xref: /qemu/hw/ppc/spapr.c (revision 1e0e11085a0d3e47cbbad8944f70b371d22b987d)
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-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define FDT_MAX_SIZE            0x100000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
222                                        PowerPCCPU *cpu,
223                                        void *fdt, int offset)
224 {
225     uint8_t pa_features_206[] = { 6, 0,
226         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227     uint8_t pa_features_207[] = { 24, 0,
228         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232     uint8_t pa_features_300[] = { 66, 0,
233         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
236         /* 6: DS207 */
237         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
238         /* 16: Vector */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248         /* 42: PM, 44: PC RA, 46: SC vec'd */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250         /* 48: SIMD, 50: QP BFP, 52: String */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252         /* 54: DecFP, 56: DecI, 58: SHA */
253         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254         /* 60: NM atomic, 62: RNG */
255         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
256     };
257     uint8_t *pa_features = NULL;
258     size_t pa_size;
259 
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_206;
262         pa_size = sizeof(pa_features_206);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_207;
266         pa_size = sizeof(pa_features_207);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_300;
270         pa_size = sizeof(pa_features_300);
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 hwaddr spapr_node0_size(MachineState *machine)
300 {
301     if (machine->numa_state->num_nodes) {
302         int i;
303         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
304             if (machine->numa_state->nodes[i].node_mem) {
305                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
306                            machine->ram_size);
307             }
308         }
309     }
310     return machine->ram_size;
311 }
312 
313 static void add_str(GString *s, const gchar *s1)
314 {
315     g_string_append_len(s, s1, strlen(s1) + 1);
316 }
317 
318 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
319                                        hwaddr size)
320 {
321     uint32_t associativity[] = {
322         cpu_to_be32(0x4), /* length */
323         cpu_to_be32(0x0), cpu_to_be32(0x0),
324         cpu_to_be32(0x0), cpu_to_be32(nodeid)
325     };
326     char mem_name[32];
327     uint64_t mem_reg_property[2];
328     int off;
329 
330     mem_reg_property[0] = cpu_to_be64(start);
331     mem_reg_property[1] = cpu_to_be64(size);
332 
333     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
334     off = fdt_add_subnode(fdt, 0, mem_name);
335     _FDT(off);
336     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
337     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
338                       sizeof(mem_reg_property))));
339     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
340                       sizeof(associativity))));
341     return off;
342 }
343 
344 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
345 {
346     MemoryDeviceInfoList *info;
347 
348     for (info = list; info; info = info->next) {
349         MemoryDeviceInfo *value = info->value;
350 
351         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
352             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
353 
354             if (addr >= pcdimm_info->addr &&
355                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
356                 return pcdimm_info->node;
357             }
358         }
359     }
360 
361     return -1;
362 }
363 
364 struct sPAPRDrconfCellV2 {
365      uint32_t seq_lmbs;
366      uint64_t base_addr;
367      uint32_t drc_index;
368      uint32_t aa_index;
369      uint32_t flags;
370 } QEMU_PACKED;
371 
372 typedef struct DrconfCellQueue {
373     struct sPAPRDrconfCellV2 cell;
374     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
375 } DrconfCellQueue;
376 
377 static DrconfCellQueue *
378 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
379                       uint32_t drc_index, uint32_t aa_index,
380                       uint32_t flags)
381 {
382     DrconfCellQueue *elem;
383 
384     elem = g_malloc0(sizeof(*elem));
385     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
386     elem->cell.base_addr = cpu_to_be64(base_addr);
387     elem->cell.drc_index = cpu_to_be32(drc_index);
388     elem->cell.aa_index = cpu_to_be32(aa_index);
389     elem->cell.flags = cpu_to_be32(flags);
390 
391     return elem;
392 }
393 
394 /* ibm,dynamic-memory-v2 */
395 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
396                                    int offset, MemoryDeviceInfoList *dimms)
397 {
398     MachineState *machine = MACHINE(spapr);
399     uint8_t *int_buf, *cur_index;
400     int ret;
401     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
402     uint64_t addr, cur_addr, size;
403     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
404     uint64_t mem_end = machine->device_memory->base +
405                        memory_region_size(&machine->device_memory->mr);
406     uint32_t node, buf_len, nr_entries = 0;
407     SpaprDrc *drc;
408     DrconfCellQueue *elem, *next;
409     MemoryDeviceInfoList *info;
410     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
411         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
412 
413     /* Entry to cover RAM and the gap area */
414     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
415                                  SPAPR_LMB_FLAGS_RESERVED |
416                                  SPAPR_LMB_FLAGS_DRC_INVALID);
417     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
418     nr_entries++;
419 
420     cur_addr = machine->device_memory->base;
421     for (info = dimms; info; info = info->next) {
422         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
423 
424         addr = di->addr;
425         size = di->size;
426         node = di->node;
427 
428         /*
429          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
430          * area is marked hotpluggable in the next iteration for the bigger
431          * chunk including the NVDIMM occupied area.
432          */
433         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
434             continue;
435 
436         /* Entry for hot-pluggable area */
437         if (cur_addr < addr) {
438             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
439             g_assert(drc);
440             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
441                                          cur_addr, spapr_drc_index(drc), -1, 0);
442             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
443             nr_entries++;
444         }
445 
446         /* Entry for DIMM */
447         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
448         g_assert(drc);
449         elem = spapr_get_drconf_cell(size / lmb_size, addr,
450                                      spapr_drc_index(drc), node,
451                                      SPAPR_LMB_FLAGS_ASSIGNED);
452         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
453         nr_entries++;
454         cur_addr = addr + size;
455     }
456 
457     /* Entry for remaining hotpluggable area */
458     if (cur_addr < mem_end) {
459         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
460         g_assert(drc);
461         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
462                                      cur_addr, spapr_drc_index(drc), -1, 0);
463         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
464         nr_entries++;
465     }
466 
467     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
468     int_buf = cur_index = g_malloc0(buf_len);
469     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
470     cur_index += sizeof(nr_entries);
471 
472     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
473         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
474         cur_index += sizeof(elem->cell);
475         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
476         g_free(elem);
477     }
478 
479     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
480     g_free(int_buf);
481     if (ret < 0) {
482         return -1;
483     }
484     return 0;
485 }
486 
487 /* ibm,dynamic-memory */
488 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
489                                    int offset, MemoryDeviceInfoList *dimms)
490 {
491     MachineState *machine = MACHINE(spapr);
492     int i, ret;
493     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
494     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
495     uint32_t nr_lmbs = (machine->device_memory->base +
496                        memory_region_size(&machine->device_memory->mr)) /
497                        lmb_size;
498     uint32_t *int_buf, *cur_index, buf_len;
499 
500     /*
501      * Allocate enough buffer size to fit in ibm,dynamic-memory
502      */
503     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
504     cur_index = int_buf = g_malloc0(buf_len);
505     int_buf[0] = cpu_to_be32(nr_lmbs);
506     cur_index++;
507     for (i = 0; i < nr_lmbs; i++) {
508         uint64_t addr = i * lmb_size;
509         uint32_t *dynamic_memory = cur_index;
510 
511         if (i >= device_lmb_start) {
512             SpaprDrc *drc;
513 
514             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
515             g_assert(drc);
516 
517             dynamic_memory[0] = cpu_to_be32(addr >> 32);
518             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
519             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
520             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
521             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
522             if (memory_region_present(get_system_memory(), addr)) {
523                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
524             } else {
525                 dynamic_memory[5] = cpu_to_be32(0);
526             }
527         } else {
528             /*
529              * LMB information for RMA, boot time RAM and gap b/n RAM and
530              * device memory region -- all these are marked as reserved
531              * and as having no valid DRC.
532              */
533             dynamic_memory[0] = cpu_to_be32(addr >> 32);
534             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
535             dynamic_memory[2] = cpu_to_be32(0);
536             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
537             dynamic_memory[4] = cpu_to_be32(-1);
538             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
539                                             SPAPR_LMB_FLAGS_DRC_INVALID);
540         }
541 
542         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
543     }
544     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
545     g_free(int_buf);
546     if (ret < 0) {
547         return -1;
548     }
549     return 0;
550 }
551 
552 /*
553  * Adds ibm,dynamic-reconfiguration-memory node.
554  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
555  * of this device tree node.
556  */
557 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
558 {
559     MachineState *machine = MACHINE(spapr);
560     int nb_numa_nodes = machine->numa_state->num_nodes;
561     int ret, i, offset;
562     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
563     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
564     uint32_t *int_buf, *cur_index, buf_len;
565     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
566     MemoryDeviceInfoList *dimms = NULL;
567 
568     /*
569      * Don't create the node if there is no device memory
570      */
571     if (machine->ram_size == machine->maxram_size) {
572         return 0;
573     }
574 
575     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
576 
577     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
578                     sizeof(prop_lmb_size));
579     if (ret < 0) {
580         return ret;
581     }
582 
583     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
584     if (ret < 0) {
585         return ret;
586     }
587 
588     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
589     if (ret < 0) {
590         return ret;
591     }
592 
593     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
594     dimms = qmp_memory_device_list();
595     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
596         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
597     } else {
598         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
599     }
600     qapi_free_MemoryDeviceInfoList(dimms);
601 
602     if (ret < 0) {
603         return ret;
604     }
605 
606     /* ibm,associativity-lookup-arrays */
607     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
608     cur_index = int_buf = g_malloc0(buf_len);
609     int_buf[0] = cpu_to_be32(nr_nodes);
610     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
611     cur_index += 2;
612     for (i = 0; i < nr_nodes; i++) {
613         uint32_t associativity[] = {
614             cpu_to_be32(0x0),
615             cpu_to_be32(0x0),
616             cpu_to_be32(0x0),
617             cpu_to_be32(i)
618         };
619         memcpy(cur_index, associativity, sizeof(associativity));
620         cur_index += 4;
621     }
622     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
623             (cur_index - int_buf) * sizeof(uint32_t));
624     g_free(int_buf);
625 
626     return ret;
627 }
628 
629 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
630 {
631     MachineState *machine = MACHINE(spapr);
632     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
633     hwaddr mem_start, node_size;
634     int i, nb_nodes = machine->numa_state->num_nodes;
635     NodeInfo *nodes = machine->numa_state->nodes;
636 
637     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
638         if (!nodes[i].node_mem) {
639             continue;
640         }
641         if (mem_start >= machine->ram_size) {
642             node_size = 0;
643         } else {
644             node_size = nodes[i].node_mem;
645             if (node_size > machine->ram_size - mem_start) {
646                 node_size = machine->ram_size - mem_start;
647             }
648         }
649         if (!mem_start) {
650             /* spapr_machine_init() checks for rma_size <= node0_size
651              * already */
652             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
653             mem_start += spapr->rma_size;
654             node_size -= spapr->rma_size;
655         }
656         for ( ; node_size; ) {
657             hwaddr sizetmp = pow2floor(node_size);
658 
659             /* mem_start != 0 here */
660             if (ctzl(mem_start) < ctzl(sizetmp)) {
661                 sizetmp = 1ULL << ctzl(mem_start);
662             }
663 
664             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
665             node_size -= sizetmp;
666             mem_start += sizetmp;
667         }
668     }
669 
670     /* Generate ibm,dynamic-reconfiguration-memory node if required */
671     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
672         int ret;
673 
674         g_assert(smc->dr_lmb_enabled);
675         ret = spapr_populate_drconf_memory(spapr, fdt);
676         if (ret) {
677             return ret;
678         }
679     }
680 
681     return 0;
682 }
683 
684 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
685                                   SpaprMachineState *spapr)
686 {
687     MachineState *ms = MACHINE(spapr);
688     PowerPCCPU *cpu = POWERPC_CPU(cs);
689     CPUPPCState *env = &cpu->env;
690     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
691     int index = spapr_get_vcpu_id(cpu);
692     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
693                        0xffffffff, 0xffffffff};
694     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
695         : SPAPR_TIMEBASE_FREQ;
696     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
697     uint32_t page_sizes_prop[64];
698     size_t page_sizes_prop_size;
699     unsigned int smp_threads = ms->smp.threads;
700     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
701     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
702     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
703     SpaprDrc *drc;
704     int drc_index;
705     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
706     int i;
707 
708     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
709     if (drc) {
710         drc_index = spapr_drc_index(drc);
711         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
712     }
713 
714     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
715     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
716 
717     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
718     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
719                            env->dcache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
721                            env->dcache_line_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
723                            env->icache_line_size)));
724     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
725                            env->icache_line_size)));
726 
727     if (pcc->l1_dcache_size) {
728         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
729                                pcc->l1_dcache_size)));
730     } else {
731         warn_report("Unknown L1 dcache size for cpu");
732     }
733     if (pcc->l1_icache_size) {
734         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
735                                pcc->l1_icache_size)));
736     } else {
737         warn_report("Unknown L1 icache size for cpu");
738     }
739 
740     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
741     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
742     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
743     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
744     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
745     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
746 
747     if (env->spr_cb[SPR_PURR].oea_read) {
748         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
749     }
750     if (env->spr_cb[SPR_SPURR].oea_read) {
751         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
752     }
753 
754     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
755         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
756                           segs, sizeof(segs))));
757     }
758 
759     /* Advertise VSX (vector extensions) if available
760      *   1               == VMX / Altivec available
761      *   2               == VSX available
762      *
763      * Only CPUs for which we create core types in spapr_cpu_core.c
764      * are possible, and all of those have VMX */
765     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
766         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
767     } else {
768         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
769     }
770 
771     /* Advertise DFP (Decimal Floating Point) if available
772      *   0 / no property == no DFP
773      *   1               == DFP available */
774     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
775         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
776     }
777 
778     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
779                                                       sizeof(page_sizes_prop));
780     if (page_sizes_prop_size) {
781         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
782                           page_sizes_prop, page_sizes_prop_size)));
783     }
784 
785     spapr_populate_pa_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_fixup_cpu_numa_dt(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_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
820 {
821     CPUState **rev;
822     CPUState *cs;
823     int n_cpus;
824     int cpus_offset;
825     char *nodename;
826     int i;
827 
828     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
829     _FDT(cpus_offset);
830     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
831     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
832 
833     /*
834      * We walk the CPUs in reverse order to ensure that CPU DT nodes
835      * created by fdt_add_subnode() end up in the right order in FDT
836      * for the guest kernel the enumerate the CPUs correctly.
837      *
838      * The CPU list cannot be traversed in reverse order, so we need
839      * to do extra work.
840      */
841     n_cpus = 0;
842     rev = NULL;
843     CPU_FOREACH(cs) {
844         rev = g_renew(CPUState *, rev, n_cpus + 1);
845         rev[n_cpus++] = cs;
846     }
847 
848     for (i = n_cpus - 1; i >= 0; i--) {
849         CPUState *cs = rev[i];
850         PowerPCCPU *cpu = POWERPC_CPU(cs);
851         int index = spapr_get_vcpu_id(cpu);
852         DeviceClass *dc = DEVICE_GET_CLASS(cs);
853         int offset;
854 
855         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
856             continue;
857         }
858 
859         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
860         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
861         g_free(nodename);
862         _FDT(offset);
863         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
864     }
865 
866     g_free(rev);
867 }
868 
869 static int spapr_rng_populate_dt(void *fdt)
870 {
871     int node;
872     int ret;
873 
874     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
875     if (node <= 0) {
876         return -1;
877     }
878     ret = fdt_setprop_string(fdt, node, "device_type",
879                              "ibm,platform-facilities");
880     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
881     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
882 
883     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
884     if (node <= 0) {
885         return -1;
886     }
887     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
888 
889     return ret ? -1 : 0;
890 }
891 
892 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
893 {
894     MachineState *ms = MACHINE(spapr);
895     int rtas;
896     GString *hypertas = g_string_sized_new(256);
897     GString *qemu_hypertas = g_string_sized_new(256);
898     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
899     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
900         memory_region_size(&MACHINE(spapr)->device_memory->mr);
901     uint32_t lrdr_capacity[] = {
902         cpu_to_be32(max_device_addr >> 32),
903         cpu_to_be32(max_device_addr & 0xffffffff),
904         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
905         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
906     };
907     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
908     uint32_t maxdomains[] = {
909         cpu_to_be32(4),
910         maxdomain,
911         maxdomain,
912         maxdomain,
913         cpu_to_be32(spapr->gpu_numa_id),
914     };
915 
916     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
917 
918     /* hypertas */
919     add_str(hypertas, "hcall-pft");
920     add_str(hypertas, "hcall-term");
921     add_str(hypertas, "hcall-dabr");
922     add_str(hypertas, "hcall-interrupt");
923     add_str(hypertas, "hcall-tce");
924     add_str(hypertas, "hcall-vio");
925     add_str(hypertas, "hcall-splpar");
926     add_str(hypertas, "hcall-join");
927     add_str(hypertas, "hcall-bulk");
928     add_str(hypertas, "hcall-set-mode");
929     add_str(hypertas, "hcall-sprg0");
930     add_str(hypertas, "hcall-copy");
931     add_str(hypertas, "hcall-debug");
932     add_str(hypertas, "hcall-vphn");
933     add_str(qemu_hypertas, "hcall-memop1");
934 
935     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
936         add_str(hypertas, "hcall-multi-tce");
937     }
938 
939     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
940         add_str(hypertas, "hcall-hpt-resize");
941     }
942 
943     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
944                      hypertas->str, hypertas->len));
945     g_string_free(hypertas, TRUE);
946     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
947                      qemu_hypertas->str, qemu_hypertas->len));
948     g_string_free(qemu_hypertas, TRUE);
949 
950     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
951                      refpoints, sizeof(refpoints)));
952 
953     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
954                      maxdomains, sizeof(maxdomains)));
955 
956     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_SIZE));
957     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
958                           RTAS_ERROR_LOG_MAX));
959     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
960                           RTAS_EVENT_SCAN_RATE));
961 
962     g_assert(msi_nonbroken);
963     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
964 
965     /*
966      * According to PAPR, rtas ibm,os-term does not guarantee a return
967      * back to the guest cpu.
968      *
969      * While an additional ibm,extended-os-term property indicates
970      * that rtas call return will always occur. Set this property.
971      */
972     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
973 
974     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
975                      lrdr_capacity, sizeof(lrdr_capacity)));
976 
977     spapr_dt_rtas_tokens(fdt, rtas);
978 }
979 
980 /*
981  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
982  * and the XIVE features that the guest may request and thus the valid
983  * values for bytes 23..26 of option vector 5:
984  */
985 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
986                                           int chosen)
987 {
988     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
989 
990     char val[2 * 4] = {
991         23, 0x00, /* XICS / XIVE mode */
992         24, 0x00, /* Hash/Radix, filled in below. */
993         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
994         26, 0x40, /* Radix options: GTSE == yes. */
995     };
996 
997     if (spapr->irq->xics && spapr->irq->xive) {
998         val[1] = SPAPR_OV5_XIVE_BOTH;
999     } else if (spapr->irq->xive) {
1000         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1001     } else {
1002         assert(spapr->irq->xics);
1003         val[1] = SPAPR_OV5_XIVE_LEGACY;
1004     }
1005 
1006     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1007                           first_ppc_cpu->compat_pvr)) {
1008         /*
1009          * If we're in a pre POWER9 compat mode then the guest should
1010          * do hash and use the legacy interrupt mode
1011          */
1012         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1013         val[3] = 0x00; /* Hash */
1014     } else if (kvm_enabled()) {
1015         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1016             val[3] = 0x80; /* OV5_MMU_BOTH */
1017         } else if (kvmppc_has_cap_mmu_radix()) {
1018             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1019         } else {
1020             val[3] = 0x00; /* Hash */
1021         }
1022     } else {
1023         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1024         val[3] = 0xC0;
1025     }
1026     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1027                      val, sizeof(val)));
1028 }
1029 
1030 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1031 {
1032     MachineState *machine = MACHINE(spapr);
1033     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1034     int chosen;
1035 
1036     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1037 
1038     if (reset) {
1039         const char *boot_device = machine->boot_order;
1040         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1041         size_t cb = 0;
1042         char *bootlist = get_boot_devices_list(&cb);
1043 
1044         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1045             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1046                                     machine->kernel_cmdline));
1047         }
1048 
1049         if (spapr->initrd_size) {
1050             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1051                                   spapr->initrd_base));
1052             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1053                                   spapr->initrd_base + spapr->initrd_size));
1054         }
1055 
1056         if (spapr->kernel_size) {
1057             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1058                                   cpu_to_be64(spapr->kernel_size) };
1059 
1060             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1061                          &kprop, sizeof(kprop)));
1062             if (spapr->kernel_le) {
1063                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1064             }
1065         }
1066         if (boot_menu) {
1067             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1068         }
1069         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1070         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1071         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1072 
1073         if (cb && bootlist) {
1074             int i;
1075 
1076             for (i = 0; i < cb; i++) {
1077                 if (bootlist[i] == '\n') {
1078                     bootlist[i] = ' ';
1079                 }
1080             }
1081             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1082         }
1083 
1084         if (boot_device && strlen(boot_device)) {
1085             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1086         }
1087 
1088         if (!spapr->has_graphics && stdout_path) {
1089             /*
1090              * "linux,stdout-path" and "stdout" properties are
1091              * deprecated by linux kernel. New platforms should only
1092              * use the "stdout-path" property. Set the new property
1093              * and continue using older property to remain compatible
1094              * with the existing firmware.
1095              */
1096             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1097             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1098         }
1099 
1100         /*
1101          * We can deal with BAR reallocation just fine, advertise it
1102          * to the guest
1103          */
1104         if (smc->linux_pci_probe) {
1105             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1106         }
1107 
1108         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1109 
1110         g_free(stdout_path);
1111         g_free(bootlist);
1112     }
1113 
1114     _FDT(spapr_ovec_populate_dt(fdt, chosen, spapr->ov5_cas,
1115                                 "ibm,architecture-vec-5"));
1116 }
1117 
1118 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1119 {
1120     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1121      * KVM to work under pHyp with some guest co-operation */
1122     int hypervisor;
1123     uint8_t hypercall[16];
1124 
1125     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1126     /* indicate KVM hypercall interface */
1127     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1128     if (kvmppc_has_cap_fixup_hcalls()) {
1129         /*
1130          * Older KVM versions with older guest kernels were broken
1131          * with the magic page, don't allow the guest to map it.
1132          */
1133         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1134                                   sizeof(hypercall))) {
1135             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1136                              hypercall, sizeof(hypercall)));
1137         }
1138     }
1139 }
1140 
1141 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1142 {
1143     MachineState *machine = MACHINE(spapr);
1144     MachineClass *mc = MACHINE_GET_CLASS(machine);
1145     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1146     int ret;
1147     void *fdt;
1148     SpaprPhbState *phb;
1149     char *buf;
1150 
1151     fdt = g_malloc0(space);
1152     _FDT((fdt_create_empty_tree(fdt, space)));
1153 
1154     /* Root node */
1155     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1156     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1157     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1158 
1159     /* Guest UUID & Name*/
1160     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1161     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1162     if (qemu_uuid_set) {
1163         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1164     }
1165     g_free(buf);
1166 
1167     if (qemu_get_vm_name()) {
1168         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1169                                 qemu_get_vm_name()));
1170     }
1171 
1172     /* Host Model & Serial Number */
1173     if (spapr->host_model) {
1174         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1175     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1176         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1177         g_free(buf);
1178     }
1179 
1180     if (spapr->host_serial) {
1181         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1182     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1183         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1184         g_free(buf);
1185     }
1186 
1187     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1188     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1189 
1190     /* /interrupt controller */
1191     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1192 
1193     ret = spapr_populate_memory(spapr, fdt);
1194     if (ret < 0) {
1195         error_report("couldn't setup memory nodes in fdt");
1196         exit(1);
1197     }
1198 
1199     /* /vdevice */
1200     spapr_dt_vdevice(spapr->vio_bus, fdt);
1201 
1202     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1203         ret = spapr_rng_populate_dt(fdt);
1204         if (ret < 0) {
1205             error_report("could not set up rng device in the fdt");
1206             exit(1);
1207         }
1208     }
1209 
1210     QLIST_FOREACH(phb, &spapr->phbs, list) {
1211         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1212         if (ret < 0) {
1213             error_report("couldn't setup PCI devices in fdt");
1214             exit(1);
1215         }
1216     }
1217 
1218     /* cpus */
1219     spapr_populate_cpus_dt_node(fdt, spapr);
1220 
1221     if (smc->dr_lmb_enabled) {
1222         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1223     }
1224 
1225     if (mc->has_hotpluggable_cpus) {
1226         int offset = fdt_path_offset(fdt, "/cpus");
1227         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1228         if (ret < 0) {
1229             error_report("Couldn't set up CPU DR device tree properties");
1230             exit(1);
1231         }
1232     }
1233 
1234     /* /event-sources */
1235     spapr_dt_events(spapr, fdt);
1236 
1237     /* /rtas */
1238     spapr_dt_rtas(spapr, fdt);
1239 
1240     /* /chosen */
1241     spapr_dt_chosen(spapr, fdt, reset);
1242 
1243     /* /hypervisor */
1244     if (kvm_enabled()) {
1245         spapr_dt_hypervisor(spapr, fdt);
1246     }
1247 
1248     /* Build memory reserve map */
1249     if (reset) {
1250         if (spapr->kernel_size) {
1251             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1252                                   spapr->kernel_size)));
1253         }
1254         if (spapr->initrd_size) {
1255             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1256                                   spapr->initrd_size)));
1257         }
1258     }
1259 
1260     if (smc->dr_phb_enabled) {
1261         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1262         if (ret < 0) {
1263             error_report("Couldn't set up PHB DR device tree properties");
1264             exit(1);
1265         }
1266     }
1267 
1268     /* NVDIMM devices */
1269     if (mc->nvdimm_supported) {
1270         spapr_dt_persistent_memory(fdt);
1271     }
1272 
1273     return fdt;
1274 }
1275 
1276 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1277 {
1278     SpaprMachineState *spapr = opaque;
1279 
1280     return (addr & 0x0fffffff) + spapr->kernel_addr;
1281 }
1282 
1283 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1284                                     PowerPCCPU *cpu)
1285 {
1286     CPUPPCState *env = &cpu->env;
1287 
1288     /* The TCG path should also be holding the BQL at this point */
1289     g_assert(qemu_mutex_iothread_locked());
1290 
1291     if (msr_pr) {
1292         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1293         env->gpr[3] = H_PRIVILEGE;
1294     } else {
1295         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1296     }
1297 }
1298 
1299 struct LPCRSyncState {
1300     target_ulong value;
1301     target_ulong mask;
1302 };
1303 
1304 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1305 {
1306     struct LPCRSyncState *s = arg.host_ptr;
1307     PowerPCCPU *cpu = POWERPC_CPU(cs);
1308     CPUPPCState *env = &cpu->env;
1309     target_ulong lpcr;
1310 
1311     cpu_synchronize_state(cs);
1312     lpcr = env->spr[SPR_LPCR];
1313     lpcr &= ~s->mask;
1314     lpcr |= s->value;
1315     ppc_store_lpcr(cpu, lpcr);
1316 }
1317 
1318 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1319 {
1320     CPUState *cs;
1321     struct LPCRSyncState s = {
1322         .value = value,
1323         .mask = mask
1324     };
1325     CPU_FOREACH(cs) {
1326         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1327     }
1328 }
1329 
1330 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1331 {
1332     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1333 
1334     /* Copy PATE1:GR into PATE0:HR */
1335     entry->dw0 = spapr->patb_entry & PATE0_HR;
1336     entry->dw1 = spapr->patb_entry;
1337 }
1338 
1339 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1340 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1341 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1342 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1343 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1344 
1345 /*
1346  * Get the fd to access the kernel htab, re-opening it if necessary
1347  */
1348 static int get_htab_fd(SpaprMachineState *spapr)
1349 {
1350     Error *local_err = NULL;
1351 
1352     if (spapr->htab_fd >= 0) {
1353         return spapr->htab_fd;
1354     }
1355 
1356     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1357     if (spapr->htab_fd < 0) {
1358         error_report_err(local_err);
1359     }
1360 
1361     return spapr->htab_fd;
1362 }
1363 
1364 void close_htab_fd(SpaprMachineState *spapr)
1365 {
1366     if (spapr->htab_fd >= 0) {
1367         close(spapr->htab_fd);
1368     }
1369     spapr->htab_fd = -1;
1370 }
1371 
1372 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1373 {
1374     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1375 
1376     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1377 }
1378 
1379 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1380 {
1381     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1382 
1383     assert(kvm_enabled());
1384 
1385     if (!spapr->htab) {
1386         return 0;
1387     }
1388 
1389     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1390 }
1391 
1392 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1393                                                 hwaddr ptex, int n)
1394 {
1395     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1396     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1397 
1398     if (!spapr->htab) {
1399         /*
1400          * HTAB is controlled by KVM. Fetch into temporary buffer
1401          */
1402         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1403         kvmppc_read_hptes(hptes, ptex, n);
1404         return hptes;
1405     }
1406 
1407     /*
1408      * HTAB is controlled by QEMU. Just point to the internally
1409      * accessible PTEG.
1410      */
1411     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1412 }
1413 
1414 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1415                               const ppc_hash_pte64_t *hptes,
1416                               hwaddr ptex, int n)
1417 {
1418     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1419 
1420     if (!spapr->htab) {
1421         g_free((void *)hptes);
1422     }
1423 
1424     /* Nothing to do for qemu managed HPT */
1425 }
1426 
1427 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1428                       uint64_t pte0, uint64_t pte1)
1429 {
1430     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1431     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1432 
1433     if (!spapr->htab) {
1434         kvmppc_write_hpte(ptex, pte0, pte1);
1435     } else {
1436         if (pte0 & HPTE64_V_VALID) {
1437             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1438             /*
1439              * When setting valid, we write PTE1 first. This ensures
1440              * proper synchronization with the reading code in
1441              * ppc_hash64_pteg_search()
1442              */
1443             smp_wmb();
1444             stq_p(spapr->htab + offset, pte0);
1445         } else {
1446             stq_p(spapr->htab + offset, pte0);
1447             /*
1448              * When clearing it we set PTE0 first. This ensures proper
1449              * synchronization with the reading code in
1450              * ppc_hash64_pteg_search()
1451              */
1452             smp_wmb();
1453             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1454         }
1455     }
1456 }
1457 
1458 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1459                              uint64_t pte1)
1460 {
1461     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1462     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1463 
1464     if (!spapr->htab) {
1465         /* There should always be a hash table when this is called */
1466         error_report("spapr_hpte_set_c called with no hash table !");
1467         return;
1468     }
1469 
1470     /* The HW performs a non-atomic byte update */
1471     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1472 }
1473 
1474 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1475                              uint64_t pte1)
1476 {
1477     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1478     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1479 
1480     if (!spapr->htab) {
1481         /* There should always be a hash table when this is called */
1482         error_report("spapr_hpte_set_r called with no hash table !");
1483         return;
1484     }
1485 
1486     /* The HW performs a non-atomic byte update */
1487     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1488 }
1489 
1490 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1491 {
1492     int shift;
1493 
1494     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1495      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1496      * that's much more than is needed for Linux guests */
1497     shift = ctz64(pow2ceil(ramsize)) - 7;
1498     shift = MAX(shift, 18); /* Minimum architected size */
1499     shift = MIN(shift, 46); /* Maximum architected size */
1500     return shift;
1501 }
1502 
1503 void spapr_free_hpt(SpaprMachineState *spapr)
1504 {
1505     g_free(spapr->htab);
1506     spapr->htab = NULL;
1507     spapr->htab_shift = 0;
1508     close_htab_fd(spapr);
1509 }
1510 
1511 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1512                           Error **errp)
1513 {
1514     long rc;
1515 
1516     /* Clean up any HPT info from a previous boot */
1517     spapr_free_hpt(spapr);
1518 
1519     rc = kvmppc_reset_htab(shift);
1520     if (rc < 0) {
1521         /* kernel-side HPT needed, but couldn't allocate one */
1522         error_setg_errno(errp, errno,
1523                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1524                          shift);
1525         /* This is almost certainly fatal, but if the caller really
1526          * wants to carry on with shift == 0, it's welcome to try */
1527     } else if (rc > 0) {
1528         /* kernel-side HPT allocated */
1529         if (rc != shift) {
1530             error_setg(errp,
1531                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1532                        shift, rc);
1533         }
1534 
1535         spapr->htab_shift = shift;
1536         spapr->htab = NULL;
1537     } else {
1538         /* kernel-side HPT not needed, allocate in userspace instead */
1539         size_t size = 1ULL << shift;
1540         int i;
1541 
1542         spapr->htab = qemu_memalign(size, size);
1543         if (!spapr->htab) {
1544             error_setg_errno(errp, errno,
1545                              "Could not allocate HPT of order %d", shift);
1546             return;
1547         }
1548 
1549         memset(spapr->htab, 0, size);
1550         spapr->htab_shift = shift;
1551 
1552         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1553             DIRTY_HPTE(HPTE(spapr->htab, i));
1554         }
1555     }
1556     /* We're setting up a hash table, so that means we're not radix */
1557     spapr->patb_entry = 0;
1558     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1559 }
1560 
1561 void spapr_setup_hpt(SpaprMachineState *spapr)
1562 {
1563     int hpt_shift;
1564 
1565     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1566         || (spapr->cas_reboot
1567             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1568         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1569     } else {
1570         uint64_t current_ram_size;
1571 
1572         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1573         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1574     }
1575     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1576 
1577     if (kvm_enabled()) {
1578         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1579 
1580         /* Check our RMA fits in the possible VRMA */
1581         if (vrma_limit < spapr->rma_size) {
1582             error_report("Unable to create %" HWADDR_PRIu
1583                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1584                          spapr->rma_size / MiB, vrma_limit / MiB);
1585             exit(EXIT_FAILURE);
1586         }
1587     }
1588 }
1589 
1590 static int spapr_reset_drcs(Object *child, void *opaque)
1591 {
1592     SpaprDrc *drc =
1593         (SpaprDrc *) object_dynamic_cast(child,
1594                                                  TYPE_SPAPR_DR_CONNECTOR);
1595 
1596     if (drc) {
1597         spapr_drc_reset(drc);
1598     }
1599 
1600     return 0;
1601 }
1602 
1603 static void spapr_machine_reset(MachineState *machine)
1604 {
1605     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1606     PowerPCCPU *first_ppc_cpu;
1607     hwaddr fdt_addr;
1608     void *fdt;
1609     int rc;
1610 
1611     kvmppc_svm_off(&error_fatal);
1612     spapr_caps_apply(spapr);
1613 
1614     first_ppc_cpu = POWERPC_CPU(first_cpu);
1615     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1616         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1617                               spapr->max_compat_pvr)) {
1618         /*
1619          * If using KVM with radix mode available, VCPUs can be started
1620          * without a HPT because KVM will start them in radix mode.
1621          * Set the GR bit in PATE so that we know there is no HPT.
1622          */
1623         spapr->patb_entry = PATE1_GR;
1624         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1625     } else {
1626         spapr_setup_hpt(spapr);
1627     }
1628 
1629     qemu_devices_reset();
1630 
1631     /*
1632      * If this reset wasn't generated by CAS, we should reset our
1633      * negotiated options and start from scratch
1634      */
1635     if (!spapr->cas_reboot) {
1636         spapr_ovec_cleanup(spapr->ov5_cas);
1637         spapr->ov5_cas = spapr_ovec_new();
1638 
1639         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1640     }
1641 
1642     /*
1643      * This is fixing some of the default configuration of the XIVE
1644      * devices. To be called after the reset of the machine devices.
1645      */
1646     spapr_irq_reset(spapr, &error_fatal);
1647 
1648     /*
1649      * There is no CAS under qtest. Simulate one to please the code that
1650      * depends on spapr->ov5_cas. This is especially needed to test device
1651      * unplug, so we do that before resetting the DRCs.
1652      */
1653     if (qtest_enabled()) {
1654         spapr_ovec_cleanup(spapr->ov5_cas);
1655         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1656     }
1657 
1658     /* DRC reset may cause a device to be unplugged. This will cause troubles
1659      * if this device is used by another device (eg, a running vhost backend
1660      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1661      * situations, we reset DRCs after all devices have been reset.
1662      */
1663     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1664 
1665     spapr_clear_pending_events(spapr);
1666 
1667     /*
1668      * We place the device tree and RTAS just below either the top of the RMA,
1669      * or just below 2GB, whichever is lower, so that it can be
1670      * processed with 32-bit real mode code if necessary
1671      */
1672     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1673 
1674     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1675 
1676     rc = fdt_pack(fdt);
1677 
1678     /* Should only fail if we've built a corrupted tree */
1679     assert(rc == 0);
1680 
1681     /* Load the fdt */
1682     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1683     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1684     g_free(spapr->fdt_blob);
1685     spapr->fdt_size = fdt_totalsize(fdt);
1686     spapr->fdt_initial_size = spapr->fdt_size;
1687     spapr->fdt_blob = fdt;
1688 
1689     /* Set up the entry state */
1690     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1691     first_ppc_cpu->env.gpr[5] = 0;
1692 
1693     spapr->cas_reboot = false;
1694 
1695     spapr->mc_status = -1;
1696     spapr->guest_machine_check_addr = -1;
1697 
1698     /* Signal all vCPUs waiting on this condition */
1699     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1700 
1701     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1702 }
1703 
1704 static void spapr_create_nvram(SpaprMachineState *spapr)
1705 {
1706     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1707     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1708 
1709     if (dinfo) {
1710         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1711                             &error_fatal);
1712     }
1713 
1714     qdev_init_nofail(dev);
1715 
1716     spapr->nvram = (struct SpaprNvram *)dev;
1717 }
1718 
1719 static void spapr_rtc_create(SpaprMachineState *spapr)
1720 {
1721     object_initialize_child(OBJECT(spapr), "rtc",
1722                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1723                             &error_fatal, NULL);
1724     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1725                               &error_fatal);
1726     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1727                               "date", &error_fatal);
1728 }
1729 
1730 /* Returns whether we want to use VGA or not */
1731 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1732 {
1733     switch (vga_interface_type) {
1734     case VGA_NONE:
1735         return false;
1736     case VGA_DEVICE:
1737         return true;
1738     case VGA_STD:
1739     case VGA_VIRTIO:
1740     case VGA_CIRRUS:
1741         return pci_vga_init(pci_bus) != NULL;
1742     default:
1743         error_setg(errp,
1744                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1745         return false;
1746     }
1747 }
1748 
1749 static int spapr_pre_load(void *opaque)
1750 {
1751     int rc;
1752 
1753     rc = spapr_caps_pre_load(opaque);
1754     if (rc) {
1755         return rc;
1756     }
1757 
1758     return 0;
1759 }
1760 
1761 static int spapr_post_load(void *opaque, int version_id)
1762 {
1763     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1764     int err = 0;
1765 
1766     err = spapr_caps_post_migration(spapr);
1767     if (err) {
1768         return err;
1769     }
1770 
1771     /*
1772      * In earlier versions, there was no separate qdev for the PAPR
1773      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1774      * So when migrating from those versions, poke the incoming offset
1775      * value into the RTC device
1776      */
1777     if (version_id < 3) {
1778         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1779         if (err) {
1780             return err;
1781         }
1782     }
1783 
1784     if (kvm_enabled() && spapr->patb_entry) {
1785         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1786         bool radix = !!(spapr->patb_entry & PATE1_GR);
1787         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1788 
1789         /*
1790          * Update LPCR:HR and UPRT as they may not be set properly in
1791          * the stream
1792          */
1793         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1794                             LPCR_HR | LPCR_UPRT);
1795 
1796         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1797         if (err) {
1798             error_report("Process table config unsupported by the host");
1799             return -EINVAL;
1800         }
1801     }
1802 
1803     err = spapr_irq_post_load(spapr, version_id);
1804     if (err) {
1805         return err;
1806     }
1807 
1808     return err;
1809 }
1810 
1811 static int spapr_pre_save(void *opaque)
1812 {
1813     int rc;
1814 
1815     rc = spapr_caps_pre_save(opaque);
1816     if (rc) {
1817         return rc;
1818     }
1819 
1820     return 0;
1821 }
1822 
1823 static bool version_before_3(void *opaque, int version_id)
1824 {
1825     return version_id < 3;
1826 }
1827 
1828 static bool spapr_pending_events_needed(void *opaque)
1829 {
1830     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1831     return !QTAILQ_EMPTY(&spapr->pending_events);
1832 }
1833 
1834 static const VMStateDescription vmstate_spapr_event_entry = {
1835     .name = "spapr_event_log_entry",
1836     .version_id = 1,
1837     .minimum_version_id = 1,
1838     .fields = (VMStateField[]) {
1839         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1840         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1841         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1842                                      NULL, extended_length),
1843         VMSTATE_END_OF_LIST()
1844     },
1845 };
1846 
1847 static const VMStateDescription vmstate_spapr_pending_events = {
1848     .name = "spapr_pending_events",
1849     .version_id = 1,
1850     .minimum_version_id = 1,
1851     .needed = spapr_pending_events_needed,
1852     .fields = (VMStateField[]) {
1853         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1854                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1855         VMSTATE_END_OF_LIST()
1856     },
1857 };
1858 
1859 static bool spapr_ov5_cas_needed(void *opaque)
1860 {
1861     SpaprMachineState *spapr = opaque;
1862     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1863     bool cas_needed;
1864 
1865     /* Prior to the introduction of SpaprOptionVector, we had two option
1866      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1867      * Both of these options encode machine topology into the device-tree
1868      * in such a way that the now-booted OS should still be able to interact
1869      * appropriately with QEMU regardless of what options were actually
1870      * negotiatied on the source side.
1871      *
1872      * As such, we can avoid migrating the CAS-negotiated options if these
1873      * are the only options available on the current machine/platform.
1874      * Since these are the only options available for pseries-2.7 and
1875      * earlier, this allows us to maintain old->new/new->old migration
1876      * compatibility.
1877      *
1878      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1879      * via default pseries-2.8 machines and explicit command-line parameters.
1880      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1881      * of the actual CAS-negotiated values to continue working properly. For
1882      * example, availability of memory unplug depends on knowing whether
1883      * OV5_HP_EVT was negotiated via CAS.
1884      *
1885      * Thus, for any cases where the set of available CAS-negotiatable
1886      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1887      * include the CAS-negotiated options in the migration stream, unless
1888      * if they affect boot time behaviour only.
1889      */
1890     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1891     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1892     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1893 
1894     /* We need extra information if we have any bits outside the mask
1895      * defined above */
1896     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1897 
1898     spapr_ovec_cleanup(ov5_mask);
1899 
1900     return cas_needed;
1901 }
1902 
1903 static const VMStateDescription vmstate_spapr_ov5_cas = {
1904     .name = "spapr_option_vector_ov5_cas",
1905     .version_id = 1,
1906     .minimum_version_id = 1,
1907     .needed = spapr_ov5_cas_needed,
1908     .fields = (VMStateField[]) {
1909         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1910                                  vmstate_spapr_ovec, SpaprOptionVector),
1911         VMSTATE_END_OF_LIST()
1912     },
1913 };
1914 
1915 static bool spapr_patb_entry_needed(void *opaque)
1916 {
1917     SpaprMachineState *spapr = opaque;
1918 
1919     return !!spapr->patb_entry;
1920 }
1921 
1922 static const VMStateDescription vmstate_spapr_patb_entry = {
1923     .name = "spapr_patb_entry",
1924     .version_id = 1,
1925     .minimum_version_id = 1,
1926     .needed = spapr_patb_entry_needed,
1927     .fields = (VMStateField[]) {
1928         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1929         VMSTATE_END_OF_LIST()
1930     },
1931 };
1932 
1933 static bool spapr_irq_map_needed(void *opaque)
1934 {
1935     SpaprMachineState *spapr = opaque;
1936 
1937     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1938 }
1939 
1940 static const VMStateDescription vmstate_spapr_irq_map = {
1941     .name = "spapr_irq_map",
1942     .version_id = 1,
1943     .minimum_version_id = 1,
1944     .needed = spapr_irq_map_needed,
1945     .fields = (VMStateField[]) {
1946         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1947         VMSTATE_END_OF_LIST()
1948     },
1949 };
1950 
1951 static bool spapr_dtb_needed(void *opaque)
1952 {
1953     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1954 
1955     return smc->update_dt_enabled;
1956 }
1957 
1958 static int spapr_dtb_pre_load(void *opaque)
1959 {
1960     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1961 
1962     g_free(spapr->fdt_blob);
1963     spapr->fdt_blob = NULL;
1964     spapr->fdt_size = 0;
1965 
1966     return 0;
1967 }
1968 
1969 static const VMStateDescription vmstate_spapr_dtb = {
1970     .name = "spapr_dtb",
1971     .version_id = 1,
1972     .minimum_version_id = 1,
1973     .needed = spapr_dtb_needed,
1974     .pre_load = spapr_dtb_pre_load,
1975     .fields = (VMStateField[]) {
1976         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1977         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1978         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1979                                      fdt_size),
1980         VMSTATE_END_OF_LIST()
1981     },
1982 };
1983 
1984 static bool spapr_fwnmi_needed(void *opaque)
1985 {
1986     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1987 
1988     return spapr->guest_machine_check_addr != -1;
1989 }
1990 
1991 static int spapr_fwnmi_pre_save(void *opaque)
1992 {
1993     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1994 
1995     /*
1996      * Check if machine check handling is in progress and print a
1997      * warning message.
1998      */
1999     if (spapr->mc_status != -1) {
2000         warn_report("A machine check is being handled during migration. The"
2001                 "handler may run and log hardware error on the destination");
2002     }
2003 
2004     return 0;
2005 }
2006 
2007 static const VMStateDescription vmstate_spapr_machine_check = {
2008     .name = "spapr_machine_check",
2009     .version_id = 1,
2010     .minimum_version_id = 1,
2011     .needed = spapr_fwnmi_needed,
2012     .pre_save = spapr_fwnmi_pre_save,
2013     .fields = (VMStateField[]) {
2014         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2015         VMSTATE_INT32(mc_status, SpaprMachineState),
2016         VMSTATE_END_OF_LIST()
2017     },
2018 };
2019 
2020 static const VMStateDescription vmstate_spapr = {
2021     .name = "spapr",
2022     .version_id = 3,
2023     .minimum_version_id = 1,
2024     .pre_load = spapr_pre_load,
2025     .post_load = spapr_post_load,
2026     .pre_save = spapr_pre_save,
2027     .fields = (VMStateField[]) {
2028         /* used to be @next_irq */
2029         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2030 
2031         /* RTC offset */
2032         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2033 
2034         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2035         VMSTATE_END_OF_LIST()
2036     },
2037     .subsections = (const VMStateDescription*[]) {
2038         &vmstate_spapr_ov5_cas,
2039         &vmstate_spapr_patb_entry,
2040         &vmstate_spapr_pending_events,
2041         &vmstate_spapr_cap_htm,
2042         &vmstate_spapr_cap_vsx,
2043         &vmstate_spapr_cap_dfp,
2044         &vmstate_spapr_cap_cfpc,
2045         &vmstate_spapr_cap_sbbc,
2046         &vmstate_spapr_cap_ibs,
2047         &vmstate_spapr_cap_hpt_maxpagesize,
2048         &vmstate_spapr_irq_map,
2049         &vmstate_spapr_cap_nested_kvm_hv,
2050         &vmstate_spapr_dtb,
2051         &vmstate_spapr_cap_large_decr,
2052         &vmstate_spapr_cap_ccf_assist,
2053         &vmstate_spapr_cap_fwnmi,
2054         &vmstate_spapr_machine_check,
2055         NULL
2056     }
2057 };
2058 
2059 static int htab_save_setup(QEMUFile *f, void *opaque)
2060 {
2061     SpaprMachineState *spapr = opaque;
2062 
2063     /* "Iteration" header */
2064     if (!spapr->htab_shift) {
2065         qemu_put_be32(f, -1);
2066     } else {
2067         qemu_put_be32(f, spapr->htab_shift);
2068     }
2069 
2070     if (spapr->htab) {
2071         spapr->htab_save_index = 0;
2072         spapr->htab_first_pass = true;
2073     } else {
2074         if (spapr->htab_shift) {
2075             assert(kvm_enabled());
2076         }
2077     }
2078 
2079 
2080     return 0;
2081 }
2082 
2083 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2084                             int chunkstart, int n_valid, int n_invalid)
2085 {
2086     qemu_put_be32(f, chunkstart);
2087     qemu_put_be16(f, n_valid);
2088     qemu_put_be16(f, n_invalid);
2089     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2090                     HASH_PTE_SIZE_64 * n_valid);
2091 }
2092 
2093 static void htab_save_end_marker(QEMUFile *f)
2094 {
2095     qemu_put_be32(f, 0);
2096     qemu_put_be16(f, 0);
2097     qemu_put_be16(f, 0);
2098 }
2099 
2100 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2101                                  int64_t max_ns)
2102 {
2103     bool has_timeout = max_ns != -1;
2104     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2105     int index = spapr->htab_save_index;
2106     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2107 
2108     assert(spapr->htab_first_pass);
2109 
2110     do {
2111         int chunkstart;
2112 
2113         /* Consume invalid HPTEs */
2114         while ((index < htabslots)
2115                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2116             CLEAN_HPTE(HPTE(spapr->htab, index));
2117             index++;
2118         }
2119 
2120         /* Consume valid HPTEs */
2121         chunkstart = index;
2122         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2123                && HPTE_VALID(HPTE(spapr->htab, index))) {
2124             CLEAN_HPTE(HPTE(spapr->htab, index));
2125             index++;
2126         }
2127 
2128         if (index > chunkstart) {
2129             int n_valid = index - chunkstart;
2130 
2131             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2132 
2133             if (has_timeout &&
2134                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2135                 break;
2136             }
2137         }
2138     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2139 
2140     if (index >= htabslots) {
2141         assert(index == htabslots);
2142         index = 0;
2143         spapr->htab_first_pass = false;
2144     }
2145     spapr->htab_save_index = index;
2146 }
2147 
2148 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2149                                 int64_t max_ns)
2150 {
2151     bool final = max_ns < 0;
2152     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2153     int examined = 0, sent = 0;
2154     int index = spapr->htab_save_index;
2155     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2156 
2157     assert(!spapr->htab_first_pass);
2158 
2159     do {
2160         int chunkstart, invalidstart;
2161 
2162         /* Consume non-dirty HPTEs */
2163         while ((index < htabslots)
2164                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2165             index++;
2166             examined++;
2167         }
2168 
2169         chunkstart = index;
2170         /* Consume valid dirty HPTEs */
2171         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2172                && HPTE_DIRTY(HPTE(spapr->htab, index))
2173                && HPTE_VALID(HPTE(spapr->htab, index))) {
2174             CLEAN_HPTE(HPTE(spapr->htab, index));
2175             index++;
2176             examined++;
2177         }
2178 
2179         invalidstart = index;
2180         /* Consume invalid dirty HPTEs */
2181         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2182                && HPTE_DIRTY(HPTE(spapr->htab, index))
2183                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2184             CLEAN_HPTE(HPTE(spapr->htab, index));
2185             index++;
2186             examined++;
2187         }
2188 
2189         if (index > chunkstart) {
2190             int n_valid = invalidstart - chunkstart;
2191             int n_invalid = index - invalidstart;
2192 
2193             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2194             sent += index - chunkstart;
2195 
2196             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2197                 break;
2198             }
2199         }
2200 
2201         if (examined >= htabslots) {
2202             break;
2203         }
2204 
2205         if (index >= htabslots) {
2206             assert(index == htabslots);
2207             index = 0;
2208         }
2209     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2210 
2211     if (index >= htabslots) {
2212         assert(index == htabslots);
2213         index = 0;
2214     }
2215 
2216     spapr->htab_save_index = index;
2217 
2218     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2219 }
2220 
2221 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2222 #define MAX_KVM_BUF_SIZE    2048
2223 
2224 static int htab_save_iterate(QEMUFile *f, void *opaque)
2225 {
2226     SpaprMachineState *spapr = opaque;
2227     int fd;
2228     int rc = 0;
2229 
2230     /* Iteration header */
2231     if (!spapr->htab_shift) {
2232         qemu_put_be32(f, -1);
2233         return 1;
2234     } else {
2235         qemu_put_be32(f, 0);
2236     }
2237 
2238     if (!spapr->htab) {
2239         assert(kvm_enabled());
2240 
2241         fd = get_htab_fd(spapr);
2242         if (fd < 0) {
2243             return fd;
2244         }
2245 
2246         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2247         if (rc < 0) {
2248             return rc;
2249         }
2250     } else  if (spapr->htab_first_pass) {
2251         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2252     } else {
2253         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2254     }
2255 
2256     htab_save_end_marker(f);
2257 
2258     return rc;
2259 }
2260 
2261 static int htab_save_complete(QEMUFile *f, void *opaque)
2262 {
2263     SpaprMachineState *spapr = opaque;
2264     int fd;
2265 
2266     /* Iteration header */
2267     if (!spapr->htab_shift) {
2268         qemu_put_be32(f, -1);
2269         return 0;
2270     } else {
2271         qemu_put_be32(f, 0);
2272     }
2273 
2274     if (!spapr->htab) {
2275         int rc;
2276 
2277         assert(kvm_enabled());
2278 
2279         fd = get_htab_fd(spapr);
2280         if (fd < 0) {
2281             return fd;
2282         }
2283 
2284         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2285         if (rc < 0) {
2286             return rc;
2287         }
2288     } else {
2289         if (spapr->htab_first_pass) {
2290             htab_save_first_pass(f, spapr, -1);
2291         }
2292         htab_save_later_pass(f, spapr, -1);
2293     }
2294 
2295     /* End marker */
2296     htab_save_end_marker(f);
2297 
2298     return 0;
2299 }
2300 
2301 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2302 {
2303     SpaprMachineState *spapr = opaque;
2304     uint32_t section_hdr;
2305     int fd = -1;
2306     Error *local_err = NULL;
2307 
2308     if (version_id < 1 || version_id > 1) {
2309         error_report("htab_load() bad version");
2310         return -EINVAL;
2311     }
2312 
2313     section_hdr = qemu_get_be32(f);
2314 
2315     if (section_hdr == -1) {
2316         spapr_free_hpt(spapr);
2317         return 0;
2318     }
2319 
2320     if (section_hdr) {
2321         /* First section gives the htab size */
2322         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2323         if (local_err) {
2324             error_report_err(local_err);
2325             return -EINVAL;
2326         }
2327         return 0;
2328     }
2329 
2330     if (!spapr->htab) {
2331         assert(kvm_enabled());
2332 
2333         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2334         if (fd < 0) {
2335             error_report_err(local_err);
2336             return fd;
2337         }
2338     }
2339 
2340     while (true) {
2341         uint32_t index;
2342         uint16_t n_valid, n_invalid;
2343 
2344         index = qemu_get_be32(f);
2345         n_valid = qemu_get_be16(f);
2346         n_invalid = qemu_get_be16(f);
2347 
2348         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2349             /* End of Stream */
2350             break;
2351         }
2352 
2353         if ((index + n_valid + n_invalid) >
2354             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2355             /* Bad index in stream */
2356             error_report(
2357                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2358                 index, n_valid, n_invalid, spapr->htab_shift);
2359             return -EINVAL;
2360         }
2361 
2362         if (spapr->htab) {
2363             if (n_valid) {
2364                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2365                                 HASH_PTE_SIZE_64 * n_valid);
2366             }
2367             if (n_invalid) {
2368                 memset(HPTE(spapr->htab, index + n_valid), 0,
2369                        HASH_PTE_SIZE_64 * n_invalid);
2370             }
2371         } else {
2372             int rc;
2373 
2374             assert(fd >= 0);
2375 
2376             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2377             if (rc < 0) {
2378                 return rc;
2379             }
2380         }
2381     }
2382 
2383     if (!spapr->htab) {
2384         assert(fd >= 0);
2385         close(fd);
2386     }
2387 
2388     return 0;
2389 }
2390 
2391 static void htab_save_cleanup(void *opaque)
2392 {
2393     SpaprMachineState *spapr = opaque;
2394 
2395     close_htab_fd(spapr);
2396 }
2397 
2398 static SaveVMHandlers savevm_htab_handlers = {
2399     .save_setup = htab_save_setup,
2400     .save_live_iterate = htab_save_iterate,
2401     .save_live_complete_precopy = htab_save_complete,
2402     .save_cleanup = htab_save_cleanup,
2403     .load_state = htab_load,
2404 };
2405 
2406 static void spapr_boot_set(void *opaque, const char *boot_device,
2407                            Error **errp)
2408 {
2409     MachineState *machine = MACHINE(opaque);
2410     machine->boot_order = g_strdup(boot_device);
2411 }
2412 
2413 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2414 {
2415     MachineState *machine = MACHINE(spapr);
2416     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2417     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2418     int i;
2419 
2420     for (i = 0; i < nr_lmbs; i++) {
2421         uint64_t addr;
2422 
2423         addr = i * lmb_size + machine->device_memory->base;
2424         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2425                                addr / lmb_size);
2426     }
2427 }
2428 
2429 /*
2430  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2431  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2432  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2433  */
2434 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2435 {
2436     int i;
2437 
2438     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2439         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2440                    " is not aligned to %" PRIu64 " MiB",
2441                    machine->ram_size,
2442                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2443         return;
2444     }
2445 
2446     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2447         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2448                    " is not aligned to %" PRIu64 " MiB",
2449                    machine->ram_size,
2450                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2451         return;
2452     }
2453 
2454     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2455         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2456             error_setg(errp,
2457                        "Node %d memory size 0x%" PRIx64
2458                        " is not aligned to %" PRIu64 " MiB",
2459                        i, machine->numa_state->nodes[i].node_mem,
2460                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2461             return;
2462         }
2463     }
2464 }
2465 
2466 /* find cpu slot in machine->possible_cpus by core_id */
2467 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2468 {
2469     int index = id / ms->smp.threads;
2470 
2471     if (index >= ms->possible_cpus->len) {
2472         return NULL;
2473     }
2474     if (idx) {
2475         *idx = index;
2476     }
2477     return &ms->possible_cpus->cpus[index];
2478 }
2479 
2480 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2481 {
2482     MachineState *ms = MACHINE(spapr);
2483     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2484     Error *local_err = NULL;
2485     bool vsmt_user = !!spapr->vsmt;
2486     int kvm_smt = kvmppc_smt_threads();
2487     int ret;
2488     unsigned int smp_threads = ms->smp.threads;
2489 
2490     if (!kvm_enabled() && (smp_threads > 1)) {
2491         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2492                      "on a pseries machine");
2493         goto out;
2494     }
2495     if (!is_power_of_2(smp_threads)) {
2496         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2497                      "machine because it must be a power of 2", smp_threads);
2498         goto out;
2499     }
2500 
2501     /* Detemine the VSMT mode to use: */
2502     if (vsmt_user) {
2503         if (spapr->vsmt < smp_threads) {
2504             error_setg(&local_err, "Cannot support VSMT mode %d"
2505                          " because it must be >= threads/core (%d)",
2506                          spapr->vsmt, smp_threads);
2507             goto out;
2508         }
2509         /* In this case, spapr->vsmt has been set by the command line */
2510     } else if (!smc->smp_threads_vsmt) {
2511         /*
2512          * Default VSMT value is tricky, because we need it to be as
2513          * consistent as possible (for migration), but this requires
2514          * changing it for at least some existing cases.  We pick 8 as
2515          * the value that we'd get with KVM on POWER8, the
2516          * overwhelmingly common case in production systems.
2517          */
2518         spapr->vsmt = MAX(8, smp_threads);
2519     } else {
2520         spapr->vsmt = smp_threads;
2521     }
2522 
2523     /* KVM: If necessary, set the SMT mode: */
2524     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2525         ret = kvmppc_set_smt_threads(spapr->vsmt);
2526         if (ret) {
2527             /* Looks like KVM isn't able to change VSMT mode */
2528             error_setg(&local_err,
2529                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2530                        spapr->vsmt, ret);
2531             /* We can live with that if the default one is big enough
2532              * for the number of threads, and a submultiple of the one
2533              * we want.  In this case we'll waste some vcpu ids, but
2534              * behaviour will be correct */
2535             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2536                 warn_report_err(local_err);
2537                 local_err = NULL;
2538                 goto out;
2539             } else {
2540                 if (!vsmt_user) {
2541                     error_append_hint(&local_err,
2542                                       "On PPC, a VM with %d threads/core"
2543                                       " on a host with %d threads/core"
2544                                       " requires the use of VSMT mode %d.\n",
2545                                       smp_threads, kvm_smt, spapr->vsmt);
2546                 }
2547                 kvmppc_error_append_smt_possible_hint(&local_err);
2548                 goto out;
2549             }
2550         }
2551     }
2552     /* else TCG: nothing to do currently */
2553 out:
2554     error_propagate(errp, local_err);
2555 }
2556 
2557 static void spapr_init_cpus(SpaprMachineState *spapr)
2558 {
2559     MachineState *machine = MACHINE(spapr);
2560     MachineClass *mc = MACHINE_GET_CLASS(machine);
2561     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2562     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2563     const CPUArchIdList *possible_cpus;
2564     unsigned int smp_cpus = machine->smp.cpus;
2565     unsigned int smp_threads = machine->smp.threads;
2566     unsigned int max_cpus = machine->smp.max_cpus;
2567     int boot_cores_nr = smp_cpus / smp_threads;
2568     int i;
2569 
2570     possible_cpus = mc->possible_cpu_arch_ids(machine);
2571     if (mc->has_hotpluggable_cpus) {
2572         if (smp_cpus % smp_threads) {
2573             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2574                          smp_cpus, smp_threads);
2575             exit(1);
2576         }
2577         if (max_cpus % smp_threads) {
2578             error_report("max_cpus (%u) must be multiple of threads (%u)",
2579                          max_cpus, smp_threads);
2580             exit(1);
2581         }
2582     } else {
2583         if (max_cpus != smp_cpus) {
2584             error_report("This machine version does not support CPU hotplug");
2585             exit(1);
2586         }
2587         boot_cores_nr = possible_cpus->len;
2588     }
2589 
2590     if (smc->pre_2_10_has_unused_icps) {
2591         int i;
2592 
2593         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2594             /* Dummy entries get deregistered when real ICPState objects
2595              * are registered during CPU core hotplug.
2596              */
2597             pre_2_10_vmstate_register_dummy_icp(i);
2598         }
2599     }
2600 
2601     for (i = 0; i < possible_cpus->len; i++) {
2602         int core_id = i * smp_threads;
2603 
2604         if (mc->has_hotpluggable_cpus) {
2605             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2606                                    spapr_vcpu_id(spapr, core_id));
2607         }
2608 
2609         if (i < boot_cores_nr) {
2610             Object *core  = object_new(type);
2611             int nr_threads = smp_threads;
2612 
2613             /* Handle the partially filled core for older machine types */
2614             if ((i + 1) * smp_threads >= smp_cpus) {
2615                 nr_threads = smp_cpus - i * smp_threads;
2616             }
2617 
2618             object_property_set_int(core, nr_threads, "nr-threads",
2619                                     &error_fatal);
2620             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2621                                     &error_fatal);
2622             object_property_set_bool(core, true, "realized", &error_fatal);
2623 
2624             object_unref(core);
2625         }
2626     }
2627 }
2628 
2629 static PCIHostState *spapr_create_default_phb(void)
2630 {
2631     DeviceState *dev;
2632 
2633     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2634     qdev_prop_set_uint32(dev, "index", 0);
2635     qdev_init_nofail(dev);
2636 
2637     return PCI_HOST_BRIDGE(dev);
2638 }
2639 
2640 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2641 {
2642     MachineState *machine = MACHINE(spapr);
2643     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2644     hwaddr rma_size = machine->ram_size;
2645     hwaddr node0_size = spapr_node0_size(machine);
2646 
2647     /* RMA has to fit in the first NUMA node */
2648     rma_size = MIN(rma_size, node0_size);
2649 
2650     /*
2651      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2652      * never exceed that
2653      */
2654     rma_size = MIN(rma_size, 1 * TiB);
2655 
2656     /*
2657      * Clamp the RMA size based on machine type.  This is for
2658      * migration compatibility with older qemu versions, which limited
2659      * the RMA size for complicated and mostly bad reasons.
2660      */
2661     if (smc->rma_limit) {
2662         rma_size = MIN(rma_size, smc->rma_limit);
2663     }
2664 
2665     if (rma_size < MIN_RMA_SLOF) {
2666         error_setg(errp,
2667                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2668                    "ldMiB guest RMA (Real Mode Area memory)",
2669                    MIN_RMA_SLOF / MiB);
2670         return 0;
2671     }
2672 
2673     return rma_size;
2674 }
2675 
2676 /* pSeries LPAR / sPAPR hardware init */
2677 static void spapr_machine_init(MachineState *machine)
2678 {
2679     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2680     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2681     MachineClass *mc = MACHINE_GET_CLASS(machine);
2682     const char *kernel_filename = machine->kernel_filename;
2683     const char *initrd_filename = machine->initrd_filename;
2684     PCIHostState *phb;
2685     int i;
2686     MemoryRegion *sysmem = get_system_memory();
2687     long load_limit, fw_size;
2688     char *filename;
2689     Error *resize_hpt_err = NULL;
2690 
2691     msi_nonbroken = true;
2692 
2693     QLIST_INIT(&spapr->phbs);
2694     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2695 
2696     /* Determine capabilities to run with */
2697     spapr_caps_init(spapr);
2698 
2699     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2700     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2701         /*
2702          * If the user explicitly requested a mode we should either
2703          * supply it, or fail completely (which we do below).  But if
2704          * it's not set explicitly, we reset our mode to something
2705          * that works
2706          */
2707         if (resize_hpt_err) {
2708             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2709             error_free(resize_hpt_err);
2710             resize_hpt_err = NULL;
2711         } else {
2712             spapr->resize_hpt = smc->resize_hpt_default;
2713         }
2714     }
2715 
2716     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2717 
2718     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2719         /*
2720          * User requested HPT resize, but this host can't supply it.  Bail out
2721          */
2722         error_report_err(resize_hpt_err);
2723         exit(1);
2724     }
2725 
2726     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2727 
2728     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2729     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2730 
2731     /*
2732      * VSMT must be set in order to be able to compute VCPU ids, ie to
2733      * call spapr_max_server_number() or spapr_vcpu_id().
2734      */
2735     spapr_set_vsmt_mode(spapr, &error_fatal);
2736 
2737     /* Set up Interrupt Controller before we create the VCPUs */
2738     spapr_irq_init(spapr, &error_fatal);
2739 
2740     /* Set up containers for ibm,client-architecture-support negotiated options
2741      */
2742     spapr->ov5 = spapr_ovec_new();
2743     spapr->ov5_cas = spapr_ovec_new();
2744 
2745     if (smc->dr_lmb_enabled) {
2746         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2747         spapr_validate_node_memory(machine, &error_fatal);
2748     }
2749 
2750     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2751 
2752     /* advertise support for dedicated HP event source to guests */
2753     if (spapr->use_hotplug_event_source) {
2754         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2755     }
2756 
2757     /* advertise support for HPT resizing */
2758     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2759         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2760     }
2761 
2762     /* advertise support for ibm,dyamic-memory-v2 */
2763     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2764 
2765     /* advertise XIVE on POWER9 machines */
2766     if (spapr->irq->xive) {
2767         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2768     }
2769 
2770     /* init CPUs */
2771     spapr_init_cpus(spapr);
2772 
2773     /*
2774      * check we don't have a memory-less/cpu-less NUMA node
2775      * Firmware relies on the existing memory/cpu topology to provide the
2776      * NUMA topology to the kernel.
2777      * And the linux kernel needs to know the NUMA topology at start
2778      * to be able to hotplug CPUs later.
2779      */
2780     if (machine->numa_state->num_nodes) {
2781         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2782             /* check for memory-less node */
2783             if (machine->numa_state->nodes[i].node_mem == 0) {
2784                 CPUState *cs;
2785                 int found = 0;
2786                 /* check for cpu-less node */
2787                 CPU_FOREACH(cs) {
2788                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2789                     if (cpu->node_id == i) {
2790                         found = 1;
2791                         break;
2792                     }
2793                 }
2794                 /* memory-less and cpu-less node */
2795                 if (!found) {
2796                     error_report(
2797                        "Memory-less/cpu-less nodes are not supported (node %d)",
2798                                  i);
2799                     exit(1);
2800                 }
2801             }
2802         }
2803 
2804     }
2805 
2806     /*
2807      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2808      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2809      * called from vPHB reset handler so we initialize the counter here.
2810      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2811      * must be equally distant from any other node.
2812      * The final value of spapr->gpu_numa_id is going to be written to
2813      * max-associativity-domains in spapr_build_fdt().
2814      */
2815     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2816 
2817     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2818         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2819                               spapr->max_compat_pvr)) {
2820         /* KVM and TCG always allow GTSE with radix... */
2821         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2822     }
2823     /* ... but not with hash (currently). */
2824 
2825     if (kvm_enabled()) {
2826         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2827         kvmppc_enable_logical_ci_hcalls();
2828         kvmppc_enable_set_mode_hcall();
2829 
2830         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2831         kvmppc_enable_clear_ref_mod_hcalls();
2832 
2833         /* Enable H_PAGE_INIT */
2834         kvmppc_enable_h_page_init();
2835     }
2836 
2837     /* map RAM */
2838     memory_region_add_subregion(sysmem, 0, machine->ram);
2839 
2840     /* always allocate the device memory information */
2841     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2842 
2843     /* initialize hotplug memory address space */
2844     if (machine->ram_size < machine->maxram_size) {
2845         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2846         /*
2847          * Limit the number of hotpluggable memory slots to half the number
2848          * slots that KVM supports, leaving the other half for PCI and other
2849          * devices. However ensure that number of slots doesn't drop below 32.
2850          */
2851         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2852                            SPAPR_MAX_RAM_SLOTS;
2853 
2854         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2855             max_memslots = SPAPR_MAX_RAM_SLOTS;
2856         }
2857         if (machine->ram_slots > max_memslots) {
2858             error_report("Specified number of memory slots %"
2859                          PRIu64" exceeds max supported %d",
2860                          machine->ram_slots, max_memslots);
2861             exit(1);
2862         }
2863 
2864         machine->device_memory->base = ROUND_UP(machine->ram_size,
2865                                                 SPAPR_DEVICE_MEM_ALIGN);
2866         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2867                            "device-memory", device_mem_size);
2868         memory_region_add_subregion(sysmem, machine->device_memory->base,
2869                                     &machine->device_memory->mr);
2870     }
2871 
2872     if (smc->dr_lmb_enabled) {
2873         spapr_create_lmb_dr_connectors(spapr);
2874     }
2875 
2876     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2877         /* Create the error string for live migration blocker */
2878         error_setg(&spapr->fwnmi_migration_blocker,
2879             "A machine check is being handled during migration. The handler"
2880             "may run and log hardware error on the destination");
2881     }
2882 
2883     if (mc->nvdimm_supported) {
2884         spapr_create_nvdimm_dr_connectors(spapr);
2885     }
2886 
2887     /* Set up RTAS event infrastructure */
2888     spapr_events_init(spapr);
2889 
2890     /* Set up the RTC RTAS interfaces */
2891     spapr_rtc_create(spapr);
2892 
2893     /* Set up VIO bus */
2894     spapr->vio_bus = spapr_vio_bus_init();
2895 
2896     for (i = 0; i < serial_max_hds(); i++) {
2897         if (serial_hd(i)) {
2898             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2899         }
2900     }
2901 
2902     /* We always have at least the nvram device on VIO */
2903     spapr_create_nvram(spapr);
2904 
2905     /*
2906      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2907      * connectors (described in root DT node's "ibm,drc-types" property)
2908      * are pre-initialized here. additional child connectors (such as
2909      * connectors for a PHBs PCI slots) are added as needed during their
2910      * parent's realization.
2911      */
2912     if (smc->dr_phb_enabled) {
2913         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2914             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2915         }
2916     }
2917 
2918     /* Set up PCI */
2919     spapr_pci_rtas_init();
2920 
2921     phb = spapr_create_default_phb();
2922 
2923     for (i = 0; i < nb_nics; i++) {
2924         NICInfo *nd = &nd_table[i];
2925 
2926         if (!nd->model) {
2927             nd->model = g_strdup("spapr-vlan");
2928         }
2929 
2930         if (g_str_equal(nd->model, "spapr-vlan") ||
2931             g_str_equal(nd->model, "ibmveth")) {
2932             spapr_vlan_create(spapr->vio_bus, nd);
2933         } else {
2934             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2935         }
2936     }
2937 
2938     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2939         spapr_vscsi_create(spapr->vio_bus);
2940     }
2941 
2942     /* Graphics */
2943     if (spapr_vga_init(phb->bus, &error_fatal)) {
2944         spapr->has_graphics = true;
2945         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2946     }
2947 
2948     if (machine->usb) {
2949         if (smc->use_ohci_by_default) {
2950             pci_create_simple(phb->bus, -1, "pci-ohci");
2951         } else {
2952             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2953         }
2954 
2955         if (spapr->has_graphics) {
2956             USBBus *usb_bus = usb_bus_find(-1);
2957 
2958             usb_create_simple(usb_bus, "usb-kbd");
2959             usb_create_simple(usb_bus, "usb-mouse");
2960         }
2961     }
2962 
2963     if (kernel_filename) {
2964         uint64_t lowaddr = 0;
2965 
2966         spapr->kernel_size = load_elf(kernel_filename, NULL,
2967                                       translate_kernel_address, spapr,
2968                                       NULL, &lowaddr, NULL, NULL, 1,
2969                                       PPC_ELF_MACHINE, 0, 0);
2970         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2971             spapr->kernel_size = load_elf(kernel_filename, NULL,
2972                                           translate_kernel_address, spapr, NULL,
2973                                           &lowaddr, NULL, NULL, 0,
2974                                           PPC_ELF_MACHINE,
2975                                           0, 0);
2976             spapr->kernel_le = spapr->kernel_size > 0;
2977         }
2978         if (spapr->kernel_size < 0) {
2979             error_report("error loading %s: %s", kernel_filename,
2980                          load_elf_strerror(spapr->kernel_size));
2981             exit(1);
2982         }
2983 
2984         /* load initrd */
2985         if (initrd_filename) {
2986             /* Try to locate the initrd in the gap between the kernel
2987              * and the firmware. Add a bit of space just in case
2988              */
2989             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2990                                   + 0x1ffff) & ~0xffff;
2991             spapr->initrd_size = load_image_targphys(initrd_filename,
2992                                                      spapr->initrd_base,
2993                                                      load_limit
2994                                                      - spapr->initrd_base);
2995             if (spapr->initrd_size < 0) {
2996                 error_report("could not load initial ram disk '%s'",
2997                              initrd_filename);
2998                 exit(1);
2999             }
3000         }
3001     }
3002 
3003     if (bios_name == NULL) {
3004         bios_name = FW_FILE_NAME;
3005     }
3006     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3007     if (!filename) {
3008         error_report("Could not find LPAR firmware '%s'", bios_name);
3009         exit(1);
3010     }
3011     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3012     if (fw_size <= 0) {
3013         error_report("Could not load LPAR firmware '%s'", filename);
3014         exit(1);
3015     }
3016     g_free(filename);
3017 
3018     /* FIXME: Should register things through the MachineState's qdev
3019      * interface, this is a legacy from the sPAPREnvironment structure
3020      * which predated MachineState but had a similar function */
3021     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3022     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3023                          &savevm_htab_handlers, spapr);
3024 
3025     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3026                              &error_fatal);
3027 
3028     qemu_register_boot_set(spapr_boot_set, spapr);
3029 
3030     /*
3031      * Nothing needs to be done to resume a suspended guest because
3032      * suspending does not change the machine state, so no need for
3033      * a ->wakeup method.
3034      */
3035     qemu_register_wakeup_support();
3036 
3037     if (kvm_enabled()) {
3038         /* to stop and start vmclock */
3039         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3040                                          &spapr->tb);
3041 
3042         kvmppc_spapr_enable_inkernel_multitce();
3043     }
3044 
3045     qemu_cond_init(&spapr->mc_delivery_cond);
3046 }
3047 
3048 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3049 {
3050     if (!vm_type) {
3051         return 0;
3052     }
3053 
3054     if (!strcmp(vm_type, "HV")) {
3055         return 1;
3056     }
3057 
3058     if (!strcmp(vm_type, "PR")) {
3059         return 2;
3060     }
3061 
3062     error_report("Unknown kvm-type specified '%s'", vm_type);
3063     exit(1);
3064 }
3065 
3066 /*
3067  * Implementation of an interface to adjust firmware path
3068  * for the bootindex property handling.
3069  */
3070 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3071                                    DeviceState *dev)
3072 {
3073 #define CAST(type, obj, name) \
3074     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3075     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3076     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3077     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3078 
3079     if (d) {
3080         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3081         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3082         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3083 
3084         if (spapr) {
3085             /*
3086              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3087              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3088              * 0x8000 | (target << 8) | (bus << 5) | lun
3089              * (see the "Logical unit addressing format" table in SAM5)
3090              */
3091             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3092             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3093                                    (uint64_t)id << 48);
3094         } else if (virtio) {
3095             /*
3096              * We use SRP luns of the form 01000000 | (target << 8) | lun
3097              * in the top 32 bits of the 64-bit LUN
3098              * Note: the quote above is from SLOF and it is wrong,
3099              * the actual binding is:
3100              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3101              */
3102             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3103             if (d->lun >= 256) {
3104                 /* Use the LUN "flat space addressing method" */
3105                 id |= 0x4000;
3106             }
3107             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3108                                    (uint64_t)id << 32);
3109         } else if (usb) {
3110             /*
3111              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3112              * in the top 32 bits of the 64-bit LUN
3113              */
3114             unsigned usb_port = atoi(usb->port->path);
3115             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3116             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3117                                    (uint64_t)id << 32);
3118         }
3119     }
3120 
3121     /*
3122      * SLOF probes the USB devices, and if it recognizes that the device is a
3123      * storage device, it changes its name to "storage" instead of "usb-host",
3124      * and additionally adds a child node for the SCSI LUN, so the correct
3125      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3126      */
3127     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3128         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3129         if (usb_host_dev_is_scsi_storage(usbdev)) {
3130             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3131         }
3132     }
3133 
3134     if (phb) {
3135         /* Replace "pci" with "pci@800000020000000" */
3136         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3137     }
3138 
3139     if (vsc) {
3140         /* Same logic as virtio above */
3141         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3142         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3143     }
3144 
3145     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3146         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3147         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3148         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3149     }
3150 
3151     return NULL;
3152 }
3153 
3154 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3155 {
3156     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3157 
3158     return g_strdup(spapr->kvm_type);
3159 }
3160 
3161 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3162 {
3163     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3164 
3165     g_free(spapr->kvm_type);
3166     spapr->kvm_type = g_strdup(value);
3167 }
3168 
3169 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3170 {
3171     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3172 
3173     return spapr->use_hotplug_event_source;
3174 }
3175 
3176 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3177                                             Error **errp)
3178 {
3179     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3180 
3181     spapr->use_hotplug_event_source = value;
3182 }
3183 
3184 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3185 {
3186     return true;
3187 }
3188 
3189 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     switch (spapr->resize_hpt) {
3194     case SPAPR_RESIZE_HPT_DEFAULT:
3195         return g_strdup("default");
3196     case SPAPR_RESIZE_HPT_DISABLED:
3197         return g_strdup("disabled");
3198     case SPAPR_RESIZE_HPT_ENABLED:
3199         return g_strdup("enabled");
3200     case SPAPR_RESIZE_HPT_REQUIRED:
3201         return g_strdup("required");
3202     }
3203     g_assert_not_reached();
3204 }
3205 
3206 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3207 {
3208     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3209 
3210     if (strcmp(value, "default") == 0) {
3211         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3212     } else if (strcmp(value, "disabled") == 0) {
3213         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3214     } else if (strcmp(value, "enabled") == 0) {
3215         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3216     } else if (strcmp(value, "required") == 0) {
3217         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3218     } else {
3219         error_setg(errp, "Bad value for \"resize-hpt\" property");
3220     }
3221 }
3222 
3223 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3224                                    void *opaque, Error **errp)
3225 {
3226     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3227 }
3228 
3229 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3230                                    void *opaque, Error **errp)
3231 {
3232     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3233 }
3234 
3235 static void spapr_get_kernel_addr(Object *obj, Visitor *v, const char *name,
3236                                   void *opaque, Error **errp)
3237 {
3238     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3239 }
3240 
3241 static void spapr_set_kernel_addr(Object *obj, Visitor *v, const char *name,
3242                                   void *opaque, Error **errp)
3243 {
3244     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3245 }
3246 
3247 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3248 {
3249     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3250 
3251     if (spapr->irq == &spapr_irq_xics_legacy) {
3252         return g_strdup("legacy");
3253     } else if (spapr->irq == &spapr_irq_xics) {
3254         return g_strdup("xics");
3255     } else if (spapr->irq == &spapr_irq_xive) {
3256         return g_strdup("xive");
3257     } else if (spapr->irq == &spapr_irq_dual) {
3258         return g_strdup("dual");
3259     }
3260     g_assert_not_reached();
3261 }
3262 
3263 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3264 {
3265     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3266 
3267     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3268         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3269         return;
3270     }
3271 
3272     /* The legacy IRQ backend can not be set */
3273     if (strcmp(value, "xics") == 0) {
3274         spapr->irq = &spapr_irq_xics;
3275     } else if (strcmp(value, "xive") == 0) {
3276         spapr->irq = &spapr_irq_xive;
3277     } else if (strcmp(value, "dual") == 0) {
3278         spapr->irq = &spapr_irq_dual;
3279     } else {
3280         error_setg(errp, "Bad value for \"ic-mode\" property");
3281     }
3282 }
3283 
3284 static char *spapr_get_host_model(Object *obj, Error **errp)
3285 {
3286     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3287 
3288     return g_strdup(spapr->host_model);
3289 }
3290 
3291 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3292 {
3293     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3294 
3295     g_free(spapr->host_model);
3296     spapr->host_model = g_strdup(value);
3297 }
3298 
3299 static char *spapr_get_host_serial(Object *obj, Error **errp)
3300 {
3301     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3302 
3303     return g_strdup(spapr->host_serial);
3304 }
3305 
3306 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3307 {
3308     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3309 
3310     g_free(spapr->host_serial);
3311     spapr->host_serial = g_strdup(value);
3312 }
3313 
3314 static void spapr_instance_init(Object *obj)
3315 {
3316     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3317     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3318 
3319     spapr->htab_fd = -1;
3320     spapr->use_hotplug_event_source = true;
3321     object_property_add_str(obj, "kvm-type",
3322                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3323     object_property_set_description(obj, "kvm-type",
3324                                     "Specifies the KVM virtualization mode (HV, PR)",
3325                                     NULL);
3326     object_property_add_bool(obj, "modern-hotplug-events",
3327                             spapr_get_modern_hotplug_events,
3328                             spapr_set_modern_hotplug_events,
3329                             NULL);
3330     object_property_set_description(obj, "modern-hotplug-events",
3331                                     "Use dedicated hotplug event mechanism in"
3332                                     " place of standard EPOW events when possible"
3333                                     " (required for memory hot-unplug support)",
3334                                     NULL);
3335     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3336                             "Maximum permitted CPU compatibility mode",
3337                             &error_fatal);
3338 
3339     object_property_add_str(obj, "resize-hpt",
3340                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3341     object_property_set_description(obj, "resize-hpt",
3342                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3343                                     NULL);
3344     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3345                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3346     object_property_set_description(obj, "vsmt",
3347                                     "Virtual SMT: KVM behaves as if this were"
3348                                     " the host's SMT mode", &error_abort);
3349     object_property_add_bool(obj, "vfio-no-msix-emulation",
3350                              spapr_get_msix_emulation, NULL, NULL);
3351 
3352     object_property_add(obj, "kernel-addr", "uint64", spapr_get_kernel_addr,
3353                         spapr_set_kernel_addr, NULL, &spapr->kernel_addr,
3354                         &error_abort);
3355     object_property_set_description(obj, "kernel-addr",
3356                                     stringify(KERNEL_LOAD_ADDR)
3357                                     " for -kernel is the default",
3358                                     NULL);
3359     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3360     /* The machine class defines the default interrupt controller mode */
3361     spapr->irq = smc->irq;
3362     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3363                             spapr_set_ic_mode, NULL);
3364     object_property_set_description(obj, "ic-mode",
3365                  "Specifies the interrupt controller mode (xics, xive, dual)",
3366                  NULL);
3367 
3368     object_property_add_str(obj, "host-model",
3369         spapr_get_host_model, spapr_set_host_model,
3370         &error_abort);
3371     object_property_set_description(obj, "host-model",
3372         "Host model to advertise in guest device tree", &error_abort);
3373     object_property_add_str(obj, "host-serial",
3374         spapr_get_host_serial, spapr_set_host_serial,
3375         &error_abort);
3376     object_property_set_description(obj, "host-serial",
3377         "Host serial number to advertise in guest device tree", &error_abort);
3378 }
3379 
3380 static void spapr_machine_finalizefn(Object *obj)
3381 {
3382     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3383 
3384     g_free(spapr->kvm_type);
3385 }
3386 
3387 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3388 {
3389     cpu_synchronize_state(cs);
3390     ppc_cpu_do_system_reset(cs);
3391 }
3392 
3393 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3394 {
3395     CPUState *cs;
3396 
3397     CPU_FOREACH(cs) {
3398         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3399     }
3400 }
3401 
3402 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3403                           void *fdt, int *fdt_start_offset, Error **errp)
3404 {
3405     uint64_t addr;
3406     uint32_t node;
3407 
3408     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3409     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3410                                     &error_abort);
3411     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3412                                                    SPAPR_MEMORY_BLOCK_SIZE);
3413     return 0;
3414 }
3415 
3416 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3417                            bool dedicated_hp_event_source, Error **errp)
3418 {
3419     SpaprDrc *drc;
3420     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3421     int i;
3422     uint64_t addr = addr_start;
3423     bool hotplugged = spapr_drc_hotplugged(dev);
3424     Error *local_err = NULL;
3425 
3426     for (i = 0; i < nr_lmbs; i++) {
3427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3428                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3429         g_assert(drc);
3430 
3431         spapr_drc_attach(drc, dev, &local_err);
3432         if (local_err) {
3433             while (addr > addr_start) {
3434                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3435                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3436                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3437                 spapr_drc_detach(drc);
3438             }
3439             error_propagate(errp, local_err);
3440             return;
3441         }
3442         if (!hotplugged) {
3443             spapr_drc_reset(drc);
3444         }
3445         addr += SPAPR_MEMORY_BLOCK_SIZE;
3446     }
3447     /* send hotplug notification to the
3448      * guest only in case of hotplugged memory
3449      */
3450     if (hotplugged) {
3451         if (dedicated_hp_event_source) {
3452             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3453                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3454             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3455                                                    nr_lmbs,
3456                                                    spapr_drc_index(drc));
3457         } else {
3458             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3459                                            nr_lmbs);
3460         }
3461     }
3462 }
3463 
3464 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3465                               Error **errp)
3466 {
3467     Error *local_err = NULL;
3468     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3469     PCDIMMDevice *dimm = PC_DIMM(dev);
3470     uint64_t size, addr, slot;
3471     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3472 
3473     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3474 
3475     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3476     if (local_err) {
3477         goto out;
3478     }
3479 
3480     if (!is_nvdimm) {
3481         addr = object_property_get_uint(OBJECT(dimm),
3482                                         PC_DIMM_ADDR_PROP, &local_err);
3483         if (local_err) {
3484             goto out_unplug;
3485         }
3486         spapr_add_lmbs(dev, addr, size,
3487                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3488                        &local_err);
3489     } else {
3490         slot = object_property_get_uint(OBJECT(dimm),
3491                                         PC_DIMM_SLOT_PROP, &local_err);
3492         if (local_err) {
3493             goto out_unplug;
3494         }
3495         spapr_add_nvdimm(dev, slot, &local_err);
3496     }
3497 
3498     if (local_err) {
3499         goto out_unplug;
3500     }
3501 
3502     return;
3503 
3504 out_unplug:
3505     pc_dimm_unplug(dimm, MACHINE(ms));
3506 out:
3507     error_propagate(errp, local_err);
3508 }
3509 
3510 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3511                                   Error **errp)
3512 {
3513     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3514     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3515     const MachineClass *mc = MACHINE_CLASS(smc);
3516     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3517     PCDIMMDevice *dimm = PC_DIMM(dev);
3518     Error *local_err = NULL;
3519     uint64_t size;
3520     Object *memdev;
3521     hwaddr pagesize;
3522 
3523     if (!smc->dr_lmb_enabled) {
3524         error_setg(errp, "Memory hotplug not supported for this machine");
3525         return;
3526     }
3527 
3528     if (is_nvdimm && !mc->nvdimm_supported) {
3529         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3530         return;
3531     }
3532 
3533     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3534     if (local_err) {
3535         error_propagate(errp, local_err);
3536         return;
3537     }
3538 
3539     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3540         error_setg(errp, "Hotplugged memory size must be a multiple of "
3541                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3542         return;
3543     } else if (is_nvdimm) {
3544         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3545         if (local_err) {
3546             error_propagate(errp, local_err);
3547             return;
3548         }
3549     }
3550 
3551     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3552                                       &error_abort);
3553     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3554     spapr_check_pagesize(spapr, pagesize, &local_err);
3555     if (local_err) {
3556         error_propagate(errp, local_err);
3557         return;
3558     }
3559 
3560     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3561 }
3562 
3563 struct SpaprDimmState {
3564     PCDIMMDevice *dimm;
3565     uint32_t nr_lmbs;
3566     QTAILQ_ENTRY(SpaprDimmState) next;
3567 };
3568 
3569 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3570                                                        PCDIMMDevice *dimm)
3571 {
3572     SpaprDimmState *dimm_state = NULL;
3573 
3574     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3575         if (dimm_state->dimm == dimm) {
3576             break;
3577         }
3578     }
3579     return dimm_state;
3580 }
3581 
3582 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3583                                                       uint32_t nr_lmbs,
3584                                                       PCDIMMDevice *dimm)
3585 {
3586     SpaprDimmState *ds = NULL;
3587 
3588     /*
3589      * If this request is for a DIMM whose removal had failed earlier
3590      * (due to guest's refusal to remove the LMBs), we would have this
3591      * dimm already in the pending_dimm_unplugs list. In that
3592      * case don't add again.
3593      */
3594     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3595     if (!ds) {
3596         ds = g_malloc0(sizeof(SpaprDimmState));
3597         ds->nr_lmbs = nr_lmbs;
3598         ds->dimm = dimm;
3599         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3600     }
3601     return ds;
3602 }
3603 
3604 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3605                                               SpaprDimmState *dimm_state)
3606 {
3607     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3608     g_free(dimm_state);
3609 }
3610 
3611 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3612                                                         PCDIMMDevice *dimm)
3613 {
3614     SpaprDrc *drc;
3615     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3616                                                   &error_abort);
3617     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3618     uint32_t avail_lmbs = 0;
3619     uint64_t addr_start, addr;
3620     int i;
3621 
3622     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3623                                          &error_abort);
3624 
3625     addr = addr_start;
3626     for (i = 0; i < nr_lmbs; i++) {
3627         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3628                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3629         g_assert(drc);
3630         if (drc->dev) {
3631             avail_lmbs++;
3632         }
3633         addr += SPAPR_MEMORY_BLOCK_SIZE;
3634     }
3635 
3636     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3637 }
3638 
3639 /* Callback to be called during DRC release. */
3640 void spapr_lmb_release(DeviceState *dev)
3641 {
3642     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3643     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3644     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3645 
3646     /* This information will get lost if a migration occurs
3647      * during the unplug process. In this case recover it. */
3648     if (ds == NULL) {
3649         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3650         g_assert(ds);
3651         /* The DRC being examined by the caller at least must be counted */
3652         g_assert(ds->nr_lmbs);
3653     }
3654 
3655     if (--ds->nr_lmbs) {
3656         return;
3657     }
3658 
3659     /*
3660      * Now that all the LMBs have been removed by the guest, call the
3661      * unplug handler chain. This can never fail.
3662      */
3663     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3664     object_unparent(OBJECT(dev));
3665 }
3666 
3667 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3668 {
3669     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3670     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3671 
3672     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3673     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3674     spapr_pending_dimm_unplugs_remove(spapr, ds);
3675 }
3676 
3677 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3678                                         DeviceState *dev, Error **errp)
3679 {
3680     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3681     Error *local_err = NULL;
3682     PCDIMMDevice *dimm = PC_DIMM(dev);
3683     uint32_t nr_lmbs;
3684     uint64_t size, addr_start, addr;
3685     int i;
3686     SpaprDrc *drc;
3687 
3688     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3689         error_setg(&local_err,
3690                    "nvdimm device hot unplug is not supported yet.");
3691         goto out;
3692     }
3693 
3694     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3695     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3696 
3697     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3698                                          &local_err);
3699     if (local_err) {
3700         goto out;
3701     }
3702 
3703     /*
3704      * An existing pending dimm state for this DIMM means that there is an
3705      * unplug operation in progress, waiting for the spapr_lmb_release
3706      * callback to complete the job (BQL can't cover that far). In this case,
3707      * bail out to avoid detaching DRCs that were already released.
3708      */
3709     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3710         error_setg(&local_err,
3711                    "Memory unplug already in progress for device %s",
3712                    dev->id);
3713         goto out;
3714     }
3715 
3716     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3717 
3718     addr = addr_start;
3719     for (i = 0; i < nr_lmbs; i++) {
3720         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3721                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3722         g_assert(drc);
3723 
3724         spapr_drc_detach(drc);
3725         addr += SPAPR_MEMORY_BLOCK_SIZE;
3726     }
3727 
3728     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3729                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3730     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3731                                               nr_lmbs, spapr_drc_index(drc));
3732 out:
3733     error_propagate(errp, local_err);
3734 }
3735 
3736 /* Callback to be called during DRC release. */
3737 void spapr_core_release(DeviceState *dev)
3738 {
3739     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3740 
3741     /* Call the unplug handler chain. This can never fail. */
3742     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3743     object_unparent(OBJECT(dev));
3744 }
3745 
3746 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3747 {
3748     MachineState *ms = MACHINE(hotplug_dev);
3749     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3750     CPUCore *cc = CPU_CORE(dev);
3751     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3752 
3753     if (smc->pre_2_10_has_unused_icps) {
3754         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3755         int i;
3756 
3757         for (i = 0; i < cc->nr_threads; i++) {
3758             CPUState *cs = CPU(sc->threads[i]);
3759 
3760             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3761         }
3762     }
3763 
3764     assert(core_slot);
3765     core_slot->cpu = NULL;
3766     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3767 }
3768 
3769 static
3770 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3771                                Error **errp)
3772 {
3773     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3774     int index;
3775     SpaprDrc *drc;
3776     CPUCore *cc = CPU_CORE(dev);
3777 
3778     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3779         error_setg(errp, "Unable to find CPU core with core-id: %d",
3780                    cc->core_id);
3781         return;
3782     }
3783     if (index == 0) {
3784         error_setg(errp, "Boot CPU core may not be unplugged");
3785         return;
3786     }
3787 
3788     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3789                           spapr_vcpu_id(spapr, cc->core_id));
3790     g_assert(drc);
3791 
3792     if (!spapr_drc_unplug_requested(drc)) {
3793         spapr_drc_detach(drc);
3794         spapr_hotplug_req_remove_by_index(drc);
3795     }
3796 }
3797 
3798 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3799                            void *fdt, int *fdt_start_offset, Error **errp)
3800 {
3801     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3802     CPUState *cs = CPU(core->threads[0]);
3803     PowerPCCPU *cpu = POWERPC_CPU(cs);
3804     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3805     int id = spapr_get_vcpu_id(cpu);
3806     char *nodename;
3807     int offset;
3808 
3809     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3810     offset = fdt_add_subnode(fdt, 0, nodename);
3811     g_free(nodename);
3812 
3813     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3814 
3815     *fdt_start_offset = offset;
3816     return 0;
3817 }
3818 
3819 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3820                             Error **errp)
3821 {
3822     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3823     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3824     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3825     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3826     CPUCore *cc = CPU_CORE(dev);
3827     CPUState *cs;
3828     SpaprDrc *drc;
3829     Error *local_err = NULL;
3830     CPUArchId *core_slot;
3831     int index;
3832     bool hotplugged = spapr_drc_hotplugged(dev);
3833     int i;
3834 
3835     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3836     if (!core_slot) {
3837         error_setg(errp, "Unable to find CPU core with core-id: %d",
3838                    cc->core_id);
3839         return;
3840     }
3841     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3842                           spapr_vcpu_id(spapr, cc->core_id));
3843 
3844     g_assert(drc || !mc->has_hotpluggable_cpus);
3845 
3846     if (drc) {
3847         spapr_drc_attach(drc, dev, &local_err);
3848         if (local_err) {
3849             error_propagate(errp, local_err);
3850             return;
3851         }
3852 
3853         if (hotplugged) {
3854             /*
3855              * Send hotplug notification interrupt to the guest only
3856              * in case of hotplugged CPUs.
3857              */
3858             spapr_hotplug_req_add_by_index(drc);
3859         } else {
3860             spapr_drc_reset(drc);
3861         }
3862     }
3863 
3864     core_slot->cpu = OBJECT(dev);
3865 
3866     if (smc->pre_2_10_has_unused_icps) {
3867         for (i = 0; i < cc->nr_threads; i++) {
3868             cs = CPU(core->threads[i]);
3869             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3870         }
3871     }
3872 
3873     /*
3874      * Set compatibility mode to match the boot CPU, which was either set
3875      * by the machine reset code or by CAS.
3876      */
3877     if (hotplugged) {
3878         for (i = 0; i < cc->nr_threads; i++) {
3879             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3880                            &local_err);
3881             if (local_err) {
3882                 error_propagate(errp, local_err);
3883                 return;
3884             }
3885         }
3886     }
3887 }
3888 
3889 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3890                                 Error **errp)
3891 {
3892     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3893     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3894     Error *local_err = NULL;
3895     CPUCore *cc = CPU_CORE(dev);
3896     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3897     const char *type = object_get_typename(OBJECT(dev));
3898     CPUArchId *core_slot;
3899     int index;
3900     unsigned int smp_threads = machine->smp.threads;
3901 
3902     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3903         error_setg(&local_err, "CPU hotplug not supported for this machine");
3904         goto out;
3905     }
3906 
3907     if (strcmp(base_core_type, type)) {
3908         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3909         goto out;
3910     }
3911 
3912     if (cc->core_id % smp_threads) {
3913         error_setg(&local_err, "invalid core id %d", cc->core_id);
3914         goto out;
3915     }
3916 
3917     /*
3918      * In general we should have homogeneous threads-per-core, but old
3919      * (pre hotplug support) machine types allow the last core to have
3920      * reduced threads as a compatibility hack for when we allowed
3921      * total vcpus not a multiple of threads-per-core.
3922      */
3923     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3924         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3925                    cc->nr_threads, smp_threads);
3926         goto out;
3927     }
3928 
3929     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3930     if (!core_slot) {
3931         error_setg(&local_err, "core id %d out of range", cc->core_id);
3932         goto out;
3933     }
3934 
3935     if (core_slot->cpu) {
3936         error_setg(&local_err, "core %d already populated", cc->core_id);
3937         goto out;
3938     }
3939 
3940     numa_cpu_pre_plug(core_slot, dev, &local_err);
3941 
3942 out:
3943     error_propagate(errp, local_err);
3944 }
3945 
3946 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3947                           void *fdt, int *fdt_start_offset, Error **errp)
3948 {
3949     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3950     int intc_phandle;
3951 
3952     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3953     if (intc_phandle <= 0) {
3954         return -1;
3955     }
3956 
3957     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3958         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3959         return -1;
3960     }
3961 
3962     /* generally SLOF creates these, for hotplug it's up to QEMU */
3963     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3964 
3965     return 0;
3966 }
3967 
3968 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3969                                Error **errp)
3970 {
3971     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3972     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3973     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3974     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3975 
3976     if (dev->hotplugged && !smc->dr_phb_enabled) {
3977         error_setg(errp, "PHB hotplug not supported for this machine");
3978         return;
3979     }
3980 
3981     if (sphb->index == (uint32_t)-1) {
3982         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3983         return;
3984     }
3985 
3986     /*
3987      * This will check that sphb->index doesn't exceed the maximum number of
3988      * PHBs for the current machine type.
3989      */
3990     smc->phb_placement(spapr, sphb->index,
3991                        &sphb->buid, &sphb->io_win_addr,
3992                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3993                        windows_supported, sphb->dma_liobn,
3994                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3995                        errp);
3996 }
3997 
3998 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3999                            Error **errp)
4000 {
4001     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4002     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4003     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4004     SpaprDrc *drc;
4005     bool hotplugged = spapr_drc_hotplugged(dev);
4006     Error *local_err = NULL;
4007 
4008     if (!smc->dr_phb_enabled) {
4009         return;
4010     }
4011 
4012     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4013     /* hotplug hooks should check it's enabled before getting this far */
4014     assert(drc);
4015 
4016     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4017     if (local_err) {
4018         error_propagate(errp, local_err);
4019         return;
4020     }
4021 
4022     if (hotplugged) {
4023         spapr_hotplug_req_add_by_index(drc);
4024     } else {
4025         spapr_drc_reset(drc);
4026     }
4027 }
4028 
4029 void spapr_phb_release(DeviceState *dev)
4030 {
4031     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4032 
4033     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4034     object_unparent(OBJECT(dev));
4035 }
4036 
4037 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4038 {
4039     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4040 }
4041 
4042 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4043                                      DeviceState *dev, Error **errp)
4044 {
4045     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4046     SpaprDrc *drc;
4047 
4048     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4049     assert(drc);
4050 
4051     if (!spapr_drc_unplug_requested(drc)) {
4052         spapr_drc_detach(drc);
4053         spapr_hotplug_req_remove_by_index(drc);
4054     }
4055 }
4056 
4057 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4058                                  Error **errp)
4059 {
4060     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4061     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4062 
4063     if (spapr->tpm_proxy != NULL) {
4064         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4065         return;
4066     }
4067 
4068     spapr->tpm_proxy = tpm_proxy;
4069 }
4070 
4071 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4072 {
4073     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4074 
4075     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4076     object_unparent(OBJECT(dev));
4077     spapr->tpm_proxy = NULL;
4078 }
4079 
4080 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4081                                       DeviceState *dev, Error **errp)
4082 {
4083     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4084         spapr_memory_plug(hotplug_dev, dev, errp);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4086         spapr_core_plug(hotplug_dev, dev, errp);
4087     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4088         spapr_phb_plug(hotplug_dev, dev, errp);
4089     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4090         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4091     }
4092 }
4093 
4094 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4095                                         DeviceState *dev, Error **errp)
4096 {
4097     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4098         spapr_memory_unplug(hotplug_dev, dev);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4100         spapr_core_unplug(hotplug_dev, dev);
4101     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4102         spapr_phb_unplug(hotplug_dev, dev);
4103     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4104         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4105     }
4106 }
4107 
4108 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4109                                                 DeviceState *dev, Error **errp)
4110 {
4111     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4112     MachineClass *mc = MACHINE_GET_CLASS(sms);
4113     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4114 
4115     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4116         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4117             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4118         } else {
4119             /* NOTE: this means there is a window after guest reset, prior to
4120              * CAS negotiation, where unplug requests will fail due to the
4121              * capability not being detected yet. This is a bit different than
4122              * the case with PCI unplug, where the events will be queued and
4123              * eventually handled by the guest after boot
4124              */
4125             error_setg(errp, "Memory hot unplug not supported for this guest");
4126         }
4127     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4128         if (!mc->has_hotpluggable_cpus) {
4129             error_setg(errp, "CPU hot unplug not supported on this machine");
4130             return;
4131         }
4132         spapr_core_unplug_request(hotplug_dev, dev, errp);
4133     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4134         if (!smc->dr_phb_enabled) {
4135             error_setg(errp, "PHB hot unplug not supported on this machine");
4136             return;
4137         }
4138         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4139     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4140         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4141     }
4142 }
4143 
4144 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4145                                           DeviceState *dev, Error **errp)
4146 {
4147     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4148         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4149     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4150         spapr_core_pre_plug(hotplug_dev, dev, errp);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4152         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4153     }
4154 }
4155 
4156 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4157                                                  DeviceState *dev)
4158 {
4159     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4160         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4161         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4162         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4163         return HOTPLUG_HANDLER(machine);
4164     }
4165     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4166         PCIDevice *pcidev = PCI_DEVICE(dev);
4167         PCIBus *root = pci_device_root_bus(pcidev);
4168         SpaprPhbState *phb =
4169             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4170                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4171 
4172         if (phb) {
4173             return HOTPLUG_HANDLER(phb);
4174         }
4175     }
4176     return NULL;
4177 }
4178 
4179 static CpuInstanceProperties
4180 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4181 {
4182     CPUArchId *core_slot;
4183     MachineClass *mc = MACHINE_GET_CLASS(machine);
4184 
4185     /* make sure possible_cpu are intialized */
4186     mc->possible_cpu_arch_ids(machine);
4187     /* get CPU core slot containing thread that matches cpu_index */
4188     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4189     assert(core_slot);
4190     return core_slot->props;
4191 }
4192 
4193 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4194 {
4195     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4196 }
4197 
4198 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4199 {
4200     int i;
4201     unsigned int smp_threads = machine->smp.threads;
4202     unsigned int smp_cpus = machine->smp.cpus;
4203     const char *core_type;
4204     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4205     MachineClass *mc = MACHINE_GET_CLASS(machine);
4206 
4207     if (!mc->has_hotpluggable_cpus) {
4208         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4209     }
4210     if (machine->possible_cpus) {
4211         assert(machine->possible_cpus->len == spapr_max_cores);
4212         return machine->possible_cpus;
4213     }
4214 
4215     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4216     if (!core_type) {
4217         error_report("Unable to find sPAPR CPU Core definition");
4218         exit(1);
4219     }
4220 
4221     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4222                              sizeof(CPUArchId) * spapr_max_cores);
4223     machine->possible_cpus->len = spapr_max_cores;
4224     for (i = 0; i < machine->possible_cpus->len; i++) {
4225         int core_id = i * smp_threads;
4226 
4227         machine->possible_cpus->cpus[i].type = core_type;
4228         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4229         machine->possible_cpus->cpus[i].arch_id = core_id;
4230         machine->possible_cpus->cpus[i].props.has_core_id = true;
4231         machine->possible_cpus->cpus[i].props.core_id = core_id;
4232     }
4233     return machine->possible_cpus;
4234 }
4235 
4236 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4237                                 uint64_t *buid, hwaddr *pio,
4238                                 hwaddr *mmio32, hwaddr *mmio64,
4239                                 unsigned n_dma, uint32_t *liobns,
4240                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4241 {
4242     /*
4243      * New-style PHB window placement.
4244      *
4245      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4246      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4247      * windows.
4248      *
4249      * Some guest kernels can't work with MMIO windows above 1<<46
4250      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4251      *
4252      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4253      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4254      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4255      * 1TiB 64-bit MMIO windows for each PHB.
4256      */
4257     const uint64_t base_buid = 0x800000020000000ULL;
4258     int i;
4259 
4260     /* Sanity check natural alignments */
4261     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4262     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4263     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4264     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4265     /* Sanity check bounds */
4266     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4267                       SPAPR_PCI_MEM32_WIN_SIZE);
4268     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4269                       SPAPR_PCI_MEM64_WIN_SIZE);
4270 
4271     if (index >= SPAPR_MAX_PHBS) {
4272         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4273                    SPAPR_MAX_PHBS - 1);
4274         return;
4275     }
4276 
4277     *buid = base_buid + index;
4278     for (i = 0; i < n_dma; ++i) {
4279         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4280     }
4281 
4282     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4283     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4284     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4285 
4286     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4287     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4288 }
4289 
4290 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4291 {
4292     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4293 
4294     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4295 }
4296 
4297 static void spapr_ics_resend(XICSFabric *dev)
4298 {
4299     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4300 
4301     ics_resend(spapr->ics);
4302 }
4303 
4304 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4305 {
4306     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4307 
4308     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4309 }
4310 
4311 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4312                                  Monitor *mon)
4313 {
4314     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4315 
4316     spapr_irq_print_info(spapr, mon);
4317     monitor_printf(mon, "irqchip: %s\n",
4318                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4319 }
4320 
4321 /*
4322  * This is a XIVE only operation
4323  */
4324 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4325                            uint8_t nvt_blk, uint32_t nvt_idx,
4326                            bool cam_ignore, uint8_t priority,
4327                            uint32_t logic_serv, XiveTCTXMatch *match)
4328 {
4329     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4330     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4331     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4332     int count;
4333 
4334     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4335                            priority, logic_serv, match);
4336     if (count < 0) {
4337         return count;
4338     }
4339 
4340     /*
4341      * When we implement the save and restore of the thread interrupt
4342      * contexts in the enter/exit CPU handlers of the machine and the
4343      * escalations in QEMU, we should be able to handle non dispatched
4344      * vCPUs.
4345      *
4346      * Until this is done, the sPAPR machine should find at least one
4347      * matching context always.
4348      */
4349     if (count == 0) {
4350         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4351                       nvt_blk, nvt_idx);
4352     }
4353 
4354     return count;
4355 }
4356 
4357 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4358 {
4359     return cpu->vcpu_id;
4360 }
4361 
4362 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4363 {
4364     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4365     MachineState *ms = MACHINE(spapr);
4366     int vcpu_id;
4367 
4368     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4369 
4370     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4371         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4372         error_append_hint(errp, "Adjust the number of cpus to %d "
4373                           "or try to raise the number of threads per core\n",
4374                           vcpu_id * ms->smp.threads / spapr->vsmt);
4375         return;
4376     }
4377 
4378     cpu->vcpu_id = vcpu_id;
4379 }
4380 
4381 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4382 {
4383     CPUState *cs;
4384 
4385     CPU_FOREACH(cs) {
4386         PowerPCCPU *cpu = POWERPC_CPU(cs);
4387 
4388         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4389             return cpu;
4390         }
4391     }
4392 
4393     return NULL;
4394 }
4395 
4396 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4397 {
4398     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4399 
4400     /* These are only called by TCG, KVM maintains dispatch state */
4401 
4402     spapr_cpu->prod = false;
4403     if (spapr_cpu->vpa_addr) {
4404         CPUState *cs = CPU(cpu);
4405         uint32_t dispatch;
4406 
4407         dispatch = ldl_be_phys(cs->as,
4408                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4409         dispatch++;
4410         if ((dispatch & 1) != 0) {
4411             qemu_log_mask(LOG_GUEST_ERROR,
4412                           "VPA: incorrect dispatch counter value for "
4413                           "dispatched partition %u, correcting.\n", dispatch);
4414             dispatch++;
4415         }
4416         stl_be_phys(cs->as,
4417                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4418     }
4419 }
4420 
4421 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4422 {
4423     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4424 
4425     if (spapr_cpu->vpa_addr) {
4426         CPUState *cs = CPU(cpu);
4427         uint32_t dispatch;
4428 
4429         dispatch = ldl_be_phys(cs->as,
4430                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4431         dispatch++;
4432         if ((dispatch & 1) != 1) {
4433             qemu_log_mask(LOG_GUEST_ERROR,
4434                           "VPA: incorrect dispatch counter value for "
4435                           "preempted partition %u, correcting.\n", dispatch);
4436             dispatch++;
4437         }
4438         stl_be_phys(cs->as,
4439                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4440     }
4441 }
4442 
4443 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4444 {
4445     MachineClass *mc = MACHINE_CLASS(oc);
4446     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4447     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4448     NMIClass *nc = NMI_CLASS(oc);
4449     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4450     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4451     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4452     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4453     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4454 
4455     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4456     mc->ignore_boot_device_suffixes = true;
4457 
4458     /*
4459      * We set up the default / latest behaviour here.  The class_init
4460      * functions for the specific versioned machine types can override
4461      * these details for backwards compatibility
4462      */
4463     mc->init = spapr_machine_init;
4464     mc->reset = spapr_machine_reset;
4465     mc->block_default_type = IF_SCSI;
4466     mc->max_cpus = 1024;
4467     mc->no_parallel = 1;
4468     mc->default_boot_order = "";
4469     mc->default_ram_size = 512 * MiB;
4470     mc->default_ram_id = "ppc_spapr.ram";
4471     mc->default_display = "std";
4472     mc->kvm_type = spapr_kvm_type;
4473     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4474     mc->pci_allow_0_address = true;
4475     assert(!mc->get_hotplug_handler);
4476     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4477     hc->pre_plug = spapr_machine_device_pre_plug;
4478     hc->plug = spapr_machine_device_plug;
4479     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4480     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4481     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4482     hc->unplug_request = spapr_machine_device_unplug_request;
4483     hc->unplug = spapr_machine_device_unplug;
4484 
4485     smc->dr_lmb_enabled = true;
4486     smc->update_dt_enabled = true;
4487     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4488     mc->has_hotpluggable_cpus = true;
4489     mc->nvdimm_supported = true;
4490     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4491     fwc->get_dev_path = spapr_get_fw_dev_path;
4492     nc->nmi_monitor_handler = spapr_nmi;
4493     smc->phb_placement = spapr_phb_placement;
4494     vhc->hypercall = emulate_spapr_hypercall;
4495     vhc->hpt_mask = spapr_hpt_mask;
4496     vhc->map_hptes = spapr_map_hptes;
4497     vhc->unmap_hptes = spapr_unmap_hptes;
4498     vhc->hpte_set_c = spapr_hpte_set_c;
4499     vhc->hpte_set_r = spapr_hpte_set_r;
4500     vhc->get_pate = spapr_get_pate;
4501     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4502     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4503     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4504     xic->ics_get = spapr_ics_get;
4505     xic->ics_resend = spapr_ics_resend;
4506     xic->icp_get = spapr_icp_get;
4507     ispc->print_info = spapr_pic_print_info;
4508     /* Force NUMA node memory size to be a multiple of
4509      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4510      * in which LMBs are represented and hot-added
4511      */
4512     mc->numa_mem_align_shift = 28;
4513     mc->numa_mem_supported = true;
4514     mc->auto_enable_numa = true;
4515 
4516     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4517     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4518     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4519     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4520     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4521     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4522     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4523     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4524     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4525     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4526     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4527     spapr_caps_add_properties(smc, &error_abort);
4528     smc->irq = &spapr_irq_dual;
4529     smc->dr_phb_enabled = true;
4530     smc->linux_pci_probe = true;
4531     smc->smp_threads_vsmt = true;
4532     smc->nr_xirqs = SPAPR_NR_XIRQS;
4533     xfc->match_nvt = spapr_match_nvt;
4534 }
4535 
4536 static const TypeInfo spapr_machine_info = {
4537     .name          = TYPE_SPAPR_MACHINE,
4538     .parent        = TYPE_MACHINE,
4539     .abstract      = true,
4540     .instance_size = sizeof(SpaprMachineState),
4541     .instance_init = spapr_instance_init,
4542     .instance_finalize = spapr_machine_finalizefn,
4543     .class_size    = sizeof(SpaprMachineClass),
4544     .class_init    = spapr_machine_class_init,
4545     .interfaces = (InterfaceInfo[]) {
4546         { TYPE_FW_PATH_PROVIDER },
4547         { TYPE_NMI },
4548         { TYPE_HOTPLUG_HANDLER },
4549         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4550         { TYPE_XICS_FABRIC },
4551         { TYPE_INTERRUPT_STATS_PROVIDER },
4552         { TYPE_XIVE_FABRIC },
4553         { }
4554     },
4555 };
4556 
4557 static void spapr_machine_latest_class_options(MachineClass *mc)
4558 {
4559     mc->alias = "pseries";
4560     mc->is_default = true;
4561 }
4562 
4563 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4564     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4565                                                     void *data)      \
4566     {                                                                \
4567         MachineClass *mc = MACHINE_CLASS(oc);                        \
4568         spapr_machine_##suffix##_class_options(mc);                  \
4569         if (latest) {                                                \
4570             spapr_machine_latest_class_options(mc);                  \
4571         }                                                            \
4572     }                                                                \
4573     static const TypeInfo spapr_machine_##suffix##_info = {          \
4574         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4575         .parent = TYPE_SPAPR_MACHINE,                                \
4576         .class_init = spapr_machine_##suffix##_class_init,           \
4577     };                                                               \
4578     static void spapr_machine_register_##suffix(void)                \
4579     {                                                                \
4580         type_register(&spapr_machine_##suffix##_info);               \
4581     }                                                                \
4582     type_init(spapr_machine_register_##suffix)
4583 
4584 /*
4585  * pseries-5.0
4586  */
4587 static void spapr_machine_5_0_class_options(MachineClass *mc)
4588 {
4589     /* Defaults for the latest behaviour inherited from the base class */
4590 }
4591 
4592 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4593 
4594 /*
4595  * pseries-4.2
4596  */
4597 static void spapr_machine_4_2_class_options(MachineClass *mc)
4598 {
4599     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4600 
4601     spapr_machine_5_0_class_options(mc);
4602     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4603     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4604     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4605     smc->rma_limit = 16 * GiB;
4606     mc->nvdimm_supported = false;
4607 }
4608 
4609 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4610 
4611 /*
4612  * pseries-4.1
4613  */
4614 static void spapr_machine_4_1_class_options(MachineClass *mc)
4615 {
4616     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4617     static GlobalProperty compat[] = {
4618         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4619         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4620     };
4621 
4622     spapr_machine_4_2_class_options(mc);
4623     smc->linux_pci_probe = false;
4624     smc->smp_threads_vsmt = false;
4625     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4626     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4627 }
4628 
4629 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4630 
4631 /*
4632  * pseries-4.0
4633  */
4634 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4635                               uint64_t *buid, hwaddr *pio,
4636                               hwaddr *mmio32, hwaddr *mmio64,
4637                               unsigned n_dma, uint32_t *liobns,
4638                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4639 {
4640     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4641                         nv2gpa, nv2atsd, errp);
4642     *nv2gpa = 0;
4643     *nv2atsd = 0;
4644 }
4645 
4646 static void spapr_machine_4_0_class_options(MachineClass *mc)
4647 {
4648     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4649 
4650     spapr_machine_4_1_class_options(mc);
4651     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4652     smc->phb_placement = phb_placement_4_0;
4653     smc->irq = &spapr_irq_xics;
4654     smc->pre_4_1_migration = true;
4655 }
4656 
4657 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4658 
4659 /*
4660  * pseries-3.1
4661  */
4662 static void spapr_machine_3_1_class_options(MachineClass *mc)
4663 {
4664     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4665 
4666     spapr_machine_4_0_class_options(mc);
4667     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4668 
4669     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4670     smc->update_dt_enabled = false;
4671     smc->dr_phb_enabled = false;
4672     smc->broken_host_serial_model = true;
4673     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4674     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4675     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4676     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4677 }
4678 
4679 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4680 
4681 /*
4682  * pseries-3.0
4683  */
4684 
4685 static void spapr_machine_3_0_class_options(MachineClass *mc)
4686 {
4687     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4688 
4689     spapr_machine_3_1_class_options(mc);
4690     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4691 
4692     smc->legacy_irq_allocation = true;
4693     smc->nr_xirqs = 0x400;
4694     smc->irq = &spapr_irq_xics_legacy;
4695 }
4696 
4697 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4698 
4699 /*
4700  * pseries-2.12
4701  */
4702 static void spapr_machine_2_12_class_options(MachineClass *mc)
4703 {
4704     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4705     static GlobalProperty compat[] = {
4706         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4707         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4708     };
4709 
4710     spapr_machine_3_0_class_options(mc);
4711     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4712     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4713 
4714     /* We depend on kvm_enabled() to choose a default value for the
4715      * hpt-max-page-size capability. Of course we can't do it here
4716      * because this is too early and the HW accelerator isn't initialzed
4717      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4718      */
4719     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4720 }
4721 
4722 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4723 
4724 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4725 {
4726     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4727 
4728     spapr_machine_2_12_class_options(mc);
4729     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4730     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4731     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4732 }
4733 
4734 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4735 
4736 /*
4737  * pseries-2.11
4738  */
4739 
4740 static void spapr_machine_2_11_class_options(MachineClass *mc)
4741 {
4742     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4743 
4744     spapr_machine_2_12_class_options(mc);
4745     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4746     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4747 }
4748 
4749 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4750 
4751 /*
4752  * pseries-2.10
4753  */
4754 
4755 static void spapr_machine_2_10_class_options(MachineClass *mc)
4756 {
4757     spapr_machine_2_11_class_options(mc);
4758     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4759 }
4760 
4761 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4762 
4763 /*
4764  * pseries-2.9
4765  */
4766 
4767 static void spapr_machine_2_9_class_options(MachineClass *mc)
4768 {
4769     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4770     static GlobalProperty compat[] = {
4771         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4772     };
4773 
4774     spapr_machine_2_10_class_options(mc);
4775     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4776     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4777     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4778     smc->pre_2_10_has_unused_icps = true;
4779     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4780 }
4781 
4782 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4783 
4784 /*
4785  * pseries-2.8
4786  */
4787 
4788 static void spapr_machine_2_8_class_options(MachineClass *mc)
4789 {
4790     static GlobalProperty compat[] = {
4791         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4792     };
4793 
4794     spapr_machine_2_9_class_options(mc);
4795     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4796     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4797     mc->numa_mem_align_shift = 23;
4798 }
4799 
4800 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4801 
4802 /*
4803  * pseries-2.7
4804  */
4805 
4806 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4807                               uint64_t *buid, hwaddr *pio,
4808                               hwaddr *mmio32, hwaddr *mmio64,
4809                               unsigned n_dma, uint32_t *liobns,
4810                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4811 {
4812     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4813     const uint64_t base_buid = 0x800000020000000ULL;
4814     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4815     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4816     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4817     const uint32_t max_index = 255;
4818     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4819 
4820     uint64_t ram_top = MACHINE(spapr)->ram_size;
4821     hwaddr phb0_base, phb_base;
4822     int i;
4823 
4824     /* Do we have device memory? */
4825     if (MACHINE(spapr)->maxram_size > ram_top) {
4826         /* Can't just use maxram_size, because there may be an
4827          * alignment gap between normal and device memory regions
4828          */
4829         ram_top = MACHINE(spapr)->device_memory->base +
4830             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4831     }
4832 
4833     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4834 
4835     if (index > max_index) {
4836         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4837                    max_index);
4838         return;
4839     }
4840 
4841     *buid = base_buid + index;
4842     for (i = 0; i < n_dma; ++i) {
4843         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4844     }
4845 
4846     phb_base = phb0_base + index * phb_spacing;
4847     *pio = phb_base + pio_offset;
4848     *mmio32 = phb_base + mmio_offset;
4849     /*
4850      * We don't set the 64-bit MMIO window, relying on the PHB's
4851      * fallback behaviour of automatically splitting a large "32-bit"
4852      * window into contiguous 32-bit and 64-bit windows
4853      */
4854 
4855     *nv2gpa = 0;
4856     *nv2atsd = 0;
4857 }
4858 
4859 static void spapr_machine_2_7_class_options(MachineClass *mc)
4860 {
4861     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4862     static GlobalProperty compat[] = {
4863         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4864         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4865         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4867     };
4868 
4869     spapr_machine_2_8_class_options(mc);
4870     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4871     mc->default_machine_opts = "modern-hotplug-events=off";
4872     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4873     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4874     smc->phb_placement = phb_placement_2_7;
4875 }
4876 
4877 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4878 
4879 /*
4880  * pseries-2.6
4881  */
4882 
4883 static void spapr_machine_2_6_class_options(MachineClass *mc)
4884 {
4885     static GlobalProperty compat[] = {
4886         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4887     };
4888 
4889     spapr_machine_2_7_class_options(mc);
4890     mc->has_hotpluggable_cpus = false;
4891     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4892     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4893 }
4894 
4895 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4896 
4897 /*
4898  * pseries-2.5
4899  */
4900 
4901 static void spapr_machine_2_5_class_options(MachineClass *mc)
4902 {
4903     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4904     static GlobalProperty compat[] = {
4905         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4906     };
4907 
4908     spapr_machine_2_6_class_options(mc);
4909     smc->use_ohci_by_default = true;
4910     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4911     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4912 }
4913 
4914 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4915 
4916 /*
4917  * pseries-2.4
4918  */
4919 
4920 static void spapr_machine_2_4_class_options(MachineClass *mc)
4921 {
4922     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4923 
4924     spapr_machine_2_5_class_options(mc);
4925     smc->dr_lmb_enabled = false;
4926     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4927 }
4928 
4929 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4930 
4931 /*
4932  * pseries-2.3
4933  */
4934 
4935 static void spapr_machine_2_3_class_options(MachineClass *mc)
4936 {
4937     static GlobalProperty compat[] = {
4938         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4939     };
4940     spapr_machine_2_4_class_options(mc);
4941     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4942     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4943 }
4944 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4945 
4946 /*
4947  * pseries-2.2
4948  */
4949 
4950 static void spapr_machine_2_2_class_options(MachineClass *mc)
4951 {
4952     static GlobalProperty compat[] = {
4953         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4954     };
4955 
4956     spapr_machine_2_3_class_options(mc);
4957     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4958     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4959     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4960 }
4961 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4962 
4963 /*
4964  * pseries-2.1
4965  */
4966 
4967 static void spapr_machine_2_1_class_options(MachineClass *mc)
4968 {
4969     spapr_machine_2_2_class_options(mc);
4970     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4971 }
4972 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4973 
4974 static void spapr_machine_register_types(void)
4975 {
4976     type_register_static(&spapr_machine_info);
4977 }
4978 
4979 type_init(spapr_machine_register_types)
4980