                                                \
4805     static const TypeInfo MACHINE_VER_SYM(info, spapr, __VA_ARGS__) = \
4806     {                                                                \
4807         .name = MACHINE_VER_TYPE_NAME("pseries", __VA_ARGS__),       \
4808         .parent = TYPE_SPAPR_MACHINE,                                \
4809         .class_init = MACHINE_VER_SYM(class_init, spapr, __VA_ARGS__), \
4810     };                                                               \
4811     static void MACHINE_VER_SYM(register, spapr, __VA_ARGS__)(void)  \
4812     {                                                                \
4813         MACHINE_VER_DELETION(__VA_ARGS__);                           \
4814         type_register(&MACHINE_VER_SYM(info, spapr, __VA_ARGS__));   \
4815     }                                                                \
4816     type_init(MACHINE_VER_SYM(register, spapr, __VA_ARGS__))
4817 
4818 #define DEFINE_SPAPR_MACHINE_AS_LATEST(major, minor) \
4819     DEFINE_SPAPR_MACHINE_IMPL(true, major, minor)
4820 #define DEFINE_SPAPR_MACHINE(major, minor) \
4821     DEFINE_SPAPR_MACHINE_IMPL(false, major, minor)
4822 #define DEFINE_SPAPR_MACHINE_TAGGED(major, minor, tag) \
4823     DEFINE_SPAPR_MACHINE_IMPL(false, major, minor, _, tag)
4824 
4825 /*
4826  * pseries-9.2
4827  */
4828 static void spapr_machine_9_2_class_options(MachineClass *mc)
4829 {
4830     /* Defaults for the latest behaviour inherited from the base class */
4831 }
4832 
4833 DEFINE_SPAPR_MACHINE_AS_LATEST(9, 2);
4834 
4835 /*
4836  * pseries-9.1
4837  */
4838 static void spapr_machine_9_1_class_options(MachineClass *mc)
4839 {
4840     spapr_machine_9_2_class_options(mc);
4841     compat_props_add(mc->compat_props, hw_compat_9_1, hw_compat_9_1_len);
4842 }
4843 
4844 DEFINE_SPAPR_MACHINE(9, 1);
4845 
4846 /*
4847  * pseries-9.0
4848  */
4849 static void spapr_machine_9_0_class_options(MachineClass *mc)
4850 {
4851     spapr_machine_9_1_class_options(mc);
4852     compat_props_add(mc->compat_props, hw_compat_9_0, hw_compat_9_0_len);
4853 }
4854 
4855 DEFINE_SPAPR_MACHINE(9, 0);
4856 
4857 /*
4858  * pseries-8.2
4859  */
4860 static void spapr_machine_8_2_class_options(MachineClass *mc)
4861 {
4862     spapr_machine_9_0_class_options(mc);
4863     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4864 }
4865 
4866 DEFINE_SPAPR_MACHINE(8, 2);
4867 
4868 /*
4869  * pseries-8.1
4870  */
4871 static void spapr_machine_8_1_class_options(MachineClass *mc)
4872 {
4873     spapr_machine_8_2_class_options(mc);
4874     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4875 }
4876 
4877 DEFINE_SPAPR_MACHINE(8, 1);
4878 
4879 /*
4880  * pseries-8.0
4881  */
4882 static void spapr_machine_8_0_class_options(MachineClass *mc)
4883 {
4884     spapr_machine_8_1_class_options(mc);
4885     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4886 }
4887 
4888 DEFINE_SPAPR_MACHINE(8, 0);
4889 
4890 /*
4891  * pseries-7.2
4892  */
4893 static void spapr_machine_7_2_class_options(MachineClass *mc)
4894 {
4895     spapr_machine_8_0_class_options(mc);
4896     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4897 }
4898 
4899 DEFINE_SPAPR_MACHINE(7, 2);
4900 
4901 /*
4902  * pseries-7.1
4903  */
4904 static void spapr_machine_7_1_class_options(MachineClass *mc)
4905 {
4906     spapr_machine_7_2_class_options(mc);
4907     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4908 }
4909 
4910 DEFINE_SPAPR_MACHINE(7, 1);
4911 
4912 /*
4913  * pseries-7.0
4914  */
4915 static void spapr_machine_7_0_class_options(MachineClass *mc)
4916 {
4917     spapr_machine_7_1_class_options(mc);
4918     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4919 }
4920 
4921 DEFINE_SPAPR_MACHINE(7, 0);
4922 
4923 /*
4924  * pseries-6.2
4925  */
4926 static void spapr_machine_6_2_class_options(MachineClass *mc)
4927 {
4928     spapr_machine_7_0_class_options(mc);
4929     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4930 }
4931 
4932 DEFINE_SPAPR_MACHINE(6, 2);
4933 
4934 /*
4935  * pseries-6.1
4936  */
4937 static void spapr_machine_6_1_class_options(MachineClass *mc)
4938 {
4939     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4940 
4941     spapr_machine_6_2_class_options(mc);
4942     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4943     smc->pre_6_2_numa_affinity = true;
4944     mc->smp_props.prefer_sockets = true;
4945 }
4946 
4947 DEFINE_SPAPR_MACHINE(6, 1);
4948 
4949 /*
4950  * pseries-6.0
4951  */
4952 static void spapr_machine_6_0_class_options(MachineClass *mc)
4953 {
4954     spapr_machine_6_1_class_options(mc);
4955     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4956 }
4957 
4958 DEFINE_SPAPR_MACHINE(6, 0);
4959 
4960 /*
4961  * pseries-5.2
4962  */
4963 static void spapr_machine_5_2_class_options(MachineClass *mc)
4964 {
4965     spapr_machine_6_0_class_options(mc);
4966     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4967 }
4968 
4969 DEFINE_SPAPR_MACHINE(5, 2);
4970 
4971 /*
4972  * pseries-5.1
4973  */
4974 static void spapr_machine_5_1_class_options(MachineClass *mc)
4975 {
4976     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4977 
4978     spapr_machine_5_2_class_options(mc);
4979     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4980     smc->pre_5_2_numa_associativity = true;
4981 }
4982 
4983 DEFINE_SPAPR_MACHINE(5, 1);
4984 
4985 /*
4986  * pseries-5.0
4987  */
4988 static void spapr_machine_5_0_class_options(MachineClass *mc)
4989 {
4990     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4991     static GlobalProperty compat[] = {
4992         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4993     };
4994 
4995     spapr_machine_5_1_class_options(mc);
4996     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4997     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4998     mc->numa_mem_supported = true;
4999     smc->pre_5_1_assoc_refpoints = true;
5000 }
5001 
5002 DEFINE_SPAPR_MACHINE(5, 0);
5003 
5004 /*
5005  * pseries-4.2
5006  */
5007 static void spapr_machine_4_2_class_options(MachineClass *mc)
5008 {
5009     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5010 
5011     spapr_machine_5_0_class_options(mc);
5012     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
5013     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
5014     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
5015     smc->rma_limit = 16 * GiB;
5016     mc->nvdimm_supported = false;
5017 }
5018 
5019 DEFINE_SPAPR_MACHINE(4, 2);
5020 
5021 /*
5022  * pseries-4.1
5023  */
5024 static void spapr_machine_4_1_class_options(MachineClass *mc)
5025 {
5026     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5027     static GlobalProperty compat[] = {
5028         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
5029         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
5030     };
5031 
5032     spapr_machine_4_2_class_options(mc);
5033     smc->linux_pci_probe = false;
5034     smc->smp_threads_vsmt = false;
5035     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
5036     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5037 }
5038 
5039 DEFINE_SPAPR_MACHINE(4, 1);
5040 
5041 /*
5042  * pseries-4.0
5043  */
5044 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
5045                               uint64_t *buid, hwaddr *pio,
5046                               hwaddr *mmio32, hwaddr *mmio64,
5047                               unsigned n_dma, uint32_t *liobns, Error **errp)
5048 {
5049     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
5050                              liobns, errp)) {
5051         return false;
5052     }
5053     return true;
5054 }
5055 static void spapr_machine_4_0_class_options(MachineClass *mc)
5056 {
5057     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5058 
5059     spapr_machine_4_1_class_options(mc);
5060     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5061     smc->phb_placement = phb_placement_4_0;
5062     smc->irq = &spapr_irq_xics;
5063     smc->pre_4_1_migration = true;
5064 }
5065 
5066 DEFINE_SPAPR_MACHINE(4, 0);
5067 
5068 /*
5069  * pseries-3.1
5070  */
5071 static void spapr_machine_3_1_class_options(MachineClass *mc)
5072 {
5073     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5074 
5075     spapr_machine_4_0_class_options(mc);
5076     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5077 
5078     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5079     smc->update_dt_enabled = false;
5080     smc->dr_phb_enabled = false;
5081     smc->broken_host_serial_model = true;
5082     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5083     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5084     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5085     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5086 }
5087 
5088 DEFINE_SPAPR_MACHINE(3, 1);
5089 
5090 /*
5091  * pseries-3.0
5092  */
5093 
5094 static void spapr_machine_3_0_class_options(MachineClass *mc)
5095 {
5096     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5097 
5098     spapr_machine_3_1_class_options(mc);
5099     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5100 
5101     smc->legacy_irq_allocation = true;
5102     smc->nr_xirqs = 0x400;
5103     smc->irq = &spapr_irq_xics_legacy;
5104 }
5105 
5106 DEFINE_SPAPR_MACHINE(3, 0);
5107 
5108 /*
5109  * pseries-2.12
5110  */
5111 static void spapr_machine_2_12_class_options(MachineClass *mc)
5112 {
5113     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5114     static GlobalProperty compat[] = {
5115         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5116         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5117     };
5118 
5119     spapr_machine_3_0_class_options(mc);
5120     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5121     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5122 
5123     /* We depend on kvm_enabled() to choose a default value for the
5124      * hpt-max-page-size capability. Of course we can't do it here
5125      * because this is too early and the HW accelerator isn't initialized
5126      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5127      */
5128     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5129 }
5130 
5131 DEFINE_SPAPR_MACHINE(2, 12);
5132 
5133 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5134 {
5135     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5136 
5137     spapr_machine_2_12_class_options(mc);
5138     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5139     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5140     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5141 }
5142 
5143 DEFINE_SPAPR_MACHINE_TAGGED(2, 12, sxxm);
5144 
5145 /*
5146  * pseries-2.11
5147  */
5148 
5149 static void spapr_machine_2_11_class_options(MachineClass *mc)
5150 {
5151     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5152 
5153     spapr_machine_2_12_class_options(mc);
5154     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5155     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5156 }
5157 
5158 DEFINE_SPAPR_MACHINE(2, 11);
5159 
5160 /*
5161  * pseries-2.10
5162  */
5163 
5164 static void spapr_machine_2_10_class_options(MachineClass *mc)
5165 {
5166     spapr_machine_2_11_class_options(mc);
5167     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5168 }
5169 
5170 DEFINE_SPAPR_MACHINE(2, 10);
5171 
5172 /*
5173  * pseries-2.9
5174  */
5175 
5176 static void spapr_machine_2_9_class_options(MachineClass *mc)
5177 {
5178     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5179     static GlobalProperty compat[] = {
5180         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5181     };
5182 
5183     spapr_machine_2_10_class_options(mc);
5184     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5185     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5186     smc->pre_2_10_has_unused_icps = true;
5187     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5188 }
5189 
5190 DEFINE_SPAPR_MACHINE(2, 9);
5191 
5192 /*
5193  * pseries-2.8
5194  */
5195 
5196 static void spapr_machine_2_8_class_options(MachineClass *mc)
5197 {
5198     static GlobalProperty compat[] = {
5199         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5200     };
5201 
5202     spapr_machine_2_9_class_options(mc);
5203     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5204     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5205     mc->numa_mem_align_shift = 23;
5206 }
5207 
5208 DEFINE_SPAPR_MACHINE(2, 8);
5209 
5210 /*
5211  * pseries-2.7
5212  */
5213 
5214 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5215                               uint64_t *buid, hwaddr *pio,
5216                               hwaddr *mmio32, hwaddr *mmio64,
5217                               unsigned n_dma, uint32_t *liobns, Error **errp)
5218 {
5219     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5220     const uint64_t base_buid = 0x800000020000000ULL;
5221     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5222     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5223     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5224     const uint32_t max_index = 255;
5225     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5226 
5227     uint64_t ram_top = MACHINE(spapr)->ram_size;
5228     hwaddr phb0_base, phb_base;
5229     int i;
5230 
5231     /* Do we have device memory? */
5232     if (MACHINE(spapr)->device_memory) {
5233         /* Can't just use maxram_size, because there may be an
5234          * alignment gap between normal and device memory regions
5235          */
5236         ram_top = MACHINE(spapr)->device_memory->base +
5237             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5238     }
5239 
5240     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5241 
5242     if (index > max_index) {
5243         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5244                    max_index);
5245         return false;
5246     }
5247 
5248     *buid = base_buid + index;
5249     for (i = 0; i < n_dma; ++i) {
5250         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5251     }
5252 
5253     phb_base = phb0_base + index * phb_spacing;
5254     *pio = phb_base + pio_offset;
5255     *mmio32 = phb_base + mmio_offset;
5256     /*
5257      * We don't set the 64-bit MMIO window, relying on the PHB's
5258      * fallback behaviour of automatically splitting a large "32-bit"
5259      * window into contiguous 32-bit and 64-bit windows
5260      */
5261 
5262     return true;
5263 }
5264 
5265 static void spapr_machine_2_7_class_options(MachineClass *mc)
5266 {
5267     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5268     static GlobalProperty compat[] = {
5269         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5270         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5271         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5272         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5273     };
5274 
5275     spapr_machine_2_8_class_options(mc);
5276     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5277     mc->default_machine_opts = "modern-hotplug-events=off";
5278     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5279     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5280     smc->phb_placement = phb_placement_2_7;
5281 }
5282 
5283 DEFINE_SPAPR_MACHINE(2, 7);
5284 
5285 /*
5286  * pseries-2.6
5287  */
5288 
5289 static void spapr_machine_2_6_class_options(MachineClass *mc)
5290 {
5291     static GlobalProperty compat[] = {
5292         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5293     };
5294 
5295     spapr_machine_2_7_class_options(mc);
5296     mc->has_hotpluggable_cpus = false;
5297     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5298     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5299 }
5300 
5301 DEFINE_SPAPR_MACHINE(2, 6);
5302 
5303 /*
5304  * pseries-2.5
5305  */
5306 
5307 static void spapr_machine_2_5_class_options(MachineClass *mc)
5308 {
5309     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5310     static GlobalProperty compat[] = {
5311         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5312     };
5313 
5314     spapr_machine_2_6_class_options(mc);
5315     smc->use_ohci_by_default = true;
5316     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5317     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5318 }
5319 
5320 DEFINE_SPAPR_MACHINE(2, 5);
5321 
5322 static void spapr_machine_register_types(void)
5323 {
5324     type_register_static(&spapr_machine_info);
5325 }
5326 
5327 type_init(spapr_machine_register_types)
5328