xref: /qemu/hw/ppc/spapr.c (revision 6ff5da16000f908140723e164d33a0b51a6c4162)
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 "system/system.h"
36 #include "system/hostmem.h"
37 #include "system/numa.h"
38 #include "system/tcg.h"
39 #include "system/qtest.h"
40 #include "system/reset.h"
41 #include "system/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 "system/device_tree.h"
47 #include "system/cpus.h"
48 #include "system/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 "system/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 int spapr_max_server_number(SpaprMachineState *spapr)
136 {
137     MachineState *ms = MACHINE(spapr);
138 
139     assert(spapr->vsmt);
140     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
141 }
142 
143 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
144                                   int smt_threads)
145 {
146     int i, ret = 0;
147     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
148     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
149     int index = spapr_get_vcpu_id(cpu);
150 
151     if (cpu->compat_pvr) {
152         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
153         if (ret < 0) {
154             return ret;
155         }
156     }
157 
158     /* Build interrupt servers and gservers properties */
159     for (i = 0; i < smt_threads; i++) {
160         servers_prop[i] = cpu_to_be32(index + i);
161         /* Hack, direct the group queues back to cpu 0 */
162         gservers_prop[i*2] = cpu_to_be32(index + i);
163         gservers_prop[i*2 + 1] = 0;
164     }
165     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
166                       servers_prop, sizeof(*servers_prop) * smt_threads);
167     if (ret < 0) {
168         return ret;
169     }
170     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
171                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
172 
173     return ret;
174 }
175 
176 static void spapr_dt_pa_features(SpaprMachineState *spapr,
177                                  PowerPCCPU *cpu,
178                                  void *fdt, int offset)
179 {
180     /*
181      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
182      * but not MTTCG, so disable it. To advertise it, a cap would have
183      * to be added, or support implemented for MTTCG.
184      *
185      * Copy/paste is not supported by TCG, so it is not advertised. KVM
186      * can execute them but it has no accelerator drivers which are usable,
187      * so there isn't much need for it anyway.
188      */
189 
190     /* These should be kept in sync with pnv */
191     uint8_t pa_features_206[] = { 6, 0,
192         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
193     uint8_t pa_features_207[] = { 24, 0,
194         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
195         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
196         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
197         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
198     uint8_t pa_features_300[] = { 66, 0,
199         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
200         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
201         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
202         /* 6: DS207 */
203         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
204         /* 16: Vector */
205         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
206         /* 18: Vec. Scalar, 20: Vec. XOR */
207         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
208         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
209         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
210         /* 32: LE atomic, 34: EBB + ext EBB */
211         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
212         /* 40: Radix MMU */
213         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
214         /* 42: PM, 44: PC RA, 46: SC vec'd */
215         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
216         /* 48: SIMD, 50: QP BFP, 52: String */
217         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
218         /* 54: DecFP, 56: DecI, 58: SHA */
219         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
220         /* 60: NM atomic, 62: RNG */
221         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
222     };
223     /* 3.1 removes SAO, HTM support */
224     uint8_t pa_features_31[] = { 74, 0,
225         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
226         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
227         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
228         /* 6: DS207 */
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
230         /* 16: Vector */
231         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
232         /* 18: Vec. Scalar, 20: Vec. XOR */
233         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
234         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
236         /* 32: LE atomic, 34: EBB + ext EBB */
237         0x00, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
238         /* 40: Radix MMU */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
240         /* 42: PM, 44: PC RA, 46: SC vec'd */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
242         /* 48: SIMD, 50: QP BFP, 52: String */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
244         /* 54: DecFP, 56: DecI, 58: SHA */
245         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
246         /* 60: NM atomic, 62: RNG */
247         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
248         /* 68: DEXCR[SBHE|IBRTPDUS|SRAPD|NPHIE|PHIE] */
249         0x00, 0x00, 0xce, 0x00, 0x00, 0x00, /* 66 - 71 */
250         /* 72: [P]HASHST/[P]HASHCHK */
251         0x80, 0x00,                         /* 72 - 73 */
252     };
253     uint8_t *pa_features = NULL;
254     size_t pa_size;
255 
256     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
257         pa_features = pa_features_206;
258         pa_size = sizeof(pa_features_206);
259     }
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_207;
262         pa_size = sizeof(pa_features_207);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_300;
266         pa_size = sizeof(pa_features_300);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_10, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_31;
270         pa_size = sizeof(pa_features_31);
271     }
272     if (!pa_features) {
273         return;
274     }
275 
276     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
277         /*
278          * Note: we keep CI large pages off by default because a 64K capable
279          * guest provisioned with large pages might otherwise try to map a qemu
280          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281          * even if that qemu runs on a 4k host.
282          * We dd this bit back here if we are confident this is not an issue
283          */
284         pa_features[3] |= 0x20;
285     }
286     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287         pa_features[24] |= 0x80;    /* Transactional memory support */
288     }
289     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290         /* Workaround for broken kernels that attempt (guest) radix
291          * mode when they can't handle it, if they see the radix bit set
292          * in pa-features. So hide it from them. */
293         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
294     }
295 
296     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
297 }
298 
299 static void spapr_dt_pi_features(SpaprMachineState *spapr,
300                                  PowerPCCPU *cpu,
301                                  void *fdt, int offset)
302 {
303     uint8_t pi_features[] = { 1, 0,
304         0x00 };
305 
306     if (kvm_enabled() && ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00,
307                                           0, cpu->compat_pvr)) {
308         /*
309          * POWER9 and later CPUs with KVM run in LPAR-per-thread mode where
310          * all threads are essentially independent CPUs, and msgsndp does not
311          * work (because it is physically-addressed) and therefore is
312          * emulated by KVM, so disable it here to ensure XIVE will be used.
313          * This is both KVM and CPU implementation-specific behaviour so a KVM
314          * cap would be cleanest, but for now this works. If KVM ever permits
315          * native msgsndp execution by guests, a cap could be added at that
316          * time.
317          */
318         pi_features[2] |= 0x08; /* 4: No msgsndp */
319     }
320 
321     _FDT((fdt_setprop(fdt, offset, "ibm,pi-features", pi_features,
322                       sizeof(pi_features))));
323 }
324 
325 static hwaddr spapr_node0_size(MachineState *machine)
326 {
327     if (machine->numa_state->num_nodes) {
328         int i;
329         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
330             if (machine->numa_state->nodes[i].node_mem) {
331                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
332                            machine->ram_size);
333             }
334         }
335     }
336     return machine->ram_size;
337 }
338 
339 static void add_str(GString *s, const gchar *s1)
340 {
341     g_string_append_len(s, s1, strlen(s1) + 1);
342 }
343 
344 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
345                                 hwaddr start, hwaddr size)
346 {
347     char mem_name[32];
348     uint64_t mem_reg_property[2];
349     int off;
350 
351     mem_reg_property[0] = cpu_to_be64(start);
352     mem_reg_property[1] = cpu_to_be64(size);
353 
354     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
355     off = fdt_add_subnode(fdt, 0, mem_name);
356     _FDT(off);
357     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
358     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
359                       sizeof(mem_reg_property))));
360     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
361     return off;
362 }
363 
364 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
365 {
366     MemoryDeviceInfoList *info;
367 
368     for (info = list; info; info = info->next) {
369         MemoryDeviceInfo *value = info->value;
370 
371         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
372             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
373 
374             if (addr >= pcdimm_info->addr &&
375                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
376                 return pcdimm_info->node;
377             }
378         }
379     }
380 
381     return -1;
382 }
383 
384 struct sPAPRDrconfCellV2 {
385      uint32_t seq_lmbs;
386      uint64_t base_addr;
387      uint32_t drc_index;
388      uint32_t aa_index;
389      uint32_t flags;
390 } QEMU_PACKED;
391 
392 typedef struct DrconfCellQueue {
393     struct sPAPRDrconfCellV2 cell;
394     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
395 } DrconfCellQueue;
396 
397 static DrconfCellQueue *
398 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
399                       uint32_t drc_index, uint32_t aa_index,
400                       uint32_t flags)
401 {
402     DrconfCellQueue *elem;
403 
404     elem = g_malloc0(sizeof(*elem));
405     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
406     elem->cell.base_addr = cpu_to_be64(base_addr);
407     elem->cell.drc_index = cpu_to_be32(drc_index);
408     elem->cell.aa_index = cpu_to_be32(aa_index);
409     elem->cell.flags = cpu_to_be32(flags);
410 
411     return elem;
412 }
413 
414 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
415                                       int offset, MemoryDeviceInfoList *dimms)
416 {
417     MachineState *machine = MACHINE(spapr);
418     uint8_t *int_buf, *cur_index;
419     int ret;
420     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
421     uint64_t addr, cur_addr, size;
422     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
423     uint64_t mem_end = machine->device_memory->base +
424                        memory_region_size(&machine->device_memory->mr);
425     uint32_t node, buf_len, nr_entries = 0;
426     SpaprDrc *drc;
427     DrconfCellQueue *elem, *next;
428     MemoryDeviceInfoList *info;
429     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
430         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
431 
432     /* Entry to cover RAM and the gap area */
433     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
434                                  SPAPR_LMB_FLAGS_RESERVED |
435                                  SPAPR_LMB_FLAGS_DRC_INVALID);
436     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
437     nr_entries++;
438 
439     cur_addr = machine->device_memory->base;
440     for (info = dimms; info; info = info->next) {
441         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
442 
443         addr = di->addr;
444         size = di->size;
445         node = di->node;
446 
447         /*
448          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
449          * area is marked hotpluggable in the next iteration for the bigger
450          * chunk including the NVDIMM occupied area.
451          */
452         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
453             continue;
454 
455         /* Entry for hot-pluggable area */
456         if (cur_addr < addr) {
457             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
458             g_assert(drc);
459             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
460                                          cur_addr, spapr_drc_index(drc), -1, 0);
461             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
462             nr_entries++;
463         }
464 
465         /* Entry for DIMM */
466         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
467         g_assert(drc);
468         elem = spapr_get_drconf_cell(size / lmb_size, addr,
469                                      spapr_drc_index(drc), node,
470                                      (SPAPR_LMB_FLAGS_ASSIGNED |
471                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
472         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
473         nr_entries++;
474         cur_addr = addr + size;
475     }
476 
477     /* Entry for remaining hotpluggable area */
478     if (cur_addr < mem_end) {
479         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
480         g_assert(drc);
481         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
482                                      cur_addr, spapr_drc_index(drc), -1, 0);
483         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
484         nr_entries++;
485     }
486 
487     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
488     int_buf = cur_index = g_malloc0(buf_len);
489     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
490     cur_index += sizeof(nr_entries);
491 
492     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
493         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
494         cur_index += sizeof(elem->cell);
495         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
496         g_free(elem);
497     }
498 
499     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
500     g_free(int_buf);
501     if (ret < 0) {
502         return -1;
503     }
504     return 0;
505 }
506 
507 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
508                                    int offset, MemoryDeviceInfoList *dimms)
509 {
510     MachineState *machine = MACHINE(spapr);
511     int i, ret;
512     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
513     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
514     uint32_t nr_lmbs = (machine->device_memory->base +
515                        memory_region_size(&machine->device_memory->mr)) /
516                        lmb_size;
517     uint32_t *int_buf, *cur_index, buf_len;
518 
519     /*
520      * Allocate enough buffer size to fit in ibm,dynamic-memory
521      */
522     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
523     cur_index = int_buf = g_malloc0(buf_len);
524     int_buf[0] = cpu_to_be32(nr_lmbs);
525     cur_index++;
526     for (i = 0; i < nr_lmbs; i++) {
527         uint64_t addr = i * lmb_size;
528         uint32_t *dynamic_memory = cur_index;
529 
530         if (i >= device_lmb_start) {
531             SpaprDrc *drc;
532 
533             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
534             g_assert(drc);
535 
536             dynamic_memory[0] = cpu_to_be32(addr >> 32);
537             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
538             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
539             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
540             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
541             if (memory_region_present(get_system_memory(), addr)) {
542                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
543             } else {
544                 dynamic_memory[5] = cpu_to_be32(0);
545             }
546         } else {
547             /*
548              * LMB information for RMA, boot time RAM and gap b/n RAM and
549              * device memory region -- all these are marked as reserved
550              * and as having no valid DRC.
551              */
552             dynamic_memory[0] = cpu_to_be32(addr >> 32);
553             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
554             dynamic_memory[2] = cpu_to_be32(0);
555             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
556             dynamic_memory[4] = cpu_to_be32(-1);
557             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
558                                             SPAPR_LMB_FLAGS_DRC_INVALID);
559         }
560 
561         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
562     }
563     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
564     g_free(int_buf);
565     if (ret < 0) {
566         return -1;
567     }
568     return 0;
569 }
570 
571 /*
572  * Adds ibm,dynamic-reconfiguration-memory node.
573  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
574  * of this device tree node.
575  */
576 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
577                                                    void *fdt)
578 {
579     MachineState *machine = MACHINE(spapr);
580     int ret, offset;
581     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
582     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
583                                 cpu_to_be32(lmb_size & 0xffffffff)};
584     MemoryDeviceInfoList *dimms = NULL;
585 
586     /* Don't create the node if there is no device memory. */
587     if (!machine->device_memory) {
588         return 0;
589     }
590 
591     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
592 
593     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
594                     sizeof(prop_lmb_size));
595     if (ret < 0) {
596         return ret;
597     }
598 
599     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
600     if (ret < 0) {
601         return ret;
602     }
603 
604     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
605     if (ret < 0) {
606         return ret;
607     }
608 
609     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
610     dimms = qmp_memory_device_list();
611     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
612         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
613     } else {
614         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
615     }
616     qapi_free_MemoryDeviceInfoList(dimms);
617 
618     if (ret < 0) {
619         return ret;
620     }
621 
622     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
623 
624     return ret;
625 }
626 
627 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
628 {
629     MachineState *machine = MACHINE(spapr);
630     hwaddr mem_start, node_size;
631     int i, nb_nodes = machine->numa_state->num_nodes;
632     NodeInfo *nodes = machine->numa_state->nodes;
633 
634     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
635         if (!nodes[i].node_mem) {
636             continue;
637         }
638         if (mem_start >= machine->ram_size) {
639             node_size = 0;
640         } else {
641             node_size = nodes[i].node_mem;
642             if (node_size > machine->ram_size - mem_start) {
643                 node_size = machine->ram_size - mem_start;
644             }
645         }
646         if (!mem_start) {
647             /* spapr_machine_init() checks for rma_size <= node0_size
648              * already */
649             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
650             mem_start += spapr->rma_size;
651             node_size -= spapr->rma_size;
652         }
653         for ( ; node_size; ) {
654             hwaddr sizetmp = pow2floor(node_size);
655 
656             /* mem_start != 0 here */
657             if (ctzl(mem_start) < ctzl(sizetmp)) {
658                 sizetmp = 1ULL << ctzl(mem_start);
659             }
660 
661             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
662             node_size -= sizetmp;
663             mem_start += sizetmp;
664         }
665     }
666 
667     /* Generate ibm,dynamic-reconfiguration-memory node if required */
668     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
669         int ret;
670 
671         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
672         if (ret) {
673             return ret;
674         }
675     }
676 
677     return 0;
678 }
679 
680 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
681                          SpaprMachineState *spapr)
682 {
683     MachineState *ms = MACHINE(spapr);
684     PowerPCCPU *cpu = POWERPC_CPU(cs);
685     CPUPPCState *env = &cpu->env;
686     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
687     int index = spapr_get_vcpu_id(cpu);
688     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
689                        0xffffffff, 0xffffffff};
690     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
691         : SPAPR_TIMEBASE_FREQ;
692     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
693     uint32_t page_sizes_prop[64];
694     size_t page_sizes_prop_size;
695     unsigned int smp_threads = ms->smp.threads;
696     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
697     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
698     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
699     SpaprDrc *drc;
700     int drc_index;
701     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
702     int i;
703 
704     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, env->core_index);
705     if (drc) {
706         drc_index = spapr_drc_index(drc);
707         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
708     }
709 
710     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
711     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
712 
713     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
714     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
715                            env->dcache_line_size)));
716     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
717                            env->dcache_line_size)));
718     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
719                            env->icache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
721                            env->icache_line_size)));
722 
723     if (pcc->l1_dcache_size) {
724         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
725                                pcc->l1_dcache_size)));
726     } else {
727         warn_report("Unknown L1 dcache size for cpu");
728     }
729     if (pcc->l1_icache_size) {
730         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
731                                pcc->l1_icache_size)));
732     } else {
733         warn_report("Unknown L1 icache size for cpu");
734     }
735 
736     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
737     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
738     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
739     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
740     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
741     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
742 
743     if (ppc_has_spr(cpu, SPR_PURR)) {
744         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
745     }
746     if (ppc_has_spr(cpu, SPR_PURR)) {
747         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
748     }
749 
750     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
751         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
752                           segs, sizeof(segs))));
753     }
754 
755     /* Advertise VSX (vector extensions) if available
756      *   1               == VMX / Altivec available
757      *   2               == VSX available
758      *
759      * Only CPUs for which we create core types in spapr_cpu_core.c
760      * are possible, and all of those have VMX */
761     if (env->insns_flags & PPC_ALTIVEC) {
762         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
763             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
764         } else {
765             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
766         }
767     }
768 
769     /* Advertise DFP (Decimal Floating Point) if available
770      *   0 / no property == no DFP
771      *   1               == DFP available */
772     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
773         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
774     }
775 
776     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
777                                                       sizeof(page_sizes_prop));
778     if (page_sizes_prop_size) {
779         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
780                           page_sizes_prop, page_sizes_prop_size)));
781     }
782 
783     spapr_dt_pa_features(spapr, cpu, fdt, offset);
784 
785     spapr_dt_pi_features(spapr, cpu, fdt, offset);
786 
787     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
788                            cs->cpu_index / vcpus_per_socket)));
789 
790     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
791                       pft_size_prop, sizeof(pft_size_prop))));
792 
793     if (ms->numa_state->num_nodes > 1) {
794         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
795     }
796 
797     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
798 
799     if (pcc->radix_page_info) {
800         for (i = 0; i < pcc->radix_page_info->count; i++) {
801             radix_AP_encodings[i] =
802                 cpu_to_be32(pcc->radix_page_info->entries[i]);
803         }
804         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
805                           radix_AP_encodings,
806                           pcc->radix_page_info->count *
807                           sizeof(radix_AP_encodings[0]))));
808     }
809 
810     /*
811      * We set this property to let the guest know that it can use the large
812      * decrementer and its width in bits.
813      */
814     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
815         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
816                               pcc->lrg_decr_bits)));
817 }
818 
819 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
820                              int cpus_offset)
821 {
822     PowerPCCPU *cpu = POWERPC_CPU(cs);
823     int index = spapr_get_vcpu_id(cpu);
824     DeviceClass *dc = DEVICE_GET_CLASS(cs);
825     g_autofree char *nodename = NULL;
826     int offset;
827 
828     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
829         return;
830     }
831 
832     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
833     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
834     _FDT(offset);
835     spapr_dt_cpu(cs, fdt, offset, spapr);
836 }
837 
838 
839 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
840 {
841     CPUState **rev;
842     CPUState *cs;
843     int n_cpus;
844     int cpus_offset;
845     int i;
846 
847     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
848     _FDT(cpus_offset);
849     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
850     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
851 
852     /*
853      * We walk the CPUs in reverse order to ensure that CPU DT nodes
854      * created by fdt_add_subnode() end up in the right order in FDT
855      * for the guest kernel the enumerate the CPUs correctly.
856      *
857      * The CPU list cannot be traversed in reverse order, so we need
858      * to do extra work.
859      */
860     n_cpus = 0;
861     rev = NULL;
862     CPU_FOREACH(cs) {
863         rev = g_renew(CPUState *, rev, n_cpus + 1);
864         rev[n_cpus++] = cs;
865     }
866 
867     for (i = n_cpus - 1; i >= 0; i--) {
868         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
869     }
870 
871     g_free(rev);
872 }
873 
874 static int spapr_dt_rng(void *fdt)
875 {
876     int node;
877     int ret;
878 
879     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
880     if (node <= 0) {
881         return -1;
882     }
883     ret = fdt_setprop_string(fdt, node, "device_type",
884                              "ibm,platform-facilities");
885     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
886     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
887 
888     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
889     if (node <= 0) {
890         return -1;
891     }
892     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
893 
894     return ret ? -1 : 0;
895 }
896 
897 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
898 {
899     MachineState *ms = MACHINE(spapr);
900     int rtas;
901     GString *hypertas = g_string_sized_new(256);
902     GString *qemu_hypertas = g_string_sized_new(256);
903     uint32_t lrdr_capacity[] = {
904         0,
905         0,
906         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
907         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
908         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
909     };
910 
911     /* Do we have device memory? */
912     if (MACHINE(spapr)->device_memory) {
913         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
914             memory_region_size(&MACHINE(spapr)->device_memory->mr);
915 
916         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
917         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
918     }
919 
920     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
921 
922     /* hypertas */
923     add_str(hypertas, "hcall-pft");
924     add_str(hypertas, "hcall-term");
925     add_str(hypertas, "hcall-dabr");
926     add_str(hypertas, "hcall-interrupt");
927     add_str(hypertas, "hcall-tce");
928     add_str(hypertas, "hcall-vio");
929     add_str(hypertas, "hcall-splpar");
930     add_str(hypertas, "hcall-join");
931     add_str(hypertas, "hcall-bulk");
932     add_str(hypertas, "hcall-set-mode");
933     add_str(hypertas, "hcall-sprg0");
934     add_str(hypertas, "hcall-copy");
935     add_str(hypertas, "hcall-debug");
936     add_str(hypertas, "hcall-vphn");
937     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
938         add_str(hypertas, "hcall-rpt-invalidate");
939     }
940 
941     add_str(qemu_hypertas, "hcall-memop1");
942 
943     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
944         add_str(hypertas, "hcall-multi-tce");
945     }
946 
947     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
948         add_str(hypertas, "hcall-hpt-resize");
949     }
950 
951     add_str(hypertas, "hcall-watchdog");
952 
953     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
954                      hypertas->str, hypertas->len));
955     g_string_free(hypertas, TRUE);
956     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
957                      qemu_hypertas->str, qemu_hypertas->len));
958     g_string_free(qemu_hypertas, TRUE);
959 
960     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
961 
962     /*
963      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
964      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
965      *
966      * The system reset requirements are driven by existing Linux and PowerVM
967      * implementation which (contrary to PAPR) saves r3 in the error log
968      * structure like machine check, so Linux expects to find the saved r3
969      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
970      * does not look at the error value).
971      *
972      * System reset interrupts are not subject to interlock like machine
973      * check, so this memory area could be corrupted if the sreset is
974      * interrupted by a machine check (or vice versa) if it was shared. To
975      * prevent this, system reset uses per-CPU areas for the sreset save
976      * area. A system reset that interrupts a system reset handler could
977      * still overwrite this area, but Linux doesn't try to recover in that
978      * case anyway.
979      *
980      * The extra 8 bytes is required because Linux's FWNMI error log check
981      * is off-by-one.
982      *
983      * RTAS_MIN_SIZE is required for the RTAS blob itself.
984      */
985     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
986                           RTAS_ERROR_LOG_MAX +
987                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
988                           sizeof(uint64_t)));
989     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
990                           RTAS_ERROR_LOG_MAX));
991     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
992                           RTAS_EVENT_SCAN_RATE));
993 
994     g_assert(msi_nonbroken);
995     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
996 
997     /*
998      * According to PAPR, rtas ibm,os-term does not guarantee a return
999      * back to the guest cpu.
1000      *
1001      * While an additional ibm,extended-os-term property indicates
1002      * that rtas call return will always occur. Set this property.
1003      */
1004     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1005 
1006     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1007                      lrdr_capacity, sizeof(lrdr_capacity)));
1008 
1009     spapr_dt_rtas_tokens(fdt, rtas);
1010 }
1011 
1012 /*
1013  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1014  * and the XIVE features that the guest may request and thus the valid
1015  * values for bytes 23..26 of option vector 5:
1016  */
1017 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1018                                           int chosen)
1019 {
1020     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1021 
1022     char val[2 * 4] = {
1023         23, 0x00, /* XICS / XIVE mode */
1024         24, 0x00, /* Hash/Radix, filled in below. */
1025         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1026         26, 0x40, /* Radix options: GTSE == yes. */
1027     };
1028 
1029     if (spapr->irq->xics && spapr->irq->xive) {
1030         val[1] = SPAPR_OV5_XIVE_BOTH;
1031     } else if (spapr->irq->xive) {
1032         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1033     } else {
1034         assert(spapr->irq->xics);
1035         val[1] = SPAPR_OV5_XIVE_LEGACY;
1036     }
1037 
1038     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1039                           first_ppc_cpu->compat_pvr)) {
1040         /*
1041          * If we're in a pre POWER9 compat mode then the guest should
1042          * do hash and use the legacy interrupt mode
1043          */
1044         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1045         val[3] = 0x00; /* Hash */
1046         spapr_check_mmu_mode(false);
1047     } else if (kvm_enabled()) {
1048         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1049             val[3] = 0x80; /* OV5_MMU_BOTH */
1050         } else if (kvmppc_has_cap_mmu_radix()) {
1051             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1052         } else {
1053             val[3] = 0x00; /* Hash */
1054         }
1055     } else {
1056         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1057         val[3] = 0xC0;
1058     }
1059     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1060                      val, sizeof(val)));
1061 }
1062 
1063 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1064 {
1065     MachineState *machine = MACHINE(spapr);
1066     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1067     int chosen;
1068 
1069     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1070 
1071     if (reset) {
1072         const char *boot_device = spapr->boot_device;
1073         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1074         size_t cb = 0;
1075         g_autofree char *bootlist = get_boot_devices_list(&cb);
1076 
1077         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1078             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1079                                     machine->kernel_cmdline));
1080         }
1081 
1082         if (spapr->initrd_size) {
1083             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1084                                   spapr->initrd_base));
1085             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1086                                   spapr->initrd_base + spapr->initrd_size));
1087         }
1088 
1089         if (spapr->kernel_size) {
1090             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1091                                   cpu_to_be64(spapr->kernel_size) };
1092 
1093             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1094                          &kprop, sizeof(kprop)));
1095             if (spapr->kernel_le) {
1096                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1097             }
1098         }
1099         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1100             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1101         }
1102         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1103         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1104         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1105 
1106         if (cb && bootlist) {
1107             int i;
1108 
1109             for (i = 0; i < cb; i++) {
1110                 if (bootlist[i] == '\n') {
1111                     bootlist[i] = ' ';
1112                 }
1113             }
1114             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1115         }
1116 
1117         if (boot_device && strlen(boot_device)) {
1118             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1119         }
1120 
1121         if (spapr->want_stdout_path && stdout_path) {
1122             /*
1123              * "linux,stdout-path" and "stdout" properties are
1124              * deprecated by linux kernel. New platforms should only
1125              * use the "stdout-path" property. Set the new property
1126              * and continue using older property to remain compatible
1127              * with the existing firmware.
1128              */
1129             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1130             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1131         }
1132 
1133         /*
1134          * We can deal with BAR reallocation just fine, advertise it
1135          * to the guest
1136          */
1137         if (smc->linux_pci_probe) {
1138             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1139         }
1140 
1141         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1142     }
1143 
1144     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1145 
1146     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1147 }
1148 
1149 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1150 {
1151     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1152      * KVM to work under pHyp with some guest co-operation */
1153     int hypervisor;
1154     uint8_t hypercall[16];
1155 
1156     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1157     /* indicate KVM hypercall interface */
1158     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1159     if (kvmppc_has_cap_fixup_hcalls()) {
1160         /*
1161          * Older KVM versions with older guest kernels were broken
1162          * with the magic page, don't allow the guest to map it.
1163          */
1164         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1165                                   sizeof(hypercall))) {
1166             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1167                              hypercall, sizeof(hypercall)));
1168         }
1169     }
1170 }
1171 
1172 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1173 {
1174     MachineState *machine = MACHINE(spapr);
1175     MachineClass *mc = MACHINE_GET_CLASS(machine);
1176     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1177     uint32_t root_drc_type_mask = 0;
1178     int ret;
1179     void *fdt;
1180     SpaprPhbState *phb;
1181     char *buf;
1182 
1183     fdt = g_malloc0(space);
1184     _FDT((fdt_create_empty_tree(fdt, space)));
1185 
1186     /* Root node */
1187     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1188     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1189     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1190 
1191     /* Guest UUID & Name*/
1192     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1193     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1194     if (qemu_uuid_set) {
1195         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1196     }
1197     g_free(buf);
1198 
1199     if (qemu_get_vm_name()) {
1200         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1201                                 qemu_get_vm_name()));
1202     }
1203 
1204     /* Host Model & Serial Number */
1205     if (spapr->host_model) {
1206         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1207     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1208         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1209         g_free(buf);
1210     }
1211 
1212     if (spapr->host_serial) {
1213         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1214     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1215         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1216         g_free(buf);
1217     }
1218 
1219     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1220     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1221 
1222     /* /interrupt controller */
1223     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1224 
1225     ret = spapr_dt_memory(spapr, fdt);
1226     if (ret < 0) {
1227         error_report("couldn't setup memory nodes in fdt");
1228         exit(1);
1229     }
1230 
1231     /* /vdevice */
1232     spapr_dt_vdevice(spapr->vio_bus, fdt);
1233 
1234     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1235         ret = spapr_dt_rng(fdt);
1236         if (ret < 0) {
1237             error_report("could not set up rng device in the fdt");
1238             exit(1);
1239         }
1240     }
1241 
1242     QLIST_FOREACH(phb, &spapr->phbs, list) {
1243         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1244         if (ret < 0) {
1245             error_report("couldn't setup PCI devices in fdt");
1246             exit(1);
1247         }
1248     }
1249 
1250     spapr_dt_cpus(fdt, spapr);
1251 
1252     /* ibm,drc-indexes and friends */
1253     root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1254     if (smc->dr_phb_enabled) {
1255         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1256     }
1257     if (mc->nvdimm_supported) {
1258         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1259     }
1260     if (root_drc_type_mask) {
1261         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1262     }
1263 
1264     if (mc->has_hotpluggable_cpus) {
1265         int offset = fdt_path_offset(fdt, "/cpus");
1266         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1267         if (ret < 0) {
1268             error_report("Couldn't set up CPU DR device tree properties");
1269             exit(1);
1270         }
1271     }
1272 
1273     /* /event-sources */
1274     spapr_dt_events(spapr, fdt);
1275 
1276     /* /rtas */
1277     spapr_dt_rtas(spapr, fdt);
1278 
1279     /* /chosen */
1280     spapr_dt_chosen(spapr, fdt, reset);
1281 
1282     /* /hypervisor */
1283     if (kvm_enabled()) {
1284         spapr_dt_hypervisor(spapr, fdt);
1285     }
1286 
1287     /* Build memory reserve map */
1288     if (reset) {
1289         if (spapr->kernel_size) {
1290             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1291                                   spapr->kernel_size)));
1292         }
1293         if (spapr->initrd_size) {
1294             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1295                                   spapr->initrd_size)));
1296         }
1297     }
1298 
1299     /* NVDIMM devices */
1300     if (mc->nvdimm_supported) {
1301         spapr_dt_persistent_memory(spapr, fdt);
1302     }
1303 
1304     return fdt;
1305 }
1306 
1307 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1308 {
1309     SpaprMachineState *spapr = opaque;
1310 
1311     return (addr & 0x0fffffff) + spapr->kernel_addr;
1312 }
1313 
1314 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1315                                     PowerPCCPU *cpu)
1316 {
1317     CPUPPCState *env = &cpu->env;
1318 
1319     /* The TCG path should also be holding the BQL at this point */
1320     g_assert(bql_locked());
1321 
1322     g_assert(!vhyp_cpu_in_nested(cpu));
1323 
1324     if (FIELD_EX64(env->msr, MSR, PR)) {
1325         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1326         env->gpr[3] = H_PRIVILEGE;
1327     } else {
1328         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1329     }
1330 }
1331 
1332 struct LPCRSyncState {
1333     target_ulong value;
1334     target_ulong mask;
1335 };
1336 
1337 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1338 {
1339     struct LPCRSyncState *s = arg.host_ptr;
1340     PowerPCCPU *cpu = POWERPC_CPU(cs);
1341     CPUPPCState *env = &cpu->env;
1342     target_ulong lpcr;
1343 
1344     cpu_synchronize_state(cs);
1345     lpcr = env->spr[SPR_LPCR];
1346     lpcr &= ~s->mask;
1347     lpcr |= s->value;
1348     ppc_store_lpcr(cpu, lpcr);
1349 }
1350 
1351 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1352 {
1353     CPUState *cs;
1354     struct LPCRSyncState s = {
1355         .value = value,
1356         .mask = mask
1357     };
1358     CPU_FOREACH(cs) {
1359         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1360     }
1361 }
1362 
1363 /* May be used when the machine is not running */
1364 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1365 {
1366     CPUState *cs;
1367     CPU_FOREACH(cs) {
1368         PowerPCCPU *cpu = POWERPC_CPU(cs);
1369         CPUPPCState *env = &cpu->env;
1370         target_ulong lpcr;
1371 
1372         lpcr = env->spr[SPR_LPCR];
1373         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1374         ppc_store_lpcr(cpu, lpcr);
1375     }
1376 }
1377 
1378 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1379                            target_ulong lpid, ppc_v3_pate_t *entry)
1380 {
1381     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1382     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1383 
1384     if (!spapr_cpu->in_nested) {
1385         assert(lpid == 0);
1386 
1387         /* Copy PATE1:GR into PATE0:HR */
1388         entry->dw0 = spapr->patb_entry & PATE0_HR;
1389         entry->dw1 = spapr->patb_entry;
1390         return true;
1391     } else {
1392         if (spapr_nested_api(spapr) == NESTED_API_KVM_HV) {
1393             return spapr_get_pate_nested_hv(spapr, cpu, lpid, entry);
1394         } else if (spapr_nested_api(spapr) == NESTED_API_PAPR) {
1395             return spapr_get_pate_nested_papr(spapr, cpu, lpid, entry);
1396         } else {
1397             g_assert_not_reached();
1398         }
1399     }
1400 }
1401 
1402 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1403 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1404 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1405 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1406 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1407 
1408 /*
1409  * Get the fd to access the kernel htab, re-opening it if necessary
1410  */
1411 static int get_htab_fd(SpaprMachineState *spapr)
1412 {
1413     Error *local_err = NULL;
1414 
1415     if (spapr->htab_fd >= 0) {
1416         return spapr->htab_fd;
1417     }
1418 
1419     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1420     if (spapr->htab_fd < 0) {
1421         error_report_err(local_err);
1422     }
1423 
1424     return spapr->htab_fd;
1425 }
1426 
1427 void close_htab_fd(SpaprMachineState *spapr)
1428 {
1429     if (spapr->htab_fd >= 0) {
1430         close(spapr->htab_fd);
1431     }
1432     spapr->htab_fd = -1;
1433 }
1434 
1435 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1436 {
1437     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1438 
1439     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1440 }
1441 
1442 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1443 {
1444     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1445 
1446     assert(kvm_enabled());
1447 
1448     if (!spapr->htab) {
1449         return 0;
1450     }
1451 
1452     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1453 }
1454 
1455 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1456                                                 hwaddr ptex, int n)
1457 {
1458     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1459     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1460 
1461     if (!spapr->htab) {
1462         /*
1463          * HTAB is controlled by KVM. Fetch into temporary buffer
1464          */
1465         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1466         kvmppc_read_hptes(hptes, ptex, n);
1467         return hptes;
1468     }
1469 
1470     /*
1471      * HTAB is controlled by QEMU. Just point to the internally
1472      * accessible PTEG.
1473      */
1474     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1475 }
1476 
1477 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1478                               const ppc_hash_pte64_t *hptes,
1479                               hwaddr ptex, int n)
1480 {
1481     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1482 
1483     if (!spapr->htab) {
1484         g_free((void *)hptes);
1485     }
1486 
1487     /* Nothing to do for qemu managed HPT */
1488 }
1489 
1490 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1491                       uint64_t pte0, uint64_t pte1)
1492 {
1493     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1494     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1495 
1496     if (!spapr->htab) {
1497         kvmppc_write_hpte(ptex, pte0, pte1);
1498     } else {
1499         if (pte0 & HPTE64_V_VALID) {
1500             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1501             /*
1502              * When setting valid, we write PTE1 first. This ensures
1503              * proper synchronization with the reading code in
1504              * ppc_hash64_pteg_search()
1505              */
1506             smp_wmb();
1507             stq_p(spapr->htab + offset, pte0);
1508         } else {
1509             stq_p(spapr->htab + offset, pte0);
1510             /*
1511              * When clearing it we set PTE0 first. This ensures proper
1512              * synchronization with the reading code in
1513              * ppc_hash64_pteg_search()
1514              */
1515             smp_wmb();
1516             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1517         }
1518     }
1519 }
1520 
1521 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1522                              uint64_t pte1)
1523 {
1524     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1525     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1526 
1527     if (!spapr->htab) {
1528         /* There should always be a hash table when this is called */
1529         error_report("spapr_hpte_set_c called with no hash table !");
1530         return;
1531     }
1532 
1533     /* The HW performs a non-atomic byte update */
1534     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1535 }
1536 
1537 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1538                              uint64_t pte1)
1539 {
1540     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1541     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1542 
1543     if (!spapr->htab) {
1544         /* There should always be a hash table when this is called */
1545         error_report("spapr_hpte_set_r called with no hash table !");
1546         return;
1547     }
1548 
1549     /* The HW performs a non-atomic byte update */
1550     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1551 }
1552 
1553 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1554 {
1555     int shift;
1556 
1557     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1558      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1559      * that's much more than is needed for Linux guests */
1560     shift = ctz64(pow2ceil(ramsize)) - 7;
1561     shift = MAX(shift, 18); /* Minimum architected size */
1562     shift = MIN(shift, 46); /* Maximum architected size */
1563     return shift;
1564 }
1565 
1566 void spapr_free_hpt(SpaprMachineState *spapr)
1567 {
1568     qemu_vfree(spapr->htab);
1569     spapr->htab = NULL;
1570     spapr->htab_shift = 0;
1571     close_htab_fd(spapr);
1572 }
1573 
1574 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1575 {
1576     ERRP_GUARD();
1577     long rc;
1578 
1579     /* Clean up any HPT info from a previous boot */
1580     spapr_free_hpt(spapr);
1581 
1582     rc = kvmppc_reset_htab(shift);
1583 
1584     if (rc == -EOPNOTSUPP) {
1585         error_setg(errp, "HPT not supported in nested guests");
1586         return -EOPNOTSUPP;
1587     }
1588 
1589     if (rc < 0) {
1590         /* kernel-side HPT needed, but couldn't allocate one */
1591         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1592                          shift);
1593         error_append_hint(errp, "Try smaller maxmem?\n");
1594         return -errno;
1595     } else if (rc > 0) {
1596         /* kernel-side HPT allocated */
1597         if (rc != shift) {
1598             error_setg(errp,
1599                        "Requested order %d HPT, but kernel allocated order %ld",
1600                        shift, rc);
1601             error_append_hint(errp, "Try smaller maxmem?\n");
1602             return -ENOSPC;
1603         }
1604 
1605         spapr->htab_shift = shift;
1606         spapr->htab = NULL;
1607     } else {
1608         /* kernel-side HPT not needed, allocate in userspace instead */
1609         size_t size = 1ULL << shift;
1610         int i;
1611 
1612         spapr->htab = qemu_memalign(size, size);
1613         memset(spapr->htab, 0, size);
1614         spapr->htab_shift = shift;
1615 
1616         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1617             DIRTY_HPTE(HPTE(spapr->htab, i));
1618         }
1619     }
1620     /* We're setting up a hash table, so that means we're not radix */
1621     spapr->patb_entry = 0;
1622     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1623     return 0;
1624 }
1625 
1626 void spapr_setup_hpt(SpaprMachineState *spapr)
1627 {
1628     int hpt_shift;
1629 
1630     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1631         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1632     } else {
1633         uint64_t current_ram_size;
1634 
1635         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1636         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1637     }
1638     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1639 
1640     if (kvm_enabled()) {
1641         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1642 
1643         /* Check our RMA fits in the possible VRMA */
1644         if (vrma_limit < spapr->rma_size) {
1645             error_report("Unable to create %" HWADDR_PRIu
1646                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1647                          spapr->rma_size / MiB, vrma_limit / MiB);
1648             exit(EXIT_FAILURE);
1649         }
1650     }
1651 }
1652 
1653 void spapr_check_mmu_mode(bool guest_radix)
1654 {
1655     if (guest_radix) {
1656         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1657             error_report("Guest requested unavailable MMU mode (radix).");
1658             exit(EXIT_FAILURE);
1659         }
1660     } else {
1661         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1662             && !kvmppc_has_cap_mmu_hash_v3()) {
1663             error_report("Guest requested unavailable MMU mode (hash).");
1664             exit(EXIT_FAILURE);
1665         }
1666     }
1667 }
1668 
1669 static void spapr_machine_reset(MachineState *machine, ResetType type)
1670 {
1671     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1672     PowerPCCPU *first_ppc_cpu;
1673     hwaddr fdt_addr;
1674     void *fdt;
1675     int rc;
1676 
1677     if (type != RESET_TYPE_SNAPSHOT_LOAD) {
1678         /*
1679          * Record-replay snapshot load must not consume random, this was
1680          * already replayed from initial machine reset.
1681          */
1682         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1683     }
1684 
1685     if (machine->cgs) {
1686         confidential_guest_kvm_reset(machine->cgs, &error_fatal);
1687     }
1688     spapr_caps_apply(spapr);
1689     spapr_nested_reset(spapr);
1690 
1691     first_ppc_cpu = POWERPC_CPU(first_cpu);
1692     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1693         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1694                               spapr->max_compat_pvr)) {
1695         /*
1696          * If using KVM with radix mode available, VCPUs can be started
1697          * without a HPT because KVM will start them in radix mode.
1698          * Set the GR bit in PATE so that we know there is no HPT.
1699          */
1700         spapr->patb_entry = PATE1_GR;
1701         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1702     } else {
1703         spapr_setup_hpt(spapr);
1704     }
1705 
1706     qemu_devices_reset(type);
1707 
1708     spapr_ovec_cleanup(spapr->ov5_cas);
1709     spapr->ov5_cas = spapr_ovec_new();
1710 
1711     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1712 
1713     /*
1714      * This is fixing some of the default configuration of the XIVE
1715      * devices. To be called after the reset of the machine devices.
1716      */
1717     spapr_irq_reset(spapr, &error_fatal);
1718 
1719     /*
1720      * There is no CAS under qtest. Simulate one to please the code that
1721      * depends on spapr->ov5_cas. This is especially needed to test device
1722      * unplug, so we do that before resetting the DRCs.
1723      */
1724     if (qtest_enabled()) {
1725         spapr_ovec_cleanup(spapr->ov5_cas);
1726         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1727     }
1728 
1729     spapr_nvdimm_finish_flushes();
1730 
1731     /* DRC reset may cause a device to be unplugged. This will cause troubles
1732      * if this device is used by another device (eg, a running vhost backend
1733      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1734      * situations, we reset DRCs after all devices have been reset.
1735      */
1736     spapr_drc_reset_all(spapr);
1737 
1738     spapr_clear_pending_events(spapr);
1739 
1740     /*
1741      * We place the device tree just below either the top of the RMA,
1742      * or just below 2GB, whichever is lower, so that it can be
1743      * processed with 32-bit real mode code if necessary
1744      */
1745     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1746 
1747     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1748     if (spapr->vof) {
1749         spapr_vof_reset(spapr, fdt, &error_fatal);
1750         /*
1751          * Do not pack the FDT as the client may change properties.
1752          * VOF client does not expect the FDT so we do not load it to the VM.
1753          */
1754     } else {
1755         rc = fdt_pack(fdt);
1756         /* Should only fail if we've built a corrupted tree */
1757         assert(rc == 0);
1758 
1759         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1760                                   0, fdt_addr, 0);
1761         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1762     }
1763 
1764     g_free(spapr->fdt_blob);
1765     spapr->fdt_size = fdt_totalsize(fdt);
1766     spapr->fdt_initial_size = spapr->fdt_size;
1767     spapr->fdt_blob = fdt;
1768 
1769     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1770     machine->fdt = fdt;
1771 
1772     /* Set up the entry state */
1773     first_ppc_cpu->env.gpr[5] = 0;
1774 
1775     spapr->fwnmi_system_reset_addr = -1;
1776     spapr->fwnmi_machine_check_addr = -1;
1777     spapr->fwnmi_machine_check_interlock = -1;
1778 
1779     /* Signal all vCPUs waiting on this condition */
1780     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1781 
1782     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1783 }
1784 
1785 static void spapr_create_nvram(SpaprMachineState *spapr)
1786 {
1787     DeviceState *dev = qdev_new("spapr-nvram");
1788     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1789 
1790     if (dinfo) {
1791         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1792                                 &error_fatal);
1793     }
1794 
1795     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1796 
1797     spapr->nvram = (struct SpaprNvram *)dev;
1798 }
1799 
1800 static void spapr_rtc_create(SpaprMachineState *spapr)
1801 {
1802     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1803                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1804                                        &error_fatal, NULL);
1805     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1806     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1807                               "date");
1808 }
1809 
1810 /* Returns whether we want to use VGA or not */
1811 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1812 {
1813     vga_interface_created = true;
1814     switch (vga_interface_type) {
1815     case VGA_NONE:
1816         return false;
1817     case VGA_DEVICE:
1818         return true;
1819     case VGA_STD:
1820     case VGA_VIRTIO:
1821     case VGA_CIRRUS:
1822         return pci_vga_init(pci_bus) != NULL;
1823     default:
1824         error_setg(errp,
1825                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1826         return false;
1827     }
1828 }
1829 
1830 static int spapr_pre_load(void *opaque)
1831 {
1832     int rc;
1833 
1834     rc = spapr_caps_pre_load(opaque);
1835     if (rc) {
1836         return rc;
1837     }
1838 
1839     return 0;
1840 }
1841 
1842 static int spapr_post_load(void *opaque, int version_id)
1843 {
1844     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1845     int err = 0;
1846 
1847     err = spapr_caps_post_migration(spapr);
1848     if (err) {
1849         return err;
1850     }
1851 
1852     /*
1853      * In earlier versions, there was no separate qdev for the PAPR
1854      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1855      * So when migrating from those versions, poke the incoming offset
1856      * value into the RTC device
1857      */
1858     if (version_id < 3) {
1859         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1860         if (err) {
1861             return err;
1862         }
1863     }
1864 
1865     if (kvm_enabled() && spapr->patb_entry) {
1866         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1867         bool radix = !!(spapr->patb_entry & PATE1_GR);
1868         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1869 
1870         /*
1871          * Update LPCR:HR and UPRT as they may not be set properly in
1872          * the stream
1873          */
1874         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1875                             LPCR_HR | LPCR_UPRT);
1876 
1877         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1878         if (err) {
1879             error_report("Process table config unsupported by the host");
1880             return -EINVAL;
1881         }
1882     }
1883 
1884     err = spapr_irq_post_load(spapr, version_id);
1885     if (err) {
1886         return err;
1887     }
1888 
1889     return err;
1890 }
1891 
1892 static int spapr_pre_save(void *opaque)
1893 {
1894     int rc;
1895 
1896     rc = spapr_caps_pre_save(opaque);
1897     if (rc) {
1898         return rc;
1899     }
1900 
1901     return 0;
1902 }
1903 
1904 static bool version_before_3(void *opaque, int version_id)
1905 {
1906     return version_id < 3;
1907 }
1908 
1909 static bool spapr_pending_events_needed(void *opaque)
1910 {
1911     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1912     return !QTAILQ_EMPTY(&spapr->pending_events);
1913 }
1914 
1915 static const VMStateDescription vmstate_spapr_event_entry = {
1916     .name = "spapr_event_log_entry",
1917     .version_id = 1,
1918     .minimum_version_id = 1,
1919     .fields = (const VMStateField[]) {
1920         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1921         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1922         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1923                                      NULL, extended_length),
1924         VMSTATE_END_OF_LIST()
1925     },
1926 };
1927 
1928 static const VMStateDescription vmstate_spapr_pending_events = {
1929     .name = "spapr_pending_events",
1930     .version_id = 1,
1931     .minimum_version_id = 1,
1932     .needed = spapr_pending_events_needed,
1933     .fields = (const VMStateField[]) {
1934         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1935                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1936         VMSTATE_END_OF_LIST()
1937     },
1938 };
1939 
1940 static bool spapr_ov5_cas_needed(void *opaque)
1941 {
1942     SpaprMachineState *spapr = opaque;
1943     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1944     bool cas_needed;
1945 
1946     /* Prior to the introduction of SpaprOptionVector, we had two option
1947      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1948      * Both of these options encode machine topology into the device-tree
1949      * in such a way that the now-booted OS should still be able to interact
1950      * appropriately with QEMU regardless of what options were actually
1951      * negotiatied on the source side.
1952      *
1953      * As such, we can avoid migrating the CAS-negotiated options if these
1954      * are the only options available on the current machine/platform.
1955      * Since these are the only options available for pseries-2.7 and
1956      * earlier, this allows us to maintain old->new/new->old migration
1957      * compatibility.
1958      *
1959      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1960      * via default pseries-2.8 machines and explicit command-line parameters.
1961      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1962      * of the actual CAS-negotiated values to continue working properly. For
1963      * example, availability of memory unplug depends on knowing whether
1964      * OV5_HP_EVT was negotiated via CAS.
1965      *
1966      * Thus, for any cases where the set of available CAS-negotiatable
1967      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1968      * include the CAS-negotiated options in the migration stream, unless
1969      * if they affect boot time behaviour only.
1970      */
1971     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1972     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1973     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1974 
1975     /* We need extra information if we have any bits outside the mask
1976      * defined above */
1977     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1978 
1979     spapr_ovec_cleanup(ov5_mask);
1980 
1981     return cas_needed;
1982 }
1983 
1984 static const VMStateDescription vmstate_spapr_ov5_cas = {
1985     .name = "spapr_option_vector_ov5_cas",
1986     .version_id = 1,
1987     .minimum_version_id = 1,
1988     .needed = spapr_ov5_cas_needed,
1989     .fields = (const VMStateField[]) {
1990         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1991                                  vmstate_spapr_ovec, SpaprOptionVector),
1992         VMSTATE_END_OF_LIST()
1993     },
1994 };
1995 
1996 static bool spapr_patb_entry_needed(void *opaque)
1997 {
1998     SpaprMachineState *spapr = opaque;
1999 
2000     return !!spapr->patb_entry;
2001 }
2002 
2003 static const VMStateDescription vmstate_spapr_patb_entry = {
2004     .name = "spapr_patb_entry",
2005     .version_id = 1,
2006     .minimum_version_id = 1,
2007     .needed = spapr_patb_entry_needed,
2008     .fields = (const VMStateField[]) {
2009         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2010         VMSTATE_END_OF_LIST()
2011     },
2012 };
2013 
2014 static bool spapr_irq_map_needed(void *opaque)
2015 {
2016     SpaprMachineState *spapr = opaque;
2017 
2018     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2019 }
2020 
2021 static const VMStateDescription vmstate_spapr_irq_map = {
2022     .name = "spapr_irq_map",
2023     .version_id = 1,
2024     .minimum_version_id = 1,
2025     .needed = spapr_irq_map_needed,
2026     .fields = (const VMStateField[]) {
2027         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2028         VMSTATE_END_OF_LIST()
2029     },
2030 };
2031 
2032 static bool spapr_dtb_needed(void *opaque)
2033 {
2034     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2035 
2036     return smc->update_dt_enabled;
2037 }
2038 
2039 static int spapr_dtb_pre_load(void *opaque)
2040 {
2041     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2042 
2043     g_free(spapr->fdt_blob);
2044     spapr->fdt_blob = NULL;
2045     spapr->fdt_size = 0;
2046 
2047     return 0;
2048 }
2049 
2050 static const VMStateDescription vmstate_spapr_dtb = {
2051     .name = "spapr_dtb",
2052     .version_id = 1,
2053     .minimum_version_id = 1,
2054     .needed = spapr_dtb_needed,
2055     .pre_load = spapr_dtb_pre_load,
2056     .fields = (const VMStateField[]) {
2057         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2058         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2059         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2060                                      fdt_size),
2061         VMSTATE_END_OF_LIST()
2062     },
2063 };
2064 
2065 static bool spapr_fwnmi_needed(void *opaque)
2066 {
2067     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2068 
2069     return spapr->fwnmi_machine_check_addr != -1;
2070 }
2071 
2072 static int spapr_fwnmi_pre_save(void *opaque)
2073 {
2074     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2075 
2076     /*
2077      * Check if machine check handling is in progress and print a
2078      * warning message.
2079      */
2080     if (spapr->fwnmi_machine_check_interlock != -1) {
2081         warn_report("A machine check is being handled during migration. The"
2082                 "handler may run and log hardware error on the destination");
2083     }
2084 
2085     return 0;
2086 }
2087 
2088 static const VMStateDescription vmstate_spapr_fwnmi = {
2089     .name = "spapr_fwnmi",
2090     .version_id = 1,
2091     .minimum_version_id = 1,
2092     .needed = spapr_fwnmi_needed,
2093     .pre_save = spapr_fwnmi_pre_save,
2094     .fields = (const VMStateField[]) {
2095         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2096         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2097         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2098         VMSTATE_END_OF_LIST()
2099     },
2100 };
2101 
2102 static const VMStateDescription vmstate_spapr = {
2103     .name = "spapr",
2104     .version_id = 3,
2105     .minimum_version_id = 1,
2106     .pre_load = spapr_pre_load,
2107     .post_load = spapr_post_load,
2108     .pre_save = spapr_pre_save,
2109     .fields = (const VMStateField[]) {
2110         /* used to be @next_irq */
2111         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2112 
2113         /* RTC offset */
2114         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2115 
2116         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2117         VMSTATE_END_OF_LIST()
2118     },
2119     .subsections = (const VMStateDescription * const []) {
2120         &vmstate_spapr_ov5_cas,
2121         &vmstate_spapr_patb_entry,
2122         &vmstate_spapr_pending_events,
2123         &vmstate_spapr_cap_htm,
2124         &vmstate_spapr_cap_vsx,
2125         &vmstate_spapr_cap_dfp,
2126         &vmstate_spapr_cap_cfpc,
2127         &vmstate_spapr_cap_sbbc,
2128         &vmstate_spapr_cap_ibs,
2129         &vmstate_spapr_cap_hpt_maxpagesize,
2130         &vmstate_spapr_irq_map,
2131         &vmstate_spapr_cap_nested_kvm_hv,
2132         &vmstate_spapr_dtb,
2133         &vmstate_spapr_cap_large_decr,
2134         &vmstate_spapr_cap_ccf_assist,
2135         &vmstate_spapr_cap_fwnmi,
2136         &vmstate_spapr_fwnmi,
2137         &vmstate_spapr_cap_rpt_invalidate,
2138         &vmstate_spapr_cap_ail_mode_3,
2139         &vmstate_spapr_cap_nested_papr,
2140         NULL
2141     }
2142 };
2143 
2144 static int htab_save_setup(QEMUFile *f, void *opaque, Error **errp)
2145 {
2146     SpaprMachineState *spapr = opaque;
2147 
2148     /* "Iteration" header */
2149     if (!spapr->htab_shift) {
2150         qemu_put_be32(f, -1);
2151     } else {
2152         qemu_put_be32(f, spapr->htab_shift);
2153     }
2154 
2155     if (spapr->htab) {
2156         spapr->htab_save_index = 0;
2157         spapr->htab_first_pass = true;
2158     } else {
2159         if (spapr->htab_shift) {
2160             assert(kvm_enabled());
2161         }
2162     }
2163 
2164 
2165     return 0;
2166 }
2167 
2168 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2169                             int chunkstart, int n_valid, int n_invalid)
2170 {
2171     qemu_put_be32(f, chunkstart);
2172     qemu_put_be16(f, n_valid);
2173     qemu_put_be16(f, n_invalid);
2174     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2175                     HASH_PTE_SIZE_64 * n_valid);
2176 }
2177 
2178 static void htab_save_end_marker(QEMUFile *f)
2179 {
2180     qemu_put_be32(f, 0);
2181     qemu_put_be16(f, 0);
2182     qemu_put_be16(f, 0);
2183 }
2184 
2185 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2186                                  int64_t max_ns)
2187 {
2188     bool has_timeout = max_ns != -1;
2189     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2190     int index = spapr->htab_save_index;
2191     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2192 
2193     assert(spapr->htab_first_pass);
2194 
2195     do {
2196         int chunkstart;
2197 
2198         /* Consume invalid HPTEs */
2199         while ((index < htabslots)
2200                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2201             CLEAN_HPTE(HPTE(spapr->htab, index));
2202             index++;
2203         }
2204 
2205         /* Consume valid HPTEs */
2206         chunkstart = index;
2207         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2208                && HPTE_VALID(HPTE(spapr->htab, index))) {
2209             CLEAN_HPTE(HPTE(spapr->htab, index));
2210             index++;
2211         }
2212 
2213         if (index > chunkstart) {
2214             int n_valid = index - chunkstart;
2215 
2216             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2217 
2218             if (has_timeout &&
2219                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2220                 break;
2221             }
2222         }
2223     } while ((index < htabslots) && !migration_rate_exceeded(f));
2224 
2225     if (index >= htabslots) {
2226         assert(index == htabslots);
2227         index = 0;
2228         spapr->htab_first_pass = false;
2229     }
2230     spapr->htab_save_index = index;
2231 }
2232 
2233 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2234                                 int64_t max_ns)
2235 {
2236     bool final = max_ns < 0;
2237     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2238     int examined = 0, sent = 0;
2239     int index = spapr->htab_save_index;
2240     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2241 
2242     assert(!spapr->htab_first_pass);
2243 
2244     do {
2245         int chunkstart, invalidstart;
2246 
2247         /* Consume non-dirty HPTEs */
2248         while ((index < htabslots)
2249                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2250             index++;
2251             examined++;
2252         }
2253 
2254         chunkstart = index;
2255         /* Consume valid dirty HPTEs */
2256         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2257                && HPTE_DIRTY(HPTE(spapr->htab, index))
2258                && HPTE_VALID(HPTE(spapr->htab, index))) {
2259             CLEAN_HPTE(HPTE(spapr->htab, index));
2260             index++;
2261             examined++;
2262         }
2263 
2264         invalidstart = index;
2265         /* Consume invalid dirty HPTEs */
2266         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2267                && HPTE_DIRTY(HPTE(spapr->htab, index))
2268                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2269             CLEAN_HPTE(HPTE(spapr->htab, index));
2270             index++;
2271             examined++;
2272         }
2273 
2274         if (index > chunkstart) {
2275             int n_valid = invalidstart - chunkstart;
2276             int n_invalid = index - invalidstart;
2277 
2278             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2279             sent += index - chunkstart;
2280 
2281             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2282                 break;
2283             }
2284         }
2285 
2286         if (examined >= htabslots) {
2287             break;
2288         }
2289 
2290         if (index >= htabslots) {
2291             assert(index == htabslots);
2292             index = 0;
2293         }
2294     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2295 
2296     if (index >= htabslots) {
2297         assert(index == htabslots);
2298         index = 0;
2299     }
2300 
2301     spapr->htab_save_index = index;
2302 
2303     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2304 }
2305 
2306 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2307 #define MAX_KVM_BUF_SIZE    2048
2308 
2309 static int htab_save_iterate(QEMUFile *f, void *opaque)
2310 {
2311     SpaprMachineState *spapr = opaque;
2312     int fd;
2313     int rc = 0;
2314 
2315     /* Iteration header */
2316     if (!spapr->htab_shift) {
2317         qemu_put_be32(f, -1);
2318         return 1;
2319     } else {
2320         qemu_put_be32(f, 0);
2321     }
2322 
2323     if (!spapr->htab) {
2324         assert(kvm_enabled());
2325 
2326         fd = get_htab_fd(spapr);
2327         if (fd < 0) {
2328             return fd;
2329         }
2330 
2331         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2332         if (rc < 0) {
2333             return rc;
2334         }
2335     } else  if (spapr->htab_first_pass) {
2336         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2337     } else {
2338         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2339     }
2340 
2341     htab_save_end_marker(f);
2342 
2343     return rc;
2344 }
2345 
2346 static int htab_save_complete(QEMUFile *f, void *opaque)
2347 {
2348     SpaprMachineState *spapr = opaque;
2349     int fd;
2350 
2351     /* Iteration header */
2352     if (!spapr->htab_shift) {
2353         qemu_put_be32(f, -1);
2354         return 0;
2355     } else {
2356         qemu_put_be32(f, 0);
2357     }
2358 
2359     if (!spapr->htab) {
2360         int rc;
2361 
2362         assert(kvm_enabled());
2363 
2364         fd = get_htab_fd(spapr);
2365         if (fd < 0) {
2366             return fd;
2367         }
2368 
2369         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2370         if (rc < 0) {
2371             return rc;
2372         }
2373     } else {
2374         if (spapr->htab_first_pass) {
2375             htab_save_first_pass(f, spapr, -1);
2376         }
2377         htab_save_later_pass(f, spapr, -1);
2378     }
2379 
2380     /* End marker */
2381     htab_save_end_marker(f);
2382 
2383     return 0;
2384 }
2385 
2386 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2387 {
2388     SpaprMachineState *spapr = opaque;
2389     uint32_t section_hdr;
2390     int fd = -1;
2391     Error *local_err = NULL;
2392 
2393     if (version_id < 1 || version_id > 1) {
2394         error_report("htab_load() bad version");
2395         return -EINVAL;
2396     }
2397 
2398     section_hdr = qemu_get_be32(f);
2399 
2400     if (section_hdr == -1) {
2401         spapr_free_hpt(spapr);
2402         return 0;
2403     }
2404 
2405     if (section_hdr) {
2406         int ret;
2407 
2408         /* First section gives the htab size */
2409         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2410         if (ret < 0) {
2411             error_report_err(local_err);
2412             return ret;
2413         }
2414         return 0;
2415     }
2416 
2417     if (!spapr->htab) {
2418         assert(kvm_enabled());
2419 
2420         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2421         if (fd < 0) {
2422             error_report_err(local_err);
2423             return fd;
2424         }
2425     }
2426 
2427     while (true) {
2428         uint32_t index;
2429         uint16_t n_valid, n_invalid;
2430 
2431         index = qemu_get_be32(f);
2432         n_valid = qemu_get_be16(f);
2433         n_invalid = qemu_get_be16(f);
2434 
2435         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2436             /* End of Stream */
2437             break;
2438         }
2439 
2440         if ((index + n_valid + n_invalid) >
2441             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2442             /* Bad index in stream */
2443             error_report(
2444                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2445                 index, n_valid, n_invalid, spapr->htab_shift);
2446             return -EINVAL;
2447         }
2448 
2449         if (spapr->htab) {
2450             if (n_valid) {
2451                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2452                                 HASH_PTE_SIZE_64 * n_valid);
2453             }
2454             if (n_invalid) {
2455                 memset(HPTE(spapr->htab, index + n_valid), 0,
2456                        HASH_PTE_SIZE_64 * n_invalid);
2457             }
2458         } else {
2459             int rc;
2460 
2461             assert(fd >= 0);
2462 
2463             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2464                                         &local_err);
2465             if (rc < 0) {
2466                 error_report_err(local_err);
2467                 return rc;
2468             }
2469         }
2470     }
2471 
2472     if (!spapr->htab) {
2473         assert(fd >= 0);
2474         close(fd);
2475     }
2476 
2477     return 0;
2478 }
2479 
2480 static void htab_save_cleanup(void *opaque)
2481 {
2482     SpaprMachineState *spapr = opaque;
2483 
2484     close_htab_fd(spapr);
2485 }
2486 
2487 static SaveVMHandlers savevm_htab_handlers = {
2488     .save_setup = htab_save_setup,
2489     .save_live_iterate = htab_save_iterate,
2490     .save_live_complete_precopy = htab_save_complete,
2491     .save_cleanup = htab_save_cleanup,
2492     .load_state = htab_load,
2493 };
2494 
2495 static void spapr_boot_set(void *opaque, const char *boot_device,
2496                            Error **errp)
2497 {
2498     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2499 
2500     g_free(spapr->boot_device);
2501     spapr->boot_device = g_strdup(boot_device);
2502 }
2503 
2504 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2505 {
2506     MachineState *machine = MACHINE(spapr);
2507     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2508     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2509     int i;
2510 
2511     g_assert(!nr_lmbs || machine->device_memory);
2512     for (i = 0; i < nr_lmbs; i++) {
2513         uint64_t addr;
2514 
2515         addr = i * lmb_size + machine->device_memory->base;
2516         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2517                                addr / lmb_size);
2518     }
2519 }
2520 
2521 /*
2522  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2523  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2524  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2525  */
2526 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2527 {
2528     int i;
2529 
2530     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2531         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2532                    " is not aligned to %" PRIu64 " MiB",
2533                    machine->ram_size,
2534                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2535         return;
2536     }
2537 
2538     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2539         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2540                    " is not aligned to %" PRIu64 " MiB",
2541                    machine->ram_size,
2542                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2543         return;
2544     }
2545 
2546     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2547         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2548             error_setg(errp,
2549                        "Node %d memory size 0x%" PRIx64
2550                        " is not aligned to %" PRIu64 " MiB",
2551                        i, machine->numa_state->nodes[i].node_mem,
2552                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2553             return;
2554         }
2555     }
2556 }
2557 
2558 /* find cpu slot in machine->possible_cpus by core_id */
2559 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2560 {
2561     int index = id / ms->smp.threads;
2562 
2563     if (index >= ms->possible_cpus->len) {
2564         return NULL;
2565     }
2566     if (idx) {
2567         *idx = index;
2568     }
2569     return &ms->possible_cpus->cpus[index];
2570 }
2571 
2572 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2573 {
2574     MachineState *ms = MACHINE(spapr);
2575     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2576     Error *local_err = NULL;
2577     bool vsmt_user = !!spapr->vsmt;
2578     int kvm_smt = kvmppc_smt_threads();
2579     int ret;
2580     unsigned int smp_threads = ms->smp.threads;
2581 
2582     if (tcg_enabled()) {
2583         if (smp_threads > 1 &&
2584             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2585                                    spapr->max_compat_pvr)) {
2586             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2587             return;
2588         }
2589 
2590         if (smp_threads > 8) {
2591             error_setg(errp, "TCG cannot support more than 8 threads/core "
2592                        "on a pseries machine");
2593             return;
2594         }
2595     }
2596     if (!is_power_of_2(smp_threads)) {
2597         error_setg(errp, "Cannot support %d threads/core on a pseries "
2598                    "machine because it must be a power of 2", smp_threads);
2599         return;
2600     }
2601 
2602     /* Determine the VSMT mode to use: */
2603     if (vsmt_user) {
2604         if (spapr->vsmt < smp_threads) {
2605             error_setg(errp, "Cannot support VSMT mode %d"
2606                        " because it must be >= threads/core (%d)",
2607                        spapr->vsmt, smp_threads);
2608             return;
2609         }
2610         /* In this case, spapr->vsmt has been set by the command line */
2611     } else if (!smc->smp_threads_vsmt) {
2612         /*
2613          * Default VSMT value is tricky, because we need it to be as
2614          * consistent as possible (for migration), but this requires
2615          * changing it for at least some existing cases.  We pick 8 as
2616          * the value that we'd get with KVM on POWER8, the
2617          * overwhelmingly common case in production systems.
2618          */
2619         spapr->vsmt = MAX(8, smp_threads);
2620     } else {
2621         spapr->vsmt = smp_threads;
2622     }
2623 
2624     /* KVM: If necessary, set the SMT mode: */
2625     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2626         ret = kvmppc_set_smt_threads(spapr->vsmt);
2627         if (ret) {
2628             /* Looks like KVM isn't able to change VSMT mode */
2629             error_setg(&local_err,
2630                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2631                        spapr->vsmt, ret);
2632             /* We can live with that if the default one is big enough
2633              * for the number of threads, and a submultiple of the one
2634              * we want.  In this case we'll waste some vcpu ids, but
2635              * behaviour will be correct */
2636             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2637                 warn_report_err(local_err);
2638             } else {
2639                 if (!vsmt_user) {
2640                     error_append_hint(&local_err,
2641                                       "On PPC, a VM with %d threads/core"
2642                                       " on a host with %d threads/core"
2643                                       " requires the use of VSMT mode %d.\n",
2644                                       smp_threads, kvm_smt, spapr->vsmt);
2645                 }
2646                 kvmppc_error_append_smt_possible_hint(&local_err);
2647                 error_propagate(errp, local_err);
2648             }
2649         }
2650     }
2651     /* else TCG: nothing to do currently */
2652 }
2653 
2654 static void spapr_init_cpus(SpaprMachineState *spapr)
2655 {
2656     MachineState *machine = MACHINE(spapr);
2657     MachineClass *mc = MACHINE_GET_CLASS(machine);
2658     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2659     const CPUArchIdList *possible_cpus;
2660     unsigned int smp_cpus = machine->smp.cpus;
2661     unsigned int smp_threads = machine->smp.threads;
2662     unsigned int max_cpus = machine->smp.max_cpus;
2663     int boot_cores_nr = smp_cpus / smp_threads;
2664     int i;
2665 
2666     possible_cpus = mc->possible_cpu_arch_ids(machine);
2667     if (mc->has_hotpluggable_cpus) {
2668         if (smp_cpus % smp_threads) {
2669             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2670                          smp_cpus, smp_threads);
2671             exit(1);
2672         }
2673         if (max_cpus % smp_threads) {
2674             error_report("max_cpus (%u) must be multiple of threads (%u)",
2675                          max_cpus, smp_threads);
2676             exit(1);
2677         }
2678     } else {
2679         if (max_cpus != smp_cpus) {
2680             error_report("This machine version does not support CPU hotplug");
2681             exit(1);
2682         }
2683         boot_cores_nr = possible_cpus->len;
2684     }
2685 
2686     for (i = 0; i < possible_cpus->len; i++) {
2687         int core_id = i * smp_threads;
2688 
2689         if (mc->has_hotpluggable_cpus) {
2690             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2691                                    spapr_vcpu_id(spapr, core_id));
2692         }
2693 
2694         if (i < boot_cores_nr) {
2695             Object *core  = object_new(type);
2696             int nr_threads = smp_threads;
2697 
2698             /* Handle the partially filled core for older machine types */
2699             if ((i + 1) * smp_threads >= smp_cpus) {
2700                 nr_threads = smp_cpus - i * smp_threads;
2701             }
2702 
2703             object_property_set_int(core, "nr-threads", nr_threads,
2704                                     &error_fatal);
2705             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2706                                     &error_fatal);
2707             qdev_realize(DEVICE(core), NULL, &error_fatal);
2708 
2709             object_unref(core);
2710         }
2711     }
2712 }
2713 
2714 static PCIHostState *spapr_create_default_phb(void)
2715 {
2716     DeviceState *dev;
2717 
2718     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2719     qdev_prop_set_uint32(dev, "index", 0);
2720     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2721 
2722     return PCI_HOST_BRIDGE(dev);
2723 }
2724 
2725 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2726 {
2727     MachineState *machine = MACHINE(spapr);
2728     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2729     hwaddr rma_size = machine->ram_size;
2730     hwaddr node0_size = spapr_node0_size(machine);
2731 
2732     /* RMA has to fit in the first NUMA node */
2733     rma_size = MIN(rma_size, node0_size);
2734 
2735     /*
2736      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2737      * never exceed that
2738      */
2739     rma_size = MIN(rma_size, 1 * TiB);
2740 
2741     /*
2742      * Clamp the RMA size based on machine type.  This is for
2743      * migration compatibility with older qemu versions, which limited
2744      * the RMA size for complicated and mostly bad reasons.
2745      */
2746     if (smc->rma_limit) {
2747         rma_size = MIN(rma_size, smc->rma_limit);
2748     }
2749 
2750     if (rma_size < MIN_RMA_SLOF) {
2751         error_setg(errp,
2752                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2753                    "ldMiB guest RMA (Real Mode Area memory)",
2754                    MIN_RMA_SLOF / MiB);
2755         return 0;
2756     }
2757 
2758     return rma_size;
2759 }
2760 
2761 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2762 {
2763     MachineState *machine = MACHINE(spapr);
2764     int i;
2765 
2766     for (i = 0; i < machine->ram_slots; i++) {
2767         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2768     }
2769 }
2770 
2771 /* pSeries LPAR / sPAPR hardware init */
2772 static void spapr_machine_init(MachineState *machine)
2773 {
2774     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2775     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2776     MachineClass *mc = MACHINE_GET_CLASS(machine);
2777     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2778     const char *bios_name = machine->firmware ?: bios_default;
2779     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2780     const char *kernel_filename = machine->kernel_filename;
2781     const char *initrd_filename = machine->initrd_filename;
2782     PCIHostState *phb;
2783     bool has_vga;
2784     int i;
2785     MemoryRegion *sysmem = get_system_memory();
2786     long load_limit, fw_size;
2787     Error *resize_hpt_err = NULL;
2788     NICInfo *nd;
2789 
2790     if (!filename) {
2791         error_report("Could not find LPAR firmware '%s'", bios_name);
2792         exit(1);
2793     }
2794     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2795     if (fw_size <= 0) {
2796         error_report("Could not load LPAR firmware '%s'", filename);
2797         exit(1);
2798     }
2799 
2800     /*
2801      * if Secure VM (PEF) support is configured, then initialize it
2802      */
2803     if (machine->cgs) {
2804         confidential_guest_kvm_init(machine->cgs, &error_fatal);
2805     }
2806 
2807     msi_nonbroken = true;
2808 
2809     QLIST_INIT(&spapr->phbs);
2810     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2811 
2812     /* Determine capabilities to run with */
2813     spapr_caps_init(spapr);
2814 
2815     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2816     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2817         /*
2818          * If the user explicitly requested a mode we should either
2819          * supply it, or fail completely (which we do below).  But if
2820          * it's not set explicitly, we reset our mode to something
2821          * that works
2822          */
2823         if (resize_hpt_err) {
2824             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2825             error_free(resize_hpt_err);
2826             resize_hpt_err = NULL;
2827         } else {
2828             spapr->resize_hpt = smc->resize_hpt_default;
2829         }
2830     }
2831 
2832     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2833 
2834     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2835         /*
2836          * User requested HPT resize, but this host can't supply it.  Bail out
2837          */
2838         error_report_err(resize_hpt_err);
2839         exit(1);
2840     }
2841     error_free(resize_hpt_err);
2842 
2843     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2844 
2845     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2846     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2847 
2848     /*
2849      * VSMT must be set in order to be able to compute VCPU ids, ie to
2850      * call spapr_max_server_number() or spapr_vcpu_id().
2851      */
2852     spapr_set_vsmt_mode(spapr, &error_fatal);
2853 
2854     /* Set up Interrupt Controller before we create the VCPUs */
2855     spapr_irq_init(spapr, &error_fatal);
2856 
2857     /* Set up containers for ibm,client-architecture-support negotiated options
2858      */
2859     spapr->ov5 = spapr_ovec_new();
2860     spapr->ov5_cas = spapr_ovec_new();
2861 
2862     spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2863     spapr_validate_node_memory(machine, &error_fatal);
2864 
2865     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2866 
2867     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2868     if (!smc->pre_6_2_numa_affinity) {
2869         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2870     }
2871 
2872     /* advertise support for dedicated HP event source to guests */
2873     if (spapr->use_hotplug_event_source) {
2874         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2875     }
2876 
2877     /* advertise support for HPT resizing */
2878     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2879         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2880     }
2881 
2882     /* advertise support for ibm,dyamic-memory-v2 */
2883     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2884 
2885     /* advertise XIVE on POWER9 machines */
2886     if (spapr->irq->xive) {
2887         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2888     }
2889 
2890     /* init CPUs */
2891     spapr_init_cpus(spapr);
2892 
2893     /* Init numa_assoc_array */
2894     spapr_numa_associativity_init(spapr, machine);
2895 
2896     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2897         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2898                               spapr->max_compat_pvr)) {
2899         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2900         /* KVM and TCG always allow GTSE with radix... */
2901         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2902     }
2903     /* ... but not with hash (currently). */
2904 
2905     if (kvm_enabled()) {
2906         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2907         kvmppc_enable_logical_ci_hcalls();
2908         kvmppc_enable_set_mode_hcall();
2909 
2910         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2911         kvmppc_enable_clear_ref_mod_hcalls();
2912 
2913         /* Enable H_PAGE_INIT */
2914         kvmppc_enable_h_page_init();
2915     }
2916 
2917     /* map RAM */
2918     memory_region_add_subregion(sysmem, 0, machine->ram);
2919 
2920     /* initialize hotplug memory address space */
2921     if (machine->ram_size < machine->maxram_size) {
2922         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2923         hwaddr device_mem_base;
2924 
2925         /*
2926          * Limit the number of hotpluggable memory slots to half the number
2927          * slots that KVM supports, leaving the other half for PCI and other
2928          * devices. However ensure that number of slots doesn't drop below 32.
2929          */
2930         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2931                            SPAPR_MAX_RAM_SLOTS;
2932 
2933         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2934             max_memslots = SPAPR_MAX_RAM_SLOTS;
2935         }
2936         if (machine->ram_slots > max_memslots) {
2937             error_report("Specified number of memory slots %"
2938                          PRIu64" exceeds max supported %d",
2939                          machine->ram_slots, max_memslots);
2940             exit(1);
2941         }
2942 
2943         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
2944         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
2945     }
2946 
2947     spapr_create_lmb_dr_connectors(spapr);
2948 
2949     if (mc->nvdimm_supported) {
2950         spapr_create_nvdimm_dr_connectors(spapr);
2951     }
2952 
2953     /* Set up RTAS event infrastructure */
2954     spapr_events_init(spapr);
2955 
2956     /* Set up the RTC RTAS interfaces */
2957     spapr_rtc_create(spapr);
2958 
2959     /* Set up VIO bus */
2960     spapr->vio_bus = spapr_vio_bus_init();
2961 
2962     for (i = 0; serial_hd(i); i++) {
2963         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2964     }
2965 
2966     /* We always have at least the nvram device on VIO */
2967     spapr_create_nvram(spapr);
2968 
2969     /*
2970      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2971      * connectors (described in root DT node's "ibm,drc-types" property)
2972      * are pre-initialized here. additional child connectors (such as
2973      * connectors for a PHBs PCI slots) are added as needed during their
2974      * parent's realization.
2975      */
2976     if (smc->dr_phb_enabled) {
2977         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2978             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2979         }
2980     }
2981 
2982     /* Set up PCI */
2983     spapr_pci_rtas_init();
2984 
2985     phb = spapr_create_default_phb();
2986 
2987     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
2988         spapr_vlan_create(spapr->vio_bus, nd);
2989     }
2990 
2991     pci_init_nic_devices(phb->bus, NULL);
2992 
2993     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2994         spapr_vscsi_create(spapr->vio_bus);
2995     }
2996 
2997     /* Graphics */
2998     has_vga = spapr_vga_init(phb->bus, &error_fatal);
2999     if (has_vga) {
3000         spapr->want_stdout_path = !machine->enable_graphics;
3001         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3002     } else {
3003         spapr->want_stdout_path = true;
3004     }
3005 
3006     if (machine->usb) {
3007         pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3008 
3009         if (has_vga) {
3010             USBBus *usb_bus;
3011 
3012             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3013                                                               &error_abort));
3014             usb_create_simple(usb_bus, "usb-kbd");
3015             usb_create_simple(usb_bus, "usb-mouse");
3016         }
3017     }
3018 
3019     if (kernel_filename) {
3020         uint64_t loaded_addr = 0;
3021 
3022         spapr->kernel_size = load_elf(kernel_filename, NULL,
3023                                       translate_kernel_address, spapr,
3024                                       NULL, &loaded_addr, NULL, NULL,
3025                                       ELFDATA2MSB, PPC_ELF_MACHINE, 0, 0);
3026         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3027             spapr->kernel_size = load_elf(kernel_filename, NULL,
3028                                           translate_kernel_address, spapr,
3029                                           NULL, &loaded_addr, NULL, NULL,
3030                                           ELFDATA2LSB, PPC_ELF_MACHINE, 0, 0);
3031             spapr->kernel_le = spapr->kernel_size > 0;
3032         }
3033         if (spapr->kernel_size < 0) {
3034             error_report("error loading %s: %s", kernel_filename,
3035                          load_elf_strerror(spapr->kernel_size));
3036             exit(1);
3037         }
3038 
3039         if (spapr->kernel_addr != loaded_addr) {
3040             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3041                         " to 0x%"PRIx64,
3042                         spapr->kernel_addr, loaded_addr);
3043             spapr->kernel_addr = loaded_addr;
3044         }
3045 
3046         /* load initrd */
3047         if (initrd_filename) {
3048             /* Try to locate the initrd in the gap between the kernel
3049              * and the firmware. Add a bit of space just in case
3050              */
3051             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3052                                   + 0x1ffff) & ~0xffff;
3053             spapr->initrd_size = load_image_targphys(initrd_filename,
3054                                                      spapr->initrd_base,
3055                                                      load_limit
3056                                                      - spapr->initrd_base);
3057             if (spapr->initrd_size < 0) {
3058                 error_report("could not load initial ram disk '%s'",
3059                              initrd_filename);
3060                 exit(1);
3061             }
3062         }
3063     }
3064 
3065     /* FIXME: Should register things through the MachineState's qdev
3066      * interface, this is a legacy from the sPAPREnvironment structure
3067      * which predated MachineState but had a similar function */
3068     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3069     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3070                          &savevm_htab_handlers, spapr);
3071 
3072     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3073 
3074     qemu_register_boot_set(spapr_boot_set, spapr);
3075 
3076     /*
3077      * Nothing needs to be done to resume a suspended guest because
3078      * suspending does not change the machine state, so no need for
3079      * a ->wakeup method.
3080      */
3081     qemu_register_wakeup_support();
3082 
3083     if (kvm_enabled()) {
3084         /* to stop and start vmclock */
3085         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3086                                          &spapr->tb);
3087 
3088         kvmppc_spapr_enable_inkernel_multitce();
3089     }
3090 
3091     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3092     if (spapr->vof) {
3093         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3094         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3095     }
3096 
3097     spapr_watchdog_init(spapr);
3098 }
3099 
3100 #define DEFAULT_KVM_TYPE "auto"
3101 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3102 {
3103     /*
3104      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3105      * accommodate the 'HV' and 'PV' formats that exists in the
3106      * wild. The 'auto' mode is being introduced already as
3107      * lower-case, thus we don't need to bother checking for
3108      * "AUTO".
3109      */
3110     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3111         return 0;
3112     }
3113 
3114     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3115         return 1;
3116     }
3117 
3118     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3119         return 2;
3120     }
3121 
3122     error_report("Unknown kvm-type specified '%s'", vm_type);
3123     return -1;
3124 }
3125 
3126 /*
3127  * Implementation of an interface to adjust firmware path
3128  * for the bootindex property handling.
3129  */
3130 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3131                                    DeviceState *dev)
3132 {
3133 #define CAST(type, obj, name) \
3134     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3135     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3136     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3137     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3138     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3139 
3140     if (d && bus) {
3141         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3142         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3143         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3144 
3145         if (spapr) {
3146             /*
3147              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3148              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3149              * 0x8000 | (target << 8) | (bus << 5) | lun
3150              * (see the "Logical unit addressing format" table in SAM5)
3151              */
3152             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3153             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3154                                    (uint64_t)id << 48);
3155         } else if (virtio) {
3156             /*
3157              * We use SRP luns of the form 01000000 | (target << 8) | lun
3158              * in the top 32 bits of the 64-bit LUN
3159              * Note: the quote above is from SLOF and it is wrong,
3160              * the actual binding is:
3161              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3162              */
3163             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3164             if (d->lun >= 256) {
3165                 /* Use the LUN "flat space addressing method" */
3166                 id |= 0x4000;
3167             }
3168             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3169                                    (uint64_t)id << 32);
3170         } else if (usb) {
3171             /*
3172              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3173              * in the top 32 bits of the 64-bit LUN
3174              */
3175             unsigned usb_port = atoi(usb->port->path);
3176             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3177             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3178                                    (uint64_t)id << 32);
3179         }
3180     }
3181 
3182     /*
3183      * SLOF probes the USB devices, and if it recognizes that the device is a
3184      * storage device, it changes its name to "storage" instead of "usb-host",
3185      * and additionally adds a child node for the SCSI LUN, so the correct
3186      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3187      */
3188     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3189         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3190         if (usb_device_is_scsi_storage(usbdev)) {
3191             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3192         }
3193     }
3194 
3195     if (phb) {
3196         /* Replace "pci" with "pci@800000020000000" */
3197         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3198     }
3199 
3200     if (vsc) {
3201         /* Same logic as virtio above */
3202         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3203         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3204     }
3205 
3206     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3207         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3208         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3209         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3210     }
3211 
3212     if (pcidev) {
3213         return spapr_pci_fw_dev_name(pcidev);
3214     }
3215 
3216     return NULL;
3217 }
3218 
3219 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3220 {
3221     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3222 
3223     return g_strdup(spapr->kvm_type);
3224 }
3225 
3226 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3227 {
3228     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3229 
3230     g_free(spapr->kvm_type);
3231     spapr->kvm_type = g_strdup(value);
3232 }
3233 
3234 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3235 {
3236     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3237 
3238     return spapr->use_hotplug_event_source;
3239 }
3240 
3241 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3242                                             Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     spapr->use_hotplug_event_source = value;
3247 }
3248 
3249 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3250 {
3251     return true;
3252 }
3253 
3254 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3255 {
3256     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3257 
3258     switch (spapr->resize_hpt) {
3259     case SPAPR_RESIZE_HPT_DEFAULT:
3260         return g_strdup("default");
3261     case SPAPR_RESIZE_HPT_DISABLED:
3262         return g_strdup("disabled");
3263     case SPAPR_RESIZE_HPT_ENABLED:
3264         return g_strdup("enabled");
3265     case SPAPR_RESIZE_HPT_REQUIRED:
3266         return g_strdup("required");
3267     }
3268     g_assert_not_reached();
3269 }
3270 
3271 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3272 {
3273     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3274 
3275     if (strcmp(value, "default") == 0) {
3276         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3277     } else if (strcmp(value, "disabled") == 0) {
3278         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3279     } else if (strcmp(value, "enabled") == 0) {
3280         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3281     } else if (strcmp(value, "required") == 0) {
3282         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3283     } else {
3284         error_setg(errp, "Bad value for \"resize-hpt\" property");
3285     }
3286 }
3287 
3288 static bool spapr_get_vof(Object *obj, Error **errp)
3289 {
3290     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3291 
3292     return spapr->vof != NULL;
3293 }
3294 
3295 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 
3299     if (spapr->vof) {
3300         vof_cleanup(spapr->vof);
3301         g_free(spapr->vof);
3302         spapr->vof = NULL;
3303     }
3304     if (!value) {
3305         return;
3306     }
3307     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3308 }
3309 
3310 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3311 {
3312     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3313 
3314     if (spapr->irq == &spapr_irq_xics_legacy) {
3315         return g_strdup("legacy");
3316     } else if (spapr->irq == &spapr_irq_xics) {
3317         return g_strdup("xics");
3318     } else if (spapr->irq == &spapr_irq_xive) {
3319         return g_strdup("xive");
3320     } else if (spapr->irq == &spapr_irq_dual) {
3321         return g_strdup("dual");
3322     }
3323     g_assert_not_reached();
3324 }
3325 
3326 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3327 {
3328     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3329 
3330     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3331         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3332         return;
3333     }
3334 
3335     /* The legacy IRQ backend can not be set */
3336     if (strcmp(value, "xics") == 0) {
3337         spapr->irq = &spapr_irq_xics;
3338     } else if (strcmp(value, "xive") == 0) {
3339         spapr->irq = &spapr_irq_xive;
3340     } else if (strcmp(value, "dual") == 0) {
3341         spapr->irq = &spapr_irq_dual;
3342     } else {
3343         error_setg(errp, "Bad value for \"ic-mode\" property");
3344     }
3345 }
3346 
3347 static char *spapr_get_host_model(Object *obj, Error **errp)
3348 {
3349     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3350 
3351     return g_strdup(spapr->host_model);
3352 }
3353 
3354 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3355 {
3356     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3357 
3358     g_free(spapr->host_model);
3359     spapr->host_model = g_strdup(value);
3360 }
3361 
3362 static char *spapr_get_host_serial(Object *obj, Error **errp)
3363 {
3364     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3365 
3366     return g_strdup(spapr->host_serial);
3367 }
3368 
3369 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3370 {
3371     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3372 
3373     g_free(spapr->host_serial);
3374     spapr->host_serial = g_strdup(value);
3375 }
3376 
3377 static void spapr_instance_init(Object *obj)
3378 {
3379     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3380     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3381     MachineState *ms = MACHINE(spapr);
3382     MachineClass *mc = MACHINE_GET_CLASS(ms);
3383 
3384     /*
3385      * NVDIMM support went live in 5.1 without considering that, in
3386      * other archs, the user needs to enable NVDIMM support with the
3387      * 'nvdimm' machine option and the default behavior is NVDIMM
3388      * support disabled. It is too late to roll back to the standard
3389      * behavior without breaking 5.1 guests.
3390      */
3391     if (mc->nvdimm_supported) {
3392         ms->nvdimms_state->is_enabled = true;
3393     }
3394 
3395     spapr->htab_fd = -1;
3396     spapr->use_hotplug_event_source = true;
3397     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3398     object_property_add_str(obj, "kvm-type",
3399                             spapr_get_kvm_type, spapr_set_kvm_type);
3400     object_property_set_description(obj, "kvm-type",
3401                                     "Specifies the KVM virtualization mode (auto,"
3402                                     " hv, pr). Defaults to 'auto'. This mode will use"
3403                                     " any available KVM module loaded in the host,"
3404                                     " where kvm_hv takes precedence if both kvm_hv and"
3405                                     " kvm_pr are loaded.");
3406     object_property_add_bool(obj, "modern-hotplug-events",
3407                             spapr_get_modern_hotplug_events,
3408                             spapr_set_modern_hotplug_events);
3409     object_property_set_description(obj, "modern-hotplug-events",
3410                                     "Use dedicated hotplug event mechanism in"
3411                                     " place of standard EPOW events when possible"
3412                                     " (required for memory hot-unplug support)");
3413     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3414                             "Maximum permitted CPU compatibility mode");
3415 
3416     object_property_add_str(obj, "resize-hpt",
3417                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3418     object_property_set_description(obj, "resize-hpt",
3419                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3420     object_property_add_uint32_ptr(obj, "vsmt",
3421                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3422     object_property_set_description(obj, "vsmt",
3423                                     "Virtual SMT: KVM behaves as if this were"
3424                                     " the host's SMT mode");
3425 
3426     object_property_add_bool(obj, "vfio-no-msix-emulation",
3427                              spapr_get_msix_emulation, NULL);
3428 
3429     object_property_add_uint64_ptr(obj, "kernel-addr",
3430                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3431     object_property_set_description(obj, "kernel-addr",
3432                                     stringify(KERNEL_LOAD_ADDR)
3433                                     " for -kernel is the default");
3434     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3435 
3436     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3437     object_property_set_description(obj, "x-vof",
3438                                     "Enable Virtual Open Firmware (experimental)");
3439 
3440     /* The machine class defines the default interrupt controller mode */
3441     spapr->irq = smc->irq;
3442     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3443                             spapr_set_ic_mode);
3444     object_property_set_description(obj, "ic-mode",
3445                  "Specifies the interrupt controller mode (xics, xive, dual)");
3446 
3447     object_property_add_str(obj, "host-model",
3448         spapr_get_host_model, spapr_set_host_model);
3449     object_property_set_description(obj, "host-model",
3450         "Host model to advertise in guest device tree");
3451     object_property_add_str(obj, "host-serial",
3452         spapr_get_host_serial, spapr_set_host_serial);
3453     object_property_set_description(obj, "host-serial",
3454         "Host serial number to advertise in guest device tree");
3455 }
3456 
3457 static void spapr_machine_finalizefn(Object *obj)
3458 {
3459     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3460 
3461     g_free(spapr->kvm_type);
3462 }
3463 
3464 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3465 {
3466     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3467     CPUPPCState *env = cpu_env(cs);
3468 
3469     cpu_synchronize_state(cs);
3470     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3471     if (spapr->fwnmi_system_reset_addr != -1) {
3472         uint64_t rtas_addr, addr;
3473 
3474         /* get rtas addr from fdt */
3475         rtas_addr = spapr_get_rtas_addr();
3476         if (!rtas_addr) {
3477             qemu_system_guest_panicked(NULL);
3478             return;
3479         }
3480 
3481         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3482         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3483         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3484         env->gpr[3] = addr;
3485     }
3486     ppc_cpu_do_system_reset(cs);
3487     if (spapr->fwnmi_system_reset_addr != -1) {
3488         env->nip = spapr->fwnmi_system_reset_addr;
3489     }
3490 }
3491 
3492 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3493 {
3494     CPUState *cs;
3495 
3496     CPU_FOREACH(cs) {
3497         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3498     }
3499 }
3500 
3501 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3502                           void *fdt, int *fdt_start_offset, Error **errp)
3503 {
3504     uint64_t addr;
3505     uint32_t node;
3506 
3507     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3508     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3509                                     &error_abort);
3510     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3511                                              SPAPR_MEMORY_BLOCK_SIZE);
3512     return 0;
3513 }
3514 
3515 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3516                            bool dedicated_hp_event_source)
3517 {
3518     SpaprDrc *drc;
3519     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3520     int i;
3521     uint64_t addr = addr_start;
3522     bool hotplugged = spapr_drc_hotplugged(dev);
3523 
3524     for (i = 0; i < nr_lmbs; i++) {
3525         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3526                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3527         g_assert(drc);
3528 
3529         /*
3530          * memory_device_get_free_addr() provided a range of free addresses
3531          * that doesn't overlap with any existing mapping at pre-plug. The
3532          * corresponding LMB DRCs are thus assumed to be all attachable.
3533          */
3534         spapr_drc_attach(drc, dev);
3535         if (!hotplugged) {
3536             spapr_drc_reset(drc);
3537         }
3538         addr += SPAPR_MEMORY_BLOCK_SIZE;
3539     }
3540     /* send hotplug notification to the
3541      * guest only in case of hotplugged memory
3542      */
3543     if (hotplugged) {
3544         if (dedicated_hp_event_source) {
3545             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3546                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3547             g_assert(drc);
3548             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3549                                                    nr_lmbs,
3550                                                    spapr_drc_index(drc));
3551         } else {
3552             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3553                                            nr_lmbs);
3554         }
3555     }
3556 }
3557 
3558 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3559 {
3560     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3561     PCDIMMDevice *dimm = PC_DIMM(dev);
3562     uint64_t size, addr;
3563     int64_t slot;
3564     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3565 
3566     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3567 
3568     pc_dimm_plug(dimm, MACHINE(ms));
3569 
3570     if (!is_nvdimm) {
3571         addr = object_property_get_uint(OBJECT(dimm),
3572                                         PC_DIMM_ADDR_PROP, &error_abort);
3573         spapr_add_lmbs(dev, addr, size,
3574                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3575     } else {
3576         slot = object_property_get_int(OBJECT(dimm),
3577                                        PC_DIMM_SLOT_PROP, &error_abort);
3578         /* We should have valid slot number at this point */
3579         g_assert(slot >= 0);
3580         spapr_add_nvdimm(dev, slot);
3581     }
3582 }
3583 
3584 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3585                                   Error **errp)
3586 {
3587     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3588     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3589     PCDIMMDevice *dimm = PC_DIMM(dev);
3590     Error *local_err = NULL;
3591     uint64_t size;
3592     Object *memdev;
3593     hwaddr pagesize;
3594 
3595     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3596     if (local_err) {
3597         error_propagate(errp, local_err);
3598         return;
3599     }
3600 
3601     if (is_nvdimm) {
3602         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3603             return;
3604         }
3605     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3606         error_setg(errp, "Hotplugged memory size must be a multiple of "
3607                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3608         return;
3609     }
3610 
3611     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3612                                       &error_abort);
3613     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3614     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3615         return;
3616     }
3617 
3618     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), errp);
3619 }
3620 
3621 struct SpaprDimmState {
3622     PCDIMMDevice *dimm;
3623     uint32_t nr_lmbs;
3624     QTAILQ_ENTRY(SpaprDimmState) next;
3625 };
3626 
3627 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3628                                                        PCDIMMDevice *dimm)
3629 {
3630     SpaprDimmState *dimm_state = NULL;
3631 
3632     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3633         if (dimm_state->dimm == dimm) {
3634             break;
3635         }
3636     }
3637     return dimm_state;
3638 }
3639 
3640 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3641                                                       uint32_t nr_lmbs,
3642                                                       PCDIMMDevice *dimm)
3643 {
3644     SpaprDimmState *ds = NULL;
3645 
3646     /*
3647      * If this request is for a DIMM whose removal had failed earlier
3648      * (due to guest's refusal to remove the LMBs), we would have this
3649      * dimm already in the pending_dimm_unplugs list. In that
3650      * case don't add again.
3651      */
3652     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3653     if (!ds) {
3654         ds = g_new0(SpaprDimmState, 1);
3655         ds->nr_lmbs = nr_lmbs;
3656         ds->dimm = dimm;
3657         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3658     }
3659     return ds;
3660 }
3661 
3662 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3663                                               SpaprDimmState *dimm_state)
3664 {
3665     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3666     g_free(dimm_state);
3667 }
3668 
3669 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3670                                                         PCDIMMDevice *dimm)
3671 {
3672     SpaprDrc *drc;
3673     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3674                                                   &error_abort);
3675     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3676     uint32_t avail_lmbs = 0;
3677     uint64_t addr_start, addr;
3678     int i;
3679 
3680     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3681                                           &error_abort);
3682 
3683     addr = addr_start;
3684     for (i = 0; i < nr_lmbs; i++) {
3685         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3686                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3687         g_assert(drc);
3688         if (drc->dev) {
3689             avail_lmbs++;
3690         }
3691         addr += SPAPR_MEMORY_BLOCK_SIZE;
3692     }
3693 
3694     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3695 }
3696 
3697 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3698 {
3699     SpaprDimmState *ds;
3700     PCDIMMDevice *dimm;
3701     SpaprDrc *drc;
3702     uint32_t nr_lmbs;
3703     uint64_t size, addr_start, addr;
3704     int i;
3705 
3706     if (!dev) {
3707         return;
3708     }
3709 
3710     dimm = PC_DIMM(dev);
3711     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3712 
3713     /*
3714      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3715      * unplug state, but one of its DRC is marked as unplug_requested.
3716      * This is bad and weird enough to g_assert() out.
3717      */
3718     g_assert(ds);
3719 
3720     spapr_pending_dimm_unplugs_remove(spapr, ds);
3721 
3722     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3723     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3724 
3725     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3726                                           &error_abort);
3727 
3728     addr = addr_start;
3729     for (i = 0; i < nr_lmbs; i++) {
3730         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3731                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3732         g_assert(drc);
3733 
3734         drc->unplug_requested = false;
3735         addr += SPAPR_MEMORY_BLOCK_SIZE;
3736     }
3737 
3738     /*
3739      * Tell QAPI that something happened and the memory
3740      * hotunplug wasn't successful.
3741      */
3742     qapi_event_send_device_unplug_guest_error(dev->id,
3743                                               dev->canonical_path);
3744 }
3745 
3746 /* Callback to be called during DRC release. */
3747 void spapr_lmb_release(DeviceState *dev)
3748 {
3749     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3750     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3751     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3752 
3753     /* This information will get lost if a migration occurs
3754      * during the unplug process. In this case recover it. */
3755     if (ds == NULL) {
3756         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3757         g_assert(ds);
3758         /* The DRC being examined by the caller at least must be counted */
3759         g_assert(ds->nr_lmbs);
3760     }
3761 
3762     if (--ds->nr_lmbs) {
3763         return;
3764     }
3765 
3766     /*
3767      * Now that all the LMBs have been removed by the guest, call the
3768      * unplug handler chain. This can never fail.
3769      */
3770     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3771     object_unparent(OBJECT(dev));
3772 }
3773 
3774 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3775 {
3776     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3777     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3778 
3779     /* We really shouldn't get this far without anything to unplug */
3780     g_assert(ds);
3781 
3782     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3783     qdev_unrealize(dev);
3784     spapr_pending_dimm_unplugs_remove(spapr, ds);
3785 }
3786 
3787 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3788                                         DeviceState *dev, Error **errp)
3789 {
3790     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3791     PCDIMMDevice *dimm = PC_DIMM(dev);
3792     uint32_t nr_lmbs;
3793     uint64_t size, addr_start, addr;
3794     int i;
3795     SpaprDrc *drc;
3796 
3797     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3798         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3799         return;
3800     }
3801 
3802     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3803     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3804 
3805     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3806                                           &error_abort);
3807 
3808     /*
3809      * An existing pending dimm state for this DIMM means that there is an
3810      * unplug operation in progress, waiting for the spapr_lmb_release
3811      * callback to complete the job (BQL can't cover that far). In this case,
3812      * bail out to avoid detaching DRCs that were already released.
3813      */
3814     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3815         error_setg(errp, "Memory unplug already in progress for device %s",
3816                    dev->id);
3817         return;
3818     }
3819 
3820     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3821 
3822     addr = addr_start;
3823     for (i = 0; i < nr_lmbs; i++) {
3824         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3825                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3826         g_assert(drc);
3827 
3828         spapr_drc_unplug_request(drc);
3829         addr += SPAPR_MEMORY_BLOCK_SIZE;
3830     }
3831 
3832     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3833                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3834     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3835                                               nr_lmbs, spapr_drc_index(drc));
3836 }
3837 
3838 /* Callback to be called during DRC release. */
3839 void spapr_core_release(DeviceState *dev)
3840 {
3841     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3842 
3843     /* Call the unplug handler chain. This can never fail. */
3844     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3845     object_unparent(OBJECT(dev));
3846 }
3847 
3848 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3849 {
3850     MachineState *ms = MACHINE(hotplug_dev);
3851     CPUCore *cc = CPU_CORE(dev);
3852     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3853 
3854     assert(core_slot);
3855     core_slot->cpu = NULL;
3856     qdev_unrealize(dev);
3857 }
3858 
3859 static
3860 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3861                                Error **errp)
3862 {
3863     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3864     int index;
3865     SpaprDrc *drc;
3866     CPUCore *cc = CPU_CORE(dev);
3867 
3868     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3869         error_setg(errp, "Unable to find CPU core with core-id: %d",
3870                    cc->core_id);
3871         return;
3872     }
3873     if (index == 0) {
3874         error_setg(errp, "Boot CPU core may not be unplugged");
3875         return;
3876     }
3877 
3878     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3879                           spapr_vcpu_id(spapr, cc->core_id));
3880     g_assert(drc);
3881 
3882     if (!spapr_drc_unplug_requested(drc)) {
3883         spapr_drc_unplug_request(drc);
3884     }
3885 
3886     /*
3887      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3888      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3889      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3890      * attempt (e.g. the kernel will refuse to remove the last online
3891      * CPU), we will never attempt it again because unplug_requested
3892      * will still be 'true' in that case.
3893      */
3894     spapr_hotplug_req_remove_by_index(drc);
3895 }
3896 
3897 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3898                            void *fdt, int *fdt_start_offset, Error **errp)
3899 {
3900     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3901     CPUState *cs = CPU(core->threads[0]);
3902     PowerPCCPU *cpu = POWERPC_CPU(cs);
3903     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3904     int id = spapr_get_vcpu_id(cpu);
3905     g_autofree char *nodename = NULL;
3906     int offset;
3907 
3908     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3909     offset = fdt_add_subnode(fdt, 0, nodename);
3910 
3911     spapr_dt_cpu(cs, fdt, offset, spapr);
3912 
3913     /*
3914      * spapr_dt_cpu() does not fill the 'name' property in the
3915      * CPU node. The function is called during boot process, before
3916      * and after CAS, and overwriting the 'name' property written
3917      * by SLOF is not allowed.
3918      *
3919      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3920      * CPUs more compatible with the coldplugged ones, which have
3921      * the 'name' property. Linux Kernel also relies on this
3922      * property to identify CPU nodes.
3923      */
3924     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3925 
3926     *fdt_start_offset = offset;
3927     return 0;
3928 }
3929 
3930 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3931 {
3932     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3933     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3934     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3935     CPUCore *cc = CPU_CORE(dev);
3936     SpaprDrc *drc;
3937     CPUArchId *core_slot;
3938     int index;
3939     bool hotplugged = spapr_drc_hotplugged(dev);
3940     int i;
3941 
3942     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3943     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3944 
3945     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3946                           spapr_vcpu_id(spapr, cc->core_id));
3947 
3948     g_assert(drc || !mc->has_hotpluggable_cpus);
3949 
3950     if (drc) {
3951         /*
3952          * spapr_core_pre_plug() already buys us this is a brand new
3953          * core being plugged into a free slot. Nothing should already
3954          * be attached to the corresponding DRC.
3955          */
3956         spapr_drc_attach(drc, dev);
3957 
3958         if (hotplugged) {
3959             /*
3960              * Send hotplug notification interrupt to the guest only
3961              * in case of hotplugged CPUs.
3962              */
3963             spapr_hotplug_req_add_by_index(drc);
3964         } else {
3965             spapr_drc_reset(drc);
3966         }
3967     }
3968 
3969     core_slot->cpu = CPU(dev);
3970 
3971     /*
3972      * Set compatibility mode to match the boot CPU, which was either set
3973      * by the machine reset code or by CAS. This really shouldn't fail at
3974      * this point.
3975      */
3976     if (hotplugged) {
3977         for (i = 0; i < cc->nr_threads; i++) {
3978             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3979                            &error_abort);
3980         }
3981     }
3982 
3983 }
3984 
3985 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3986                                 Error **errp)
3987 {
3988     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3989     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3990     CPUCore *cc = CPU_CORE(dev);
3991     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3992     const char *type = object_get_typename(OBJECT(dev));
3993     CPUArchId *core_slot;
3994     int index;
3995     unsigned int smp_threads = machine->smp.threads;
3996 
3997     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3998         error_setg(errp, "CPU hotplug not supported for this machine");
3999         return;
4000     }
4001 
4002     if (strcmp(base_core_type, type)) {
4003         error_setg(errp, "CPU core type should be %s", base_core_type);
4004         return;
4005     }
4006 
4007     if (cc->core_id % smp_threads) {
4008         error_setg(errp, "invalid core id %d", cc->core_id);
4009         return;
4010     }
4011 
4012     /*
4013      * In general we should have homogeneous threads-per-core, but old
4014      * (pre hotplug support) machine types allow the last core to have
4015      * reduced threads as a compatibility hack for when we allowed
4016      * total vcpus not a multiple of threads-per-core.
4017      */
4018     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4019         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4020                    smp_threads);
4021         return;
4022     }
4023 
4024     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4025     if (!core_slot) {
4026         error_setg(errp, "core id %d out of range", cc->core_id);
4027         return;
4028     }
4029 
4030     if (core_slot->cpu) {
4031         error_setg(errp, "core %d already populated", cc->core_id);
4032         return;
4033     }
4034 
4035     numa_cpu_pre_plug(core_slot, dev, errp);
4036 }
4037 
4038 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4039                           void *fdt, int *fdt_start_offset, Error **errp)
4040 {
4041     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4042     int intc_phandle;
4043 
4044     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4045     if (intc_phandle <= 0) {
4046         return -1;
4047     }
4048 
4049     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4050         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4051         return -1;
4052     }
4053 
4054     /* generally SLOF creates these, for hotplug it's up to QEMU */
4055     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4056 
4057     return 0;
4058 }
4059 
4060 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4061                                Error **errp)
4062 {
4063     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4064     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4065     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4066     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4067     SpaprDrc *drc;
4068 
4069     if (dev->hotplugged && !smc->dr_phb_enabled) {
4070         error_setg(errp, "PHB hotplug not supported for this machine");
4071         return false;
4072     }
4073 
4074     if (sphb->index == (uint32_t)-1) {
4075         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4076         return false;
4077     }
4078 
4079     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4080     if (drc && drc->dev) {
4081         error_setg(errp, "PHB %d already attached", sphb->index);
4082         return false;
4083     }
4084 
4085     /*
4086      * This will check that sphb->index doesn't exceed the maximum number of
4087      * PHBs for the current machine type.
4088      */
4089     return
4090         smc->phb_placement(spapr, sphb->index,
4091                            &sphb->buid, &sphb->io_win_addr,
4092                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4093                            windows_supported, sphb->dma_liobn,
4094                            errp);
4095 }
4096 
4097 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4098 {
4099     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4100     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4101     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4102     SpaprDrc *drc;
4103     bool hotplugged = spapr_drc_hotplugged(dev);
4104 
4105     if (!smc->dr_phb_enabled) {
4106         return;
4107     }
4108 
4109     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4110     /* hotplug hooks should check it's enabled before getting this far */
4111     assert(drc);
4112 
4113     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4114     spapr_drc_attach(drc, dev);
4115 
4116     if (hotplugged) {
4117         spapr_hotplug_req_add_by_index(drc);
4118     } else {
4119         spapr_drc_reset(drc);
4120     }
4121 }
4122 
4123 void spapr_phb_release(DeviceState *dev)
4124 {
4125     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4126 
4127     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4128     object_unparent(OBJECT(dev));
4129 }
4130 
4131 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4132 {
4133     qdev_unrealize(dev);
4134 }
4135 
4136 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4137                                      DeviceState *dev, Error **errp)
4138 {
4139     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4140     SpaprDrc *drc;
4141 
4142     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4143     assert(drc);
4144 
4145     if (!spapr_drc_unplug_requested(drc)) {
4146         spapr_drc_unplug_request(drc);
4147         spapr_hotplug_req_remove_by_index(drc);
4148     } else {
4149         error_setg(errp,
4150                    "PCI Host Bridge unplug already in progress for device %s",
4151                    dev->id);
4152     }
4153 }
4154 
4155 static
4156 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4157                               Error **errp)
4158 {
4159     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4160 
4161     if (spapr->tpm_proxy != NULL) {
4162         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4163         return false;
4164     }
4165 
4166     return true;
4167 }
4168 
4169 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4170 {
4171     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4172     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4173 
4174     /* Already checked in spapr_tpm_proxy_pre_plug() */
4175     g_assert(spapr->tpm_proxy == NULL);
4176 
4177     spapr->tpm_proxy = tpm_proxy;
4178 }
4179 
4180 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4181 {
4182     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4183 
4184     qdev_unrealize(dev);
4185     object_unparent(OBJECT(dev));
4186     spapr->tpm_proxy = NULL;
4187 }
4188 
4189 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4190                                       DeviceState *dev, Error **errp)
4191 {
4192     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4193         spapr_memory_plug(hotplug_dev, dev);
4194     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4195         spapr_core_plug(hotplug_dev, dev);
4196     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4197         spapr_phb_plug(hotplug_dev, dev);
4198     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4199         spapr_tpm_proxy_plug(hotplug_dev, dev);
4200     }
4201 }
4202 
4203 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4204                                         DeviceState *dev, Error **errp)
4205 {
4206     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4207         spapr_memory_unplug(hotplug_dev, dev);
4208     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4209         spapr_core_unplug(hotplug_dev, dev);
4210     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4211         spapr_phb_unplug(hotplug_dev, dev);
4212     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4213         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4214     }
4215 }
4216 
4217 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4218 {
4219     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4220         /*
4221          * CAS will process all pending unplug requests.
4222          *
4223          * HACK: a guest could theoretically have cleared all bits in OV5,
4224          * but none of the guests we care for do.
4225          */
4226         spapr_ovec_empty(spapr->ov5_cas);
4227 }
4228 
4229 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4230                                                 DeviceState *dev, Error **errp)
4231 {
4232     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4233     MachineClass *mc = MACHINE_GET_CLASS(sms);
4234     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4235 
4236     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4237         if (spapr_memory_hot_unplug_supported(sms)) {
4238             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4239         } else {
4240             error_setg(errp, "Memory hot unplug not supported for this guest");
4241         }
4242     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4243         if (!mc->has_hotpluggable_cpus) {
4244             error_setg(errp, "CPU hot unplug not supported on this machine");
4245             return;
4246         }
4247         spapr_core_unplug_request(hotplug_dev, dev, errp);
4248     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4249         if (!smc->dr_phb_enabled) {
4250             error_setg(errp, "PHB hot unplug not supported on this machine");
4251             return;
4252         }
4253         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4254     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4255         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4256     }
4257 }
4258 
4259 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4260                                           DeviceState *dev, Error **errp)
4261 {
4262     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4263         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4264     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4265         spapr_core_pre_plug(hotplug_dev, dev, errp);
4266     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4267         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4268     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4269         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4270     }
4271 }
4272 
4273 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4274                                                  DeviceState *dev)
4275 {
4276     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4277         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4278         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4279         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4280         return HOTPLUG_HANDLER(machine);
4281     }
4282     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4283         PCIDevice *pcidev = PCI_DEVICE(dev);
4284         PCIBus *root = pci_device_root_bus(pcidev);
4285         SpaprPhbState *phb =
4286             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4287                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4288 
4289         if (phb) {
4290             return HOTPLUG_HANDLER(phb);
4291         }
4292     }
4293     return NULL;
4294 }
4295 
4296 static CpuInstanceProperties
4297 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4298 {
4299     CPUArchId *core_slot;
4300     MachineClass *mc = MACHINE_GET_CLASS(machine);
4301 
4302     /* make sure possible_cpu are initialized */
4303     mc->possible_cpu_arch_ids(machine);
4304     /* get CPU core slot containing thread that matches cpu_index */
4305     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4306     assert(core_slot);
4307     return core_slot->props;
4308 }
4309 
4310 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4311 {
4312     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4313 }
4314 
4315 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4316 {
4317     int i;
4318     unsigned int smp_threads = machine->smp.threads;
4319     unsigned int smp_cpus = machine->smp.cpus;
4320     const char *core_type;
4321     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4322     MachineClass *mc = MACHINE_GET_CLASS(machine);
4323 
4324     if (!mc->has_hotpluggable_cpus) {
4325         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4326     }
4327     if (machine->possible_cpus) {
4328         assert(machine->possible_cpus->len == spapr_max_cores);
4329         return machine->possible_cpus;
4330     }
4331 
4332     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4333     if (!core_type) {
4334         error_report("Unable to find sPAPR CPU Core definition");
4335         exit(1);
4336     }
4337 
4338     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4339                              sizeof(CPUArchId) * spapr_max_cores);
4340     machine->possible_cpus->len = spapr_max_cores;
4341     for (i = 0; i < machine->possible_cpus->len; i++) {
4342         int core_id = i * smp_threads;
4343 
4344         machine->possible_cpus->cpus[i].type = core_type;
4345         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4346         machine->possible_cpus->cpus[i].arch_id = core_id;
4347         machine->possible_cpus->cpus[i].props.has_core_id = true;
4348         machine->possible_cpus->cpus[i].props.core_id = core_id;
4349     }
4350     return machine->possible_cpus;
4351 }
4352 
4353 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4354                                 uint64_t *buid, hwaddr *pio,
4355                                 hwaddr *mmio32, hwaddr *mmio64,
4356                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4357 {
4358     /*
4359      * New-style PHB window placement.
4360      *
4361      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4362      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4363      * windows.
4364      *
4365      * Some guest kernels can't work with MMIO windows above 1<<46
4366      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4367      *
4368      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4369      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4370      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4371      * 1TiB 64-bit MMIO windows for each PHB.
4372      */
4373     const uint64_t base_buid = 0x800000020000000ULL;
4374     int i;
4375 
4376     /* Sanity check natural alignments */
4377     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4378     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4379     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4380     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4381     /* Sanity check bounds */
4382     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4383                       SPAPR_PCI_MEM32_WIN_SIZE);
4384     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4385                       SPAPR_PCI_MEM64_WIN_SIZE);
4386 
4387     if (index >= SPAPR_MAX_PHBS) {
4388         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4389                    SPAPR_MAX_PHBS - 1);
4390         return false;
4391     }
4392 
4393     *buid = base_buid + index;
4394     for (i = 0; i < n_dma; ++i) {
4395         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4396     }
4397 
4398     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4399     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4400     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4401     return true;
4402 }
4403 
4404 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4405 {
4406     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4407 
4408     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4409 }
4410 
4411 static void spapr_ics_resend(XICSFabric *dev)
4412 {
4413     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4414 
4415     ics_resend(spapr->ics);
4416 }
4417 
4418 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4419 {
4420     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4421 
4422     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4423 }
4424 
4425 static void spapr_pic_print_info(InterruptStatsProvider *obj, GString *buf)
4426 {
4427     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4428 
4429     spapr_irq_print_info(spapr, buf);
4430     g_string_append_printf(buf, "irqchip: %s\n",
4431                            kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4432 }
4433 
4434 /*
4435  * This is a XIVE only operation
4436  */
4437 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4438                            uint8_t nvt_blk, uint32_t nvt_idx,
4439                            bool cam_ignore, uint8_t priority,
4440                            uint32_t logic_serv, XiveTCTXMatch *match)
4441 {
4442     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4443     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4444     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4445     int count;
4446 
4447     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4448                            priority, logic_serv, match);
4449     if (count < 0) {
4450         return count;
4451     }
4452 
4453     /*
4454      * When we implement the save and restore of the thread interrupt
4455      * contexts in the enter/exit CPU handlers of the machine and the
4456      * escalations in QEMU, we should be able to handle non dispatched
4457      * vCPUs.
4458      *
4459      * Until this is done, the sPAPR machine should find at least one
4460      * matching context always.
4461      */
4462     if (count == 0) {
4463         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4464                       nvt_blk, nvt_idx);
4465     }
4466 
4467     return count;
4468 }
4469 
4470 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4471 {
4472     return cpu->vcpu_id;
4473 }
4474 
4475 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4476 {
4477     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4478     MachineState *ms = MACHINE(spapr);
4479     int vcpu_id;
4480 
4481     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4482 
4483     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4484         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4485         error_append_hint(errp, "Adjust the number of cpus to %d "
4486                           "or try to raise the number of threads per core\n",
4487                           vcpu_id * ms->smp.threads / spapr->vsmt);
4488         return false;
4489     }
4490 
4491     cpu->vcpu_id = vcpu_id;
4492     return true;
4493 }
4494 
4495 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4496 {
4497     CPUState *cs;
4498 
4499     CPU_FOREACH(cs) {
4500         PowerPCCPU *cpu = POWERPC_CPU(cs);
4501 
4502         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4503             return cpu;
4504         }
4505     }
4506 
4507     return NULL;
4508 }
4509 
4510 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4511 {
4512     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4513 
4514     return spapr_cpu->in_nested;
4515 }
4516 
4517 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4518 {
4519     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4520 
4521     /* These are only called by TCG, KVM maintains dispatch state */
4522 
4523     spapr_cpu->prod = false;
4524     if (spapr_cpu->vpa_addr) {
4525         CPUState *cs = CPU(cpu);
4526         uint32_t dispatch;
4527 
4528         dispatch = ldl_be_phys(cs->as,
4529                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4530         dispatch++;
4531         if ((dispatch & 1) != 0) {
4532             qemu_log_mask(LOG_GUEST_ERROR,
4533                           "VPA: incorrect dispatch counter value for "
4534                           "dispatched partition %u, correcting.\n", dispatch);
4535             dispatch++;
4536         }
4537         stl_be_phys(cs->as,
4538                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4539     }
4540 }
4541 
4542 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4543 {
4544     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4545 
4546     if (spapr_cpu->vpa_addr) {
4547         CPUState *cs = CPU(cpu);
4548         uint32_t dispatch;
4549 
4550         dispatch = ldl_be_phys(cs->as,
4551                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4552         dispatch++;
4553         if ((dispatch & 1) != 1) {
4554             qemu_log_mask(LOG_GUEST_ERROR,
4555                           "VPA: incorrect dispatch counter value for "
4556                           "preempted partition %u, correcting.\n", dispatch);
4557             dispatch++;
4558         }
4559         stl_be_phys(cs->as,
4560                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4561     }
4562 }
4563 
4564 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4565 {
4566     MachineClass *mc = MACHINE_CLASS(oc);
4567     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4568     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4569     NMIClass *nc = NMI_CLASS(oc);
4570     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4571     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4572     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4573     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4574     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4575     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4576 
4577     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4578     mc->ignore_boot_device_suffixes = true;
4579 
4580     /*
4581      * We set up the default / latest behaviour here.  The class_init
4582      * functions for the specific versioned machine types can override
4583      * these details for backwards compatibility
4584      */
4585     mc->init = spapr_machine_init;
4586     mc->reset = spapr_machine_reset;
4587     mc->block_default_type = IF_SCSI;
4588 
4589     /*
4590      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4591      * In TCG the limit is restricted by the range of CPU IPIs available.
4592      */
4593     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4594 
4595     mc->no_parallel = 1;
4596     mc->default_boot_order = "";
4597     mc->default_ram_size = 512 * MiB;
4598     mc->default_ram_id = "ppc_spapr.ram";
4599     mc->default_display = "std";
4600     mc->kvm_type = spapr_kvm_type;
4601     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4602     mc->pci_allow_0_address = true;
4603     assert(!mc->get_hotplug_handler);
4604     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4605     hc->pre_plug = spapr_machine_device_pre_plug;
4606     hc->plug = spapr_machine_device_plug;
4607     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4608     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4609     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4610     hc->unplug_request = spapr_machine_device_unplug_request;
4611     hc->unplug = spapr_machine_device_unplug;
4612 
4613     smc->update_dt_enabled = true;
4614     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4615     mc->has_hotpluggable_cpus = true;
4616     mc->nvdimm_supported = true;
4617     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4618     fwc->get_dev_path = spapr_get_fw_dev_path;
4619     nc->nmi_monitor_handler = spapr_nmi;
4620     smc->phb_placement = spapr_phb_placement;
4621     vhc->cpu_in_nested = spapr_cpu_in_nested;
4622     vhc->deliver_hv_excp = spapr_exit_nested;
4623     vhc->hypercall = emulate_spapr_hypercall;
4624     vhc->hpt_mask = spapr_hpt_mask;
4625     vhc->map_hptes = spapr_map_hptes;
4626     vhc->unmap_hptes = spapr_unmap_hptes;
4627     vhc->hpte_set_c = spapr_hpte_set_c;
4628     vhc->hpte_set_r = spapr_hpte_set_r;
4629     vhc->get_pate = spapr_get_pate;
4630     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4631     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4632     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4633     xic->ics_get = spapr_ics_get;
4634     xic->ics_resend = spapr_ics_resend;
4635     xic->icp_get = spapr_icp_get;
4636     ispc->print_info = spapr_pic_print_info;
4637     /* Force NUMA node memory size to be a multiple of
4638      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4639      * in which LMBs are represented and hot-added
4640      */
4641     mc->numa_mem_align_shift = 28;
4642     mc->auto_enable_numa = true;
4643 
4644     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4645     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4646     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4647     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4648     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4649     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4650     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4651     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4652     smc->default_caps.caps[SPAPR_CAP_NESTED_PAPR] = SPAPR_CAP_OFF;
4653     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4654     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4655     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4656     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4657 
4658     /*
4659      * This cap specifies whether the AIL 3 mode for
4660      * H_SET_RESOURCE is supported. The default is modified
4661      * by default_caps_with_cpu().
4662      */
4663     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4664     spapr_caps_add_properties(smc);
4665     smc->irq = &spapr_irq_dual;
4666     smc->dr_phb_enabled = true;
4667     smc->linux_pci_probe = true;
4668     smc->smp_threads_vsmt = true;
4669     smc->nr_xirqs = SPAPR_NR_XIRQS;
4670     xfc->match_nvt = spapr_match_nvt;
4671     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4672     vmc->quiesce = spapr_vof_quiesce;
4673     vmc->setprop = spapr_vof_setprop;
4674 }
4675 
4676 static const TypeInfo spapr_machine_info = {
4677     .name          = TYPE_SPAPR_MACHINE,
4678     .parent        = TYPE_MACHINE,
4679     .abstract      = true,
4680     .instance_size = sizeof(SpaprMachineState),
4681     .instance_init = spapr_instance_init,
4682     .instance_finalize = spapr_machine_finalizefn,
4683     .class_size    = sizeof(SpaprMachineClass),
4684     .class_init    = spapr_machine_class_init,
4685     .interfaces = (InterfaceInfo[]) {
4686         { TYPE_FW_PATH_PROVIDER },
4687         { TYPE_NMI },
4688         { TYPE_HOTPLUG_HANDLER },
4689         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4690         { TYPE_XICS_FABRIC },
4691         { TYPE_INTERRUPT_STATS_PROVIDER },
4692         { TYPE_XIVE_FABRIC },
4693         { TYPE_VOF_MACHINE_IF },
4694         { }
4695     },
4696 };
4697 
4698 static void spapr_machine_latest_class_options(MachineClass *mc)
4699 {
4700     mc->alias = "pseries";
4701     mc->is_default = true;
4702 }
4703 
4704 #define DEFINE_SPAPR_MACHINE_IMPL(latest, ...)                       \
4705     static void MACHINE_VER_SYM(class_init, spapr, __VA_ARGS__)(     \
4706         ObjectClass *oc,                                             \
4707         void *data)                                                  \
4708     {                                                                \
4709         MachineClass *mc = MACHINE_CLASS(oc);                        \
4710         MACHINE_VER_SYM(class_options, spapr, __VA_ARGS__)(mc);      \
4711         MACHINE_VER_DEPRECATION(__VA_ARGS__);                        \
4712         if (latest) {                                                \
4713             spapr_machine_latest_class_options(mc);                  \
4714         }                                                            \
4715     }                                                                \
4716     static const TypeInfo MACHINE_VER_SYM(info, spapr, __VA_ARGS__) = \
4717     {                                                                \
4718         .name = MACHINE_VER_TYPE_NAME("pseries", __VA_ARGS__),       \
4719         .parent = TYPE_SPAPR_MACHINE,                                \
4720         .class_init = MACHINE_VER_SYM(class_init, spapr, __VA_ARGS__), \
4721     };                                                               \
4722     static void MACHINE_VER_SYM(register, spapr, __VA_ARGS__)(void)  \
4723     {                                                                \
4724         MACHINE_VER_DELETION(__VA_ARGS__);                           \
4725         type_register_static(&MACHINE_VER_SYM(info, spapr, __VA_ARGS__));   \
4726     }                                                                \
4727     type_init(MACHINE_VER_SYM(register, spapr, __VA_ARGS__))
4728 
4729 #define DEFINE_SPAPR_MACHINE_AS_LATEST(major, minor) \
4730     DEFINE_SPAPR_MACHINE_IMPL(true, major, minor)
4731 #define DEFINE_SPAPR_MACHINE(major, minor) \
4732     DEFINE_SPAPR_MACHINE_IMPL(false, major, minor)
4733 
4734 /*
4735  * pseries-10.0
4736  */
4737 static void spapr_machine_10_0_class_options(MachineClass *mc)
4738 {
4739     /* Defaults for the latest behaviour inherited from the base class */
4740 }
4741 
4742 DEFINE_SPAPR_MACHINE_AS_LATEST(10, 0);
4743 
4744 /*
4745  * pseries-9.2
4746  */
4747 static void spapr_machine_9_2_class_options(MachineClass *mc)
4748 {
4749     spapr_machine_10_0_class_options(mc);
4750     compat_props_add(mc->compat_props, hw_compat_9_2, hw_compat_9_2_len);
4751 }
4752 
4753 DEFINE_SPAPR_MACHINE(9, 2);
4754 
4755 /*
4756  * pseries-9.1
4757  */
4758 static void spapr_machine_9_1_class_options(MachineClass *mc)
4759 {
4760     spapr_machine_9_2_class_options(mc);
4761     compat_props_add(mc->compat_props, hw_compat_9_1, hw_compat_9_1_len);
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(9, 1);
4765 
4766 /*
4767  * pseries-9.0
4768  */
4769 static void spapr_machine_9_0_class_options(MachineClass *mc)
4770 {
4771     spapr_machine_9_1_class_options(mc);
4772     compat_props_add(mc->compat_props, hw_compat_9_0, hw_compat_9_0_len);
4773 }
4774 
4775 DEFINE_SPAPR_MACHINE(9, 0);
4776 
4777 /*
4778  * pseries-8.2
4779  */
4780 static void spapr_machine_8_2_class_options(MachineClass *mc)
4781 {
4782     spapr_machine_9_0_class_options(mc);
4783     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4784 }
4785 
4786 DEFINE_SPAPR_MACHINE(8, 2);
4787 
4788 /*
4789  * pseries-8.1
4790  */
4791 static void spapr_machine_8_1_class_options(MachineClass *mc)
4792 {
4793     spapr_machine_8_2_class_options(mc);
4794     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4795 }
4796 
4797 DEFINE_SPAPR_MACHINE(8, 1);
4798 
4799 /*
4800  * pseries-8.0
4801  */
4802 static void spapr_machine_8_0_class_options(MachineClass *mc)
4803 {
4804     spapr_machine_8_1_class_options(mc);
4805     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4806 }
4807 
4808 DEFINE_SPAPR_MACHINE(8, 0);
4809 
4810 /*
4811  * pseries-7.2
4812  */
4813 static void spapr_machine_7_2_class_options(MachineClass *mc)
4814 {
4815     spapr_machine_8_0_class_options(mc);
4816     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4817 }
4818 
4819 DEFINE_SPAPR_MACHINE(7, 2);
4820 
4821 /*
4822  * pseries-7.1
4823  */
4824 static void spapr_machine_7_1_class_options(MachineClass *mc)
4825 {
4826     spapr_machine_7_2_class_options(mc);
4827     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4828 }
4829 
4830 DEFINE_SPAPR_MACHINE(7, 1);
4831 
4832 /*
4833  * pseries-7.0
4834  */
4835 static void spapr_machine_7_0_class_options(MachineClass *mc)
4836 {
4837     spapr_machine_7_1_class_options(mc);
4838     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4839 }
4840 
4841 DEFINE_SPAPR_MACHINE(7, 0);
4842 
4843 /*
4844  * pseries-6.2
4845  */
4846 static void spapr_machine_6_2_class_options(MachineClass *mc)
4847 {
4848     spapr_machine_7_0_class_options(mc);
4849     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4850 }
4851 
4852 DEFINE_SPAPR_MACHINE(6, 2);
4853 
4854 /*
4855  * pseries-6.1
4856  */
4857 static void spapr_machine_6_1_class_options(MachineClass *mc)
4858 {
4859     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4860 
4861     spapr_machine_6_2_class_options(mc);
4862     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4863     smc->pre_6_2_numa_affinity = true;
4864     mc->smp_props.prefer_sockets = true;
4865 }
4866 
4867 DEFINE_SPAPR_MACHINE(6, 1);
4868 
4869 /*
4870  * pseries-6.0
4871  */
4872 static void spapr_machine_6_0_class_options(MachineClass *mc)
4873 {
4874     spapr_machine_6_1_class_options(mc);
4875     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4876 }
4877 
4878 DEFINE_SPAPR_MACHINE(6, 0);
4879 
4880 /*
4881  * pseries-5.2
4882  */
4883 static void spapr_machine_5_2_class_options(MachineClass *mc)
4884 {
4885     spapr_machine_6_0_class_options(mc);
4886     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4887 }
4888 
4889 DEFINE_SPAPR_MACHINE(5, 2);
4890 
4891 /*
4892  * pseries-5.1
4893  */
4894 static void spapr_machine_5_1_class_options(MachineClass *mc)
4895 {
4896     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4897 
4898     spapr_machine_5_2_class_options(mc);
4899     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4900     smc->pre_5_2_numa_associativity = true;
4901 }
4902 
4903 DEFINE_SPAPR_MACHINE(5, 1);
4904 
4905 /*
4906  * pseries-5.0
4907  */
4908 static void spapr_machine_5_0_class_options(MachineClass *mc)
4909 {
4910     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4911     static GlobalProperty compat[] = {
4912         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4913     };
4914 
4915     spapr_machine_5_1_class_options(mc);
4916     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4917     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4918     mc->numa_mem_supported = true;
4919     smc->pre_5_1_assoc_refpoints = true;
4920 }
4921 
4922 DEFINE_SPAPR_MACHINE(5, 0);
4923 
4924 /*
4925  * pseries-4.2
4926  */
4927 static void spapr_machine_4_2_class_options(MachineClass *mc)
4928 {
4929     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4930 
4931     spapr_machine_5_0_class_options(mc);
4932     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4933     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4934     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4935     smc->rma_limit = 16 * GiB;
4936     mc->nvdimm_supported = false;
4937 }
4938 
4939 DEFINE_SPAPR_MACHINE(4, 2);
4940 
4941 /*
4942  * pseries-4.1
4943  */
4944 static void spapr_machine_4_1_class_options(MachineClass *mc)
4945 {
4946     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4947     static GlobalProperty compat[] = {
4948         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4949         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4950     };
4951 
4952     spapr_machine_4_2_class_options(mc);
4953     smc->linux_pci_probe = false;
4954     smc->smp_threads_vsmt = false;
4955     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4956     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4957 }
4958 
4959 DEFINE_SPAPR_MACHINE(4, 1);
4960 
4961 /*
4962  * pseries-4.0
4963  */
4964 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4965                               uint64_t *buid, hwaddr *pio,
4966                               hwaddr *mmio32, hwaddr *mmio64,
4967                               unsigned n_dma, uint32_t *liobns, Error **errp)
4968 {
4969     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4970                              liobns, errp)) {
4971         return false;
4972     }
4973     return true;
4974 }
4975 static void spapr_machine_4_0_class_options(MachineClass *mc)
4976 {
4977     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4978 
4979     spapr_machine_4_1_class_options(mc);
4980     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4981     smc->phb_placement = phb_placement_4_0;
4982     smc->irq = &spapr_irq_xics;
4983     smc->pre_4_1_migration = true;
4984 }
4985 
4986 DEFINE_SPAPR_MACHINE(4, 0);
4987 
4988 /*
4989  * pseries-3.1
4990  */
4991 static void spapr_machine_3_1_class_options(MachineClass *mc)
4992 {
4993     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4994 
4995     spapr_machine_4_0_class_options(mc);
4996     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4997 
4998     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4999     smc->update_dt_enabled = false;
5000     smc->dr_phb_enabled = false;
5001     smc->broken_host_serial_model = true;
5002     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5003     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5004     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5005     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5006 }
5007 
5008 DEFINE_SPAPR_MACHINE(3, 1);
5009 
5010 /*
5011  * pseries-3.0
5012  */
5013 
5014 static void spapr_machine_3_0_class_options(MachineClass *mc)
5015 {
5016     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5017 
5018     spapr_machine_3_1_class_options(mc);
5019     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5020 
5021     smc->legacy_irq_allocation = true;
5022     smc->nr_xirqs = 0x400;
5023     smc->irq = &spapr_irq_xics_legacy;
5024 }
5025 
5026 DEFINE_SPAPR_MACHINE(3, 0);
5027 
5028 static void spapr_machine_register_types(void)
5029 {
5030     type_register_static(&spapr_machine_info);
5031 }
5032 
5033 type_init(spapr_machine_register_types)
5034