xref: /openbmc/qemu/hw/ppc/spapr.c (revision 5a894dd7)
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 RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            (128 * MiB)
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 static void spapr_dt_pa_features(SpaprMachineState *spapr,
220                                  PowerPCCPU *cpu,
221                                  void *fdt, int offset)
222 {
223     uint8_t pa_features_206[] = { 6, 0,
224         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
225     uint8_t pa_features_207[] = { 24, 0,
226         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
227         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
228         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
230     uint8_t pa_features_300[] = { 66, 0,
231         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
232         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
233         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
234         /* 6: DS207 */
235         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
236         /* 16: Vector */
237         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
238         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
240         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
242         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
243         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
244         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
245         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
246         /* 42: PM, 44: PC RA, 46: SC vec'd */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
248         /* 48: SIMD, 50: QP BFP, 52: String */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
250         /* 54: DecFP, 56: DecI, 58: SHA */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
252         /* 60: NM atomic, 62: RNG */
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
254     };
255     uint8_t *pa_features = NULL;
256     size_t pa_size;
257 
258     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
259         pa_features = pa_features_206;
260         pa_size = sizeof(pa_features_206);
261     }
262     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
263         pa_features = pa_features_207;
264         pa_size = sizeof(pa_features_207);
265     }
266     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
267         pa_features = pa_features_300;
268         pa_size = sizeof(pa_features_300);
269     }
270     if (!pa_features) {
271         return;
272     }
273 
274     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
275         /*
276          * Note: we keep CI large pages off by default because a 64K capable
277          * guest provisioned with large pages might otherwise try to map a qemu
278          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
279          * even if that qemu runs on a 4k host.
280          * We dd this bit back here if we are confident this is not an issue
281          */
282         pa_features[3] |= 0x20;
283     }
284     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
285         pa_features[24] |= 0x80;    /* Transactional memory support */
286     }
287     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
288         /* Workaround for broken kernels that attempt (guest) radix
289          * mode when they can't handle it, if they see the radix bit set
290          * in pa-features. So hide it from them. */
291         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
292     }
293 
294     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
295 }
296 
297 static hwaddr spapr_node0_size(MachineState *machine)
298 {
299     if (machine->numa_state->num_nodes) {
300         int i;
301         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
302             if (machine->numa_state->nodes[i].node_mem) {
303                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
304                            machine->ram_size);
305             }
306         }
307     }
308     return machine->ram_size;
309 }
310 
311 static void add_str(GString *s, const gchar *s1)
312 {
313     g_string_append_len(s, s1, strlen(s1) + 1);
314 }
315 
316 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
317                                 hwaddr size)
318 {
319     uint32_t associativity[] = {
320         cpu_to_be32(0x4), /* length */
321         cpu_to_be32(0x0), cpu_to_be32(0x0),
322         cpu_to_be32(0x0), cpu_to_be32(nodeid)
323     };
324     char mem_name[32];
325     uint64_t mem_reg_property[2];
326     int off;
327 
328     mem_reg_property[0] = cpu_to_be64(start);
329     mem_reg_property[1] = cpu_to_be64(size);
330 
331     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
332     off = fdt_add_subnode(fdt, 0, mem_name);
333     _FDT(off);
334     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
335     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
336                       sizeof(mem_reg_property))));
337     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
338                       sizeof(associativity))));
339     return off;
340 }
341 
342 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
343 {
344     MemoryDeviceInfoList *info;
345 
346     for (info = list; info; info = info->next) {
347         MemoryDeviceInfo *value = info->value;
348 
349         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
350             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
351 
352             if (addr >= pcdimm_info->addr &&
353                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
354                 return pcdimm_info->node;
355             }
356         }
357     }
358 
359     return -1;
360 }
361 
362 struct sPAPRDrconfCellV2 {
363      uint32_t seq_lmbs;
364      uint64_t base_addr;
365      uint32_t drc_index;
366      uint32_t aa_index;
367      uint32_t flags;
368 } QEMU_PACKED;
369 
370 typedef struct DrconfCellQueue {
371     struct sPAPRDrconfCellV2 cell;
372     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
373 } DrconfCellQueue;
374 
375 static DrconfCellQueue *
376 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
377                       uint32_t drc_index, uint32_t aa_index,
378                       uint32_t flags)
379 {
380     DrconfCellQueue *elem;
381 
382     elem = g_malloc0(sizeof(*elem));
383     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
384     elem->cell.base_addr = cpu_to_be64(base_addr);
385     elem->cell.drc_index = cpu_to_be32(drc_index);
386     elem->cell.aa_index = cpu_to_be32(aa_index);
387     elem->cell.flags = cpu_to_be32(flags);
388 
389     return elem;
390 }
391 
392 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
393                                       int offset, MemoryDeviceInfoList *dimms)
394 {
395     MachineState *machine = MACHINE(spapr);
396     uint8_t *int_buf, *cur_index;
397     int ret;
398     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
399     uint64_t addr, cur_addr, size;
400     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
401     uint64_t mem_end = machine->device_memory->base +
402                        memory_region_size(&machine->device_memory->mr);
403     uint32_t node, buf_len, nr_entries = 0;
404     SpaprDrc *drc;
405     DrconfCellQueue *elem, *next;
406     MemoryDeviceInfoList *info;
407     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
408         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
409 
410     /* Entry to cover RAM and the gap area */
411     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
412                                  SPAPR_LMB_FLAGS_RESERVED |
413                                  SPAPR_LMB_FLAGS_DRC_INVALID);
414     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
415     nr_entries++;
416 
417     cur_addr = machine->device_memory->base;
418     for (info = dimms; info; info = info->next) {
419         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
420 
421         addr = di->addr;
422         size = di->size;
423         node = di->node;
424 
425         /*
426          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
427          * area is marked hotpluggable in the next iteration for the bigger
428          * chunk including the NVDIMM occupied area.
429          */
430         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
431             continue;
432 
433         /* Entry for hot-pluggable area */
434         if (cur_addr < addr) {
435             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
436             g_assert(drc);
437             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
438                                          cur_addr, spapr_drc_index(drc), -1, 0);
439             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440             nr_entries++;
441         }
442 
443         /* Entry for DIMM */
444         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
445         g_assert(drc);
446         elem = spapr_get_drconf_cell(size / lmb_size, addr,
447                                      spapr_drc_index(drc), node,
448                                      (SPAPR_LMB_FLAGS_ASSIGNED |
449                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452         cur_addr = addr + size;
453     }
454 
455     /* Entry for remaining hotpluggable area */
456     if (cur_addr < mem_end) {
457         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
458         g_assert(drc);
459         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
460                                      cur_addr, spapr_drc_index(drc), -1, 0);
461         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
462         nr_entries++;
463     }
464 
465     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
466     int_buf = cur_index = g_malloc0(buf_len);
467     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
468     cur_index += sizeof(nr_entries);
469 
470     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
471         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
472         cur_index += sizeof(elem->cell);
473         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
474         g_free(elem);
475     }
476 
477     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
478     g_free(int_buf);
479     if (ret < 0) {
480         return -1;
481     }
482     return 0;
483 }
484 
485 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
486                                    int offset, MemoryDeviceInfoList *dimms)
487 {
488     MachineState *machine = MACHINE(spapr);
489     int i, ret;
490     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
491     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
492     uint32_t nr_lmbs = (machine->device_memory->base +
493                        memory_region_size(&machine->device_memory->mr)) /
494                        lmb_size;
495     uint32_t *int_buf, *cur_index, buf_len;
496 
497     /*
498      * Allocate enough buffer size to fit in ibm,dynamic-memory
499      */
500     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
501     cur_index = int_buf = g_malloc0(buf_len);
502     int_buf[0] = cpu_to_be32(nr_lmbs);
503     cur_index++;
504     for (i = 0; i < nr_lmbs; i++) {
505         uint64_t addr = i * lmb_size;
506         uint32_t *dynamic_memory = cur_index;
507 
508         if (i >= device_lmb_start) {
509             SpaprDrc *drc;
510 
511             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
512             g_assert(drc);
513 
514             dynamic_memory[0] = cpu_to_be32(addr >> 32);
515             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
516             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
517             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
518             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
519             if (memory_region_present(get_system_memory(), addr)) {
520                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
521             } else {
522                 dynamic_memory[5] = cpu_to_be32(0);
523             }
524         } else {
525             /*
526              * LMB information for RMA, boot time RAM and gap b/n RAM and
527              * device memory region -- all these are marked as reserved
528              * and as having no valid DRC.
529              */
530             dynamic_memory[0] = cpu_to_be32(addr >> 32);
531             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
532             dynamic_memory[2] = cpu_to_be32(0);
533             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
534             dynamic_memory[4] = cpu_to_be32(-1);
535             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
536                                             SPAPR_LMB_FLAGS_DRC_INVALID);
537         }
538 
539         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
540     }
541     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
542     g_free(int_buf);
543     if (ret < 0) {
544         return -1;
545     }
546     return 0;
547 }
548 
549 /*
550  * Adds ibm,dynamic-reconfiguration-memory node.
551  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
552  * of this device tree node.
553  */
554 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
555                                                    void *fdt)
556 {
557     MachineState *machine = MACHINE(spapr);
558     int nb_numa_nodes = machine->numa_state->num_nodes;
559     int ret, i, offset;
560     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
561     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
562                                 cpu_to_be32(lmb_size & 0xffffffff)};
563     uint32_t *int_buf, *cur_index, buf_len;
564     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
565     MemoryDeviceInfoList *dimms = NULL;
566 
567     /*
568      * Don't create the node if there is no device memory
569      */
570     if (machine->ram_size == machine->maxram_size) {
571         return 0;
572     }
573 
574     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
575 
576     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
577                     sizeof(prop_lmb_size));
578     if (ret < 0) {
579         return ret;
580     }
581 
582     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
583     if (ret < 0) {
584         return ret;
585     }
586 
587     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
588     if (ret < 0) {
589         return ret;
590     }
591 
592     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
593     dimms = qmp_memory_device_list();
594     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
595         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
596     } else {
597         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
598     }
599     qapi_free_MemoryDeviceInfoList(dimms);
600 
601     if (ret < 0) {
602         return ret;
603     }
604 
605     /* ibm,associativity-lookup-arrays */
606     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
607     cur_index = int_buf = g_malloc0(buf_len);
608     int_buf[0] = cpu_to_be32(nr_nodes);
609     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
610     cur_index += 2;
611     for (i = 0; i < nr_nodes; i++) {
612         uint32_t associativity[] = {
613             cpu_to_be32(0x0),
614             cpu_to_be32(0x0),
615             cpu_to_be32(0x0),
616             cpu_to_be32(i)
617         };
618         memcpy(cur_index, associativity, sizeof(associativity));
619         cur_index += 4;
620     }
621     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
622             (cur_index - int_buf) * sizeof(uint32_t));
623     g_free(int_buf);
624 
625     return ret;
626 }
627 
628 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
629 {
630     MachineState *machine = MACHINE(spapr);
631     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
632     hwaddr mem_start, node_size;
633     int i, nb_nodes = machine->numa_state->num_nodes;
634     NodeInfo *nodes = machine->numa_state->nodes;
635 
636     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
637         if (!nodes[i].node_mem) {
638             continue;
639         }
640         if (mem_start >= machine->ram_size) {
641             node_size = 0;
642         } else {
643             node_size = nodes[i].node_mem;
644             if (node_size > machine->ram_size - mem_start) {
645                 node_size = machine->ram_size - mem_start;
646             }
647         }
648         if (!mem_start) {
649             /* spapr_machine_init() checks for rma_size <= node0_size
650              * already */
651             spapr_dt_memory_node(fdt, i, 0, spapr->rma_size);
652             mem_start += spapr->rma_size;
653             node_size -= spapr->rma_size;
654         }
655         for ( ; node_size; ) {
656             hwaddr sizetmp = pow2floor(node_size);
657 
658             /* mem_start != 0 here */
659             if (ctzl(mem_start) < ctzl(sizetmp)) {
660                 sizetmp = 1ULL << ctzl(mem_start);
661             }
662 
663             spapr_dt_memory_node(fdt, i, mem_start, sizetmp);
664             node_size -= sizetmp;
665             mem_start += sizetmp;
666         }
667     }
668 
669     /* Generate ibm,dynamic-reconfiguration-memory node if required */
670     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
671         int ret;
672 
673         g_assert(smc->dr_lmb_enabled);
674         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
675         if (ret) {
676             return ret;
677         }
678     }
679 
680     return 0;
681 }
682 
683 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
684                          SpaprMachineState *spapr)
685 {
686     MachineState *ms = MACHINE(spapr);
687     PowerPCCPU *cpu = POWERPC_CPU(cs);
688     CPUPPCState *env = &cpu->env;
689     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
690     int index = spapr_get_vcpu_id(cpu);
691     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
692                        0xffffffff, 0xffffffff};
693     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
694         : SPAPR_TIMEBASE_FREQ;
695     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
696     uint32_t page_sizes_prop[64];
697     size_t page_sizes_prop_size;
698     unsigned int smp_threads = ms->smp.threads;
699     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
700     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
701     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
702     SpaprDrc *drc;
703     int drc_index;
704     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
705     int i;
706 
707     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
708     if (drc) {
709         drc_index = spapr_drc_index(drc);
710         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
711     }
712 
713     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
714     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
715 
716     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
717     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
718                            env->dcache_line_size)));
719     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
720                            env->dcache_line_size)));
721     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
722                            env->icache_line_size)));
723     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
724                            env->icache_line_size)));
725 
726     if (pcc->l1_dcache_size) {
727         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
728                                pcc->l1_dcache_size)));
729     } else {
730         warn_report("Unknown L1 dcache size for cpu");
731     }
732     if (pcc->l1_icache_size) {
733         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
734                                pcc->l1_icache_size)));
735     } else {
736         warn_report("Unknown L1 icache size for cpu");
737     }
738 
739     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
740     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
741     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
742     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
743     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
744     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
745 
746     if (env->spr_cb[SPR_PURR].oea_read) {
747         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
748     }
749     if (env->spr_cb[SPR_SPURR].oea_read) {
750         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
751     }
752 
753     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
754         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
755                           segs, sizeof(segs))));
756     }
757 
758     /* Advertise VSX (vector extensions) if available
759      *   1               == VMX / Altivec available
760      *   2               == VSX available
761      *
762      * Only CPUs for which we create core types in spapr_cpu_core.c
763      * are possible, and all of those have VMX */
764     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
765         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
766     } else {
767         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
768     }
769 
770     /* Advertise DFP (Decimal Floating Point) if available
771      *   0 / no property == no DFP
772      *   1               == DFP available */
773     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
774         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
775     }
776 
777     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
778                                                       sizeof(page_sizes_prop));
779     if (page_sizes_prop_size) {
780         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
781                           page_sizes_prop, page_sizes_prop_size)));
782     }
783 
784     spapr_dt_pa_features(spapr, cpu, fdt, offset);
785 
786     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
787                            cs->cpu_index / vcpus_per_socket)));
788 
789     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
790                       pft_size_prop, sizeof(pft_size_prop))));
791 
792     if (ms->numa_state->num_nodes > 1) {
793         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
794     }
795 
796     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
797 
798     if (pcc->radix_page_info) {
799         for (i = 0; i < pcc->radix_page_info->count; i++) {
800             radix_AP_encodings[i] =
801                 cpu_to_be32(pcc->radix_page_info->entries[i]);
802         }
803         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
804                           radix_AP_encodings,
805                           pcc->radix_page_info->count *
806                           sizeof(radix_AP_encodings[0]))));
807     }
808 
809     /*
810      * We set this property to let the guest know that it can use the large
811      * decrementer and its width in bits.
812      */
813     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
814         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
815                               pcc->lrg_decr_bits)));
816 }
817 
818 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
819 {
820     CPUState **rev;
821     CPUState *cs;
822     int n_cpus;
823     int cpus_offset;
824     char *nodename;
825     int i;
826 
827     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
828     _FDT(cpus_offset);
829     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
830     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
831 
832     /*
833      * We walk the CPUs in reverse order to ensure that CPU DT nodes
834      * created by fdt_add_subnode() end up in the right order in FDT
835      * for the guest kernel the enumerate the CPUs correctly.
836      *
837      * The CPU list cannot be traversed in reverse order, so we need
838      * to do extra work.
839      */
840     n_cpus = 0;
841     rev = NULL;
842     CPU_FOREACH(cs) {
843         rev = g_renew(CPUState *, rev, n_cpus + 1);
844         rev[n_cpus++] = cs;
845     }
846 
847     for (i = n_cpus - 1; i >= 0; i--) {
848         CPUState *cs = rev[i];
849         PowerPCCPU *cpu = POWERPC_CPU(cs);
850         int index = spapr_get_vcpu_id(cpu);
851         DeviceClass *dc = DEVICE_GET_CLASS(cs);
852         int offset;
853 
854         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
855             continue;
856         }
857 
858         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
859         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
860         g_free(nodename);
861         _FDT(offset);
862         spapr_dt_cpu(cs, fdt, offset, spapr);
863     }
864 
865     g_free(rev);
866 }
867 
868 static int spapr_dt_rng(void *fdt)
869 {
870     int node;
871     int ret;
872 
873     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
874     if (node <= 0) {
875         return -1;
876     }
877     ret = fdt_setprop_string(fdt, node, "device_type",
878                              "ibm,platform-facilities");
879     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
880     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
881 
882     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
883     if (node <= 0) {
884         return -1;
885     }
886     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
887 
888     return ret ? -1 : 0;
889 }
890 
891 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
892 {
893     MachineState *ms = MACHINE(spapr);
894     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
895     int rtas;
896     GString *hypertas = g_string_sized_new(256);
897     GString *qemu_hypertas = g_string_sized_new(256);
898     uint32_t refpoints[] = {
899         cpu_to_be32(0x4),
900         cpu_to_be32(0x4),
901         cpu_to_be32(0x2),
902     };
903     uint32_t nr_refpoints = ARRAY_SIZE(refpoints);
904     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
905         memory_region_size(&MACHINE(spapr)->device_memory->mr);
906     uint32_t lrdr_capacity[] = {
907         cpu_to_be32(max_device_addr >> 32),
908         cpu_to_be32(max_device_addr & 0xffffffff),
909         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
910         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
911         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
912     };
913     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
914     uint32_t maxdomains[] = {
915         cpu_to_be32(4),
916         maxdomain,
917         maxdomain,
918         maxdomain,
919         cpu_to_be32(spapr->gpu_numa_id),
920     };
921 
922     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
923 
924     /* hypertas */
925     add_str(hypertas, "hcall-pft");
926     add_str(hypertas, "hcall-term");
927     add_str(hypertas, "hcall-dabr");
928     add_str(hypertas, "hcall-interrupt");
929     add_str(hypertas, "hcall-tce");
930     add_str(hypertas, "hcall-vio");
931     add_str(hypertas, "hcall-splpar");
932     add_str(hypertas, "hcall-join");
933     add_str(hypertas, "hcall-bulk");
934     add_str(hypertas, "hcall-set-mode");
935     add_str(hypertas, "hcall-sprg0");
936     add_str(hypertas, "hcall-copy");
937     add_str(hypertas, "hcall-debug");
938     add_str(hypertas, "hcall-vphn");
939     add_str(qemu_hypertas, "hcall-memop1");
940 
941     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
942         add_str(hypertas, "hcall-multi-tce");
943     }
944 
945     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
946         add_str(hypertas, "hcall-hpt-resize");
947     }
948 
949     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
950                      hypertas->str, hypertas->len));
951     g_string_free(hypertas, TRUE);
952     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
953                      qemu_hypertas->str, qemu_hypertas->len));
954     g_string_free(qemu_hypertas, TRUE);
955 
956     if (smc->pre_5_1_assoc_refpoints) {
957         nr_refpoints = 2;
958     }
959 
960     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
961                      refpoints, nr_refpoints * sizeof(refpoints[0])));
962 
963     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
964                      maxdomains, sizeof(maxdomains)));
965 
966     /*
967      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
968      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
969      *
970      * The system reset requirements are driven by existing Linux and PowerVM
971      * implementation which (contrary to PAPR) saves r3 in the error log
972      * structure like machine check, so Linux expects to find the saved r3
973      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
974      * does not look at the error value).
975      *
976      * System reset interrupts are not subject to interlock like machine
977      * check, so this memory area could be corrupted if the sreset is
978      * interrupted by a machine check (or vice versa) if it was shared. To
979      * prevent this, system reset uses per-CPU areas for the sreset save
980      * area. A system reset that interrupts a system reset handler could
981      * still overwrite this area, but Linux doesn't try to recover in that
982      * case anyway.
983      *
984      * The extra 8 bytes is required because Linux's FWNMI error log check
985      * is off-by-one.
986      */
987     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
988 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
989     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
990                           RTAS_ERROR_LOG_MAX));
991     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
992                           RTAS_EVENT_SCAN_RATE));
993 
994     g_assert(msi_nonbroken);
995     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
996 
997     /*
998      * According to PAPR, rtas ibm,os-term does not guarantee a return
999      * back to the guest cpu.
1000      *
1001      * While an additional ibm,extended-os-term property indicates
1002      * that rtas call return will always occur. Set this property.
1003      */
1004     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1005 
1006     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1007                      lrdr_capacity, sizeof(lrdr_capacity)));
1008 
1009     spapr_dt_rtas_tokens(fdt, rtas);
1010 }
1011 
1012 /*
1013  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1014  * and the XIVE features that the guest may request and thus the valid
1015  * values for bytes 23..26 of option vector 5:
1016  */
1017 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1018                                           int chosen)
1019 {
1020     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1021 
1022     char val[2 * 4] = {
1023         23, 0x00, /* XICS / XIVE mode */
1024         24, 0x00, /* Hash/Radix, filled in below. */
1025         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1026         26, 0x40, /* Radix options: GTSE == yes. */
1027     };
1028 
1029     if (spapr->irq->xics && spapr->irq->xive) {
1030         val[1] = SPAPR_OV5_XIVE_BOTH;
1031     } else if (spapr->irq->xive) {
1032         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1033     } else {
1034         assert(spapr->irq->xics);
1035         val[1] = SPAPR_OV5_XIVE_LEGACY;
1036     }
1037 
1038     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1039                           first_ppc_cpu->compat_pvr)) {
1040         /*
1041          * If we're in a pre POWER9 compat mode then the guest should
1042          * do hash and use the legacy interrupt mode
1043          */
1044         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1045         val[3] = 0x00; /* Hash */
1046     } else if (kvm_enabled()) {
1047         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1048             val[3] = 0x80; /* OV5_MMU_BOTH */
1049         } else if (kvmppc_has_cap_mmu_radix()) {
1050             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1051         } else {
1052             val[3] = 0x00; /* Hash */
1053         }
1054     } else {
1055         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1056         val[3] = 0xC0;
1057     }
1058     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1059                      val, sizeof(val)));
1060 }
1061 
1062 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1063 {
1064     MachineState *machine = MACHINE(spapr);
1065     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1066     int chosen;
1067 
1068     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1069 
1070     if (reset) {
1071         const char *boot_device = machine->boot_order;
1072         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1073         size_t cb = 0;
1074         char *bootlist = get_boot_devices_list(&cb);
1075 
1076         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1077             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1078                                     machine->kernel_cmdline));
1079         }
1080 
1081         if (spapr->initrd_size) {
1082             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1083                                   spapr->initrd_base));
1084             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1085                                   spapr->initrd_base + spapr->initrd_size));
1086         }
1087 
1088         if (spapr->kernel_size) {
1089             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1090                                   cpu_to_be64(spapr->kernel_size) };
1091 
1092             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1093                          &kprop, sizeof(kprop)));
1094             if (spapr->kernel_le) {
1095                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1096             }
1097         }
1098         if (boot_menu) {
1099             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1100         }
1101         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1102         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1103         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1104 
1105         if (cb && bootlist) {
1106             int i;
1107 
1108             for (i = 0; i < cb; i++) {
1109                 if (bootlist[i] == '\n') {
1110                     bootlist[i] = ' ';
1111                 }
1112             }
1113             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1114         }
1115 
1116         if (boot_device && strlen(boot_device)) {
1117             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1118         }
1119 
1120         if (!spapr->has_graphics && stdout_path) {
1121             /*
1122              * "linux,stdout-path" and "stdout" properties are
1123              * deprecated by linux kernel. New platforms should only
1124              * use the "stdout-path" property. Set the new property
1125              * and continue using older property to remain compatible
1126              * with the existing firmware.
1127              */
1128             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1129             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1130         }
1131 
1132         /*
1133          * We can deal with BAR reallocation just fine, advertise it
1134          * to the guest
1135          */
1136         if (smc->linux_pci_probe) {
1137             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1138         }
1139 
1140         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1141 
1142         g_free(stdout_path);
1143         g_free(bootlist);
1144     }
1145 
1146     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1147 }
1148 
1149 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1150 {
1151     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1152      * KVM to work under pHyp with some guest co-operation */
1153     int hypervisor;
1154     uint8_t hypercall[16];
1155 
1156     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1157     /* indicate KVM hypercall interface */
1158     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1159     if (kvmppc_has_cap_fixup_hcalls()) {
1160         /*
1161          * Older KVM versions with older guest kernels were broken
1162          * with the magic page, don't allow the guest to map it.
1163          */
1164         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1165                                   sizeof(hypercall))) {
1166             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1167                              hypercall, sizeof(hypercall)));
1168         }
1169     }
1170 }
1171 
1172 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1173 {
1174     MachineState *machine = MACHINE(spapr);
1175     MachineClass *mc = MACHINE_GET_CLASS(machine);
1176     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1177     int ret;
1178     void *fdt;
1179     SpaprPhbState *phb;
1180     char *buf;
1181 
1182     fdt = g_malloc0(space);
1183     _FDT((fdt_create_empty_tree(fdt, space)));
1184 
1185     /* Root node */
1186     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1187     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1188     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1189 
1190     /* Guest UUID & Name*/
1191     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1192     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1193     if (qemu_uuid_set) {
1194         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1195     }
1196     g_free(buf);
1197 
1198     if (qemu_get_vm_name()) {
1199         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1200                                 qemu_get_vm_name()));
1201     }
1202 
1203     /* Host Model & Serial Number */
1204     if (spapr->host_model) {
1205         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1206     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1207         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1208         g_free(buf);
1209     }
1210 
1211     if (spapr->host_serial) {
1212         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1213     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1214         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1215         g_free(buf);
1216     }
1217 
1218     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1219     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1220 
1221     /* /interrupt controller */
1222     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1223 
1224     ret = spapr_dt_memory(spapr, fdt);
1225     if (ret < 0) {
1226         error_report("couldn't setup memory nodes in fdt");
1227         exit(1);
1228     }
1229 
1230     /* /vdevice */
1231     spapr_dt_vdevice(spapr->vio_bus, fdt);
1232 
1233     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1234         ret = spapr_dt_rng(fdt);
1235         if (ret < 0) {
1236             error_report("could not set up rng device in the fdt");
1237             exit(1);
1238         }
1239     }
1240 
1241     QLIST_FOREACH(phb, &spapr->phbs, list) {
1242         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1243         if (ret < 0) {
1244             error_report("couldn't setup PCI devices in fdt");
1245             exit(1);
1246         }
1247     }
1248 
1249     spapr_dt_cpus(fdt, spapr);
1250 
1251     if (smc->dr_lmb_enabled) {
1252         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1253     }
1254 
1255     if (mc->has_hotpluggable_cpus) {
1256         int offset = fdt_path_offset(fdt, "/cpus");
1257         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1258         if (ret < 0) {
1259             error_report("Couldn't set up CPU DR device tree properties");
1260             exit(1);
1261         }
1262     }
1263 
1264     /* /event-sources */
1265     spapr_dt_events(spapr, fdt);
1266 
1267     /* /rtas */
1268     spapr_dt_rtas(spapr, fdt);
1269 
1270     /* /chosen */
1271     spapr_dt_chosen(spapr, fdt, reset);
1272 
1273     /* /hypervisor */
1274     if (kvm_enabled()) {
1275         spapr_dt_hypervisor(spapr, fdt);
1276     }
1277 
1278     /* Build memory reserve map */
1279     if (reset) {
1280         if (spapr->kernel_size) {
1281             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1282                                   spapr->kernel_size)));
1283         }
1284         if (spapr->initrd_size) {
1285             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1286                                   spapr->initrd_size)));
1287         }
1288     }
1289 
1290     if (smc->dr_phb_enabled) {
1291         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1292         if (ret < 0) {
1293             error_report("Couldn't set up PHB DR device tree properties");
1294             exit(1);
1295         }
1296     }
1297 
1298     /* NVDIMM devices */
1299     if (mc->nvdimm_supported) {
1300         spapr_dt_persistent_memory(fdt);
1301     }
1302 
1303     return fdt;
1304 }
1305 
1306 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1307 {
1308     SpaprMachineState *spapr = opaque;
1309 
1310     return (addr & 0x0fffffff) + spapr->kernel_addr;
1311 }
1312 
1313 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1314                                     PowerPCCPU *cpu)
1315 {
1316     CPUPPCState *env = &cpu->env;
1317 
1318     /* The TCG path should also be holding the BQL at this point */
1319     g_assert(qemu_mutex_iothread_locked());
1320 
1321     if (msr_pr) {
1322         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1323         env->gpr[3] = H_PRIVILEGE;
1324     } else {
1325         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1326     }
1327 }
1328 
1329 struct LPCRSyncState {
1330     target_ulong value;
1331     target_ulong mask;
1332 };
1333 
1334 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1335 {
1336     struct LPCRSyncState *s = arg.host_ptr;
1337     PowerPCCPU *cpu = POWERPC_CPU(cs);
1338     CPUPPCState *env = &cpu->env;
1339     target_ulong lpcr;
1340 
1341     cpu_synchronize_state(cs);
1342     lpcr = env->spr[SPR_LPCR];
1343     lpcr &= ~s->mask;
1344     lpcr |= s->value;
1345     ppc_store_lpcr(cpu, lpcr);
1346 }
1347 
1348 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1349 {
1350     CPUState *cs;
1351     struct LPCRSyncState s = {
1352         .value = value,
1353         .mask = mask
1354     };
1355     CPU_FOREACH(cs) {
1356         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1357     }
1358 }
1359 
1360 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1361 {
1362     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1363 
1364     /* Copy PATE1:GR into PATE0:HR */
1365     entry->dw0 = spapr->patb_entry & PATE0_HR;
1366     entry->dw1 = spapr->patb_entry;
1367 }
1368 
1369 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1370 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1371 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1372 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1373 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1374 
1375 /*
1376  * Get the fd to access the kernel htab, re-opening it if necessary
1377  */
1378 static int get_htab_fd(SpaprMachineState *spapr)
1379 {
1380     Error *local_err = NULL;
1381 
1382     if (spapr->htab_fd >= 0) {
1383         return spapr->htab_fd;
1384     }
1385 
1386     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1387     if (spapr->htab_fd < 0) {
1388         error_report_err(local_err);
1389     }
1390 
1391     return spapr->htab_fd;
1392 }
1393 
1394 void close_htab_fd(SpaprMachineState *spapr)
1395 {
1396     if (spapr->htab_fd >= 0) {
1397         close(spapr->htab_fd);
1398     }
1399     spapr->htab_fd = -1;
1400 }
1401 
1402 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1403 {
1404     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1405 
1406     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1407 }
1408 
1409 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1410 {
1411     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1412 
1413     assert(kvm_enabled());
1414 
1415     if (!spapr->htab) {
1416         return 0;
1417     }
1418 
1419     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1420 }
1421 
1422 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1423                                                 hwaddr ptex, int n)
1424 {
1425     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1426     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1427 
1428     if (!spapr->htab) {
1429         /*
1430          * HTAB is controlled by KVM. Fetch into temporary buffer
1431          */
1432         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1433         kvmppc_read_hptes(hptes, ptex, n);
1434         return hptes;
1435     }
1436 
1437     /*
1438      * HTAB is controlled by QEMU. Just point to the internally
1439      * accessible PTEG.
1440      */
1441     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1442 }
1443 
1444 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1445                               const ppc_hash_pte64_t *hptes,
1446                               hwaddr ptex, int n)
1447 {
1448     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1449 
1450     if (!spapr->htab) {
1451         g_free((void *)hptes);
1452     }
1453 
1454     /* Nothing to do for qemu managed HPT */
1455 }
1456 
1457 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1458                       uint64_t pte0, uint64_t pte1)
1459 {
1460     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1461     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1462 
1463     if (!spapr->htab) {
1464         kvmppc_write_hpte(ptex, pte0, pte1);
1465     } else {
1466         if (pte0 & HPTE64_V_VALID) {
1467             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1468             /*
1469              * When setting valid, we write PTE1 first. This ensures
1470              * proper synchronization with the reading code in
1471              * ppc_hash64_pteg_search()
1472              */
1473             smp_wmb();
1474             stq_p(spapr->htab + offset, pte0);
1475         } else {
1476             stq_p(spapr->htab + offset, pte0);
1477             /*
1478              * When clearing it we set PTE0 first. This ensures proper
1479              * synchronization with the reading code in
1480              * ppc_hash64_pteg_search()
1481              */
1482             smp_wmb();
1483             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1484         }
1485     }
1486 }
1487 
1488 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1489                              uint64_t pte1)
1490 {
1491     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1492     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1493 
1494     if (!spapr->htab) {
1495         /* There should always be a hash table when this is called */
1496         error_report("spapr_hpte_set_c called with no hash table !");
1497         return;
1498     }
1499 
1500     /* The HW performs a non-atomic byte update */
1501     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1502 }
1503 
1504 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1505                              uint64_t pte1)
1506 {
1507     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1508     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1509 
1510     if (!spapr->htab) {
1511         /* There should always be a hash table when this is called */
1512         error_report("spapr_hpte_set_r called with no hash table !");
1513         return;
1514     }
1515 
1516     /* The HW performs a non-atomic byte update */
1517     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1518 }
1519 
1520 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1521 {
1522     int shift;
1523 
1524     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1525      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1526      * that's much more than is needed for Linux guests */
1527     shift = ctz64(pow2ceil(ramsize)) - 7;
1528     shift = MAX(shift, 18); /* Minimum architected size */
1529     shift = MIN(shift, 46); /* Maximum architected size */
1530     return shift;
1531 }
1532 
1533 void spapr_free_hpt(SpaprMachineState *spapr)
1534 {
1535     g_free(spapr->htab);
1536     spapr->htab = NULL;
1537     spapr->htab_shift = 0;
1538     close_htab_fd(spapr);
1539 }
1540 
1541 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1542                           Error **errp)
1543 {
1544     long rc;
1545 
1546     /* Clean up any HPT info from a previous boot */
1547     spapr_free_hpt(spapr);
1548 
1549     rc = kvmppc_reset_htab(shift);
1550     if (rc < 0) {
1551         /* kernel-side HPT needed, but couldn't allocate one */
1552         error_setg_errno(errp, errno,
1553                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1554                          shift);
1555         /* This is almost certainly fatal, but if the caller really
1556          * wants to carry on with shift == 0, it's welcome to try */
1557     } else if (rc > 0) {
1558         /* kernel-side HPT allocated */
1559         if (rc != shift) {
1560             error_setg(errp,
1561                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1562                        shift, rc);
1563         }
1564 
1565         spapr->htab_shift = shift;
1566         spapr->htab = NULL;
1567     } else {
1568         /* kernel-side HPT not needed, allocate in userspace instead */
1569         size_t size = 1ULL << shift;
1570         int i;
1571 
1572         spapr->htab = qemu_memalign(size, size);
1573         if (!spapr->htab) {
1574             error_setg_errno(errp, errno,
1575                              "Could not allocate HPT of order %d", shift);
1576             return;
1577         }
1578 
1579         memset(spapr->htab, 0, size);
1580         spapr->htab_shift = shift;
1581 
1582         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1583             DIRTY_HPTE(HPTE(spapr->htab, i));
1584         }
1585     }
1586     /* We're setting up a hash table, so that means we're not radix */
1587     spapr->patb_entry = 0;
1588     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1589 }
1590 
1591 void spapr_setup_hpt(SpaprMachineState *spapr)
1592 {
1593     int hpt_shift;
1594 
1595     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1596         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1597     } else {
1598         uint64_t current_ram_size;
1599 
1600         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1601         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1602     }
1603     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1604 
1605     if (kvm_enabled()) {
1606         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1607 
1608         /* Check our RMA fits in the possible VRMA */
1609         if (vrma_limit < spapr->rma_size) {
1610             error_report("Unable to create %" HWADDR_PRIu
1611                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1612                          spapr->rma_size / MiB, vrma_limit / MiB);
1613             exit(EXIT_FAILURE);
1614         }
1615     }
1616 }
1617 
1618 static int spapr_reset_drcs(Object *child, void *opaque)
1619 {
1620     SpaprDrc *drc =
1621         (SpaprDrc *) object_dynamic_cast(child,
1622                                                  TYPE_SPAPR_DR_CONNECTOR);
1623 
1624     if (drc) {
1625         spapr_drc_reset(drc);
1626     }
1627 
1628     return 0;
1629 }
1630 
1631 static void spapr_machine_reset(MachineState *machine)
1632 {
1633     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1634     PowerPCCPU *first_ppc_cpu;
1635     hwaddr fdt_addr;
1636     void *fdt;
1637     int rc;
1638 
1639     kvmppc_svm_off(&error_fatal);
1640     spapr_caps_apply(spapr);
1641 
1642     first_ppc_cpu = POWERPC_CPU(first_cpu);
1643     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1644         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1645                               spapr->max_compat_pvr)) {
1646         /*
1647          * If using KVM with radix mode available, VCPUs can be started
1648          * without a HPT because KVM will start them in radix mode.
1649          * Set the GR bit in PATE so that we know there is no HPT.
1650          */
1651         spapr->patb_entry = PATE1_GR;
1652         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1653     } else {
1654         spapr_setup_hpt(spapr);
1655     }
1656 
1657     qemu_devices_reset();
1658 
1659     spapr_ovec_cleanup(spapr->ov5_cas);
1660     spapr->ov5_cas = spapr_ovec_new();
1661 
1662     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1663 
1664     /*
1665      * This is fixing some of the default configuration of the XIVE
1666      * devices. To be called after the reset of the machine devices.
1667      */
1668     spapr_irq_reset(spapr, &error_fatal);
1669 
1670     /*
1671      * There is no CAS under qtest. Simulate one to please the code that
1672      * depends on spapr->ov5_cas. This is especially needed to test device
1673      * unplug, so we do that before resetting the DRCs.
1674      */
1675     if (qtest_enabled()) {
1676         spapr_ovec_cleanup(spapr->ov5_cas);
1677         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1678     }
1679 
1680     /* DRC reset may cause a device to be unplugged. This will cause troubles
1681      * if this device is used by another device (eg, a running vhost backend
1682      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1683      * situations, we reset DRCs after all devices have been reset.
1684      */
1685     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1686 
1687     spapr_clear_pending_events(spapr);
1688 
1689     /*
1690      * We place the device tree and RTAS just below either the top of the RMA,
1691      * or just below 2GB, whichever is lower, so that it can be
1692      * processed with 32-bit real mode code if necessary
1693      */
1694     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1695 
1696     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1697 
1698     rc = fdt_pack(fdt);
1699 
1700     /* Should only fail if we've built a corrupted tree */
1701     assert(rc == 0);
1702 
1703     /* Load the fdt */
1704     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1705     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1706     g_free(spapr->fdt_blob);
1707     spapr->fdt_size = fdt_totalsize(fdt);
1708     spapr->fdt_initial_size = spapr->fdt_size;
1709     spapr->fdt_blob = fdt;
1710 
1711     /* Set up the entry state */
1712     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1713     first_ppc_cpu->env.gpr[5] = 0;
1714 
1715     spapr->fwnmi_system_reset_addr = -1;
1716     spapr->fwnmi_machine_check_addr = -1;
1717     spapr->fwnmi_machine_check_interlock = -1;
1718 
1719     /* Signal all vCPUs waiting on this condition */
1720     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1721 
1722     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1723 }
1724 
1725 static void spapr_create_nvram(SpaprMachineState *spapr)
1726 {
1727     DeviceState *dev = qdev_new("spapr-nvram");
1728     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1729 
1730     if (dinfo) {
1731         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1732                                 &error_fatal);
1733     }
1734 
1735     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1736 
1737     spapr->nvram = (struct SpaprNvram *)dev;
1738 }
1739 
1740 static void spapr_rtc_create(SpaprMachineState *spapr)
1741 {
1742     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1743                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1744                                        &error_fatal, NULL);
1745     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1746     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1747                               "date");
1748 }
1749 
1750 /* Returns whether we want to use VGA or not */
1751 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1752 {
1753     switch (vga_interface_type) {
1754     case VGA_NONE:
1755         return false;
1756     case VGA_DEVICE:
1757         return true;
1758     case VGA_STD:
1759     case VGA_VIRTIO:
1760     case VGA_CIRRUS:
1761         return pci_vga_init(pci_bus) != NULL;
1762     default:
1763         error_setg(errp,
1764                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1765         return false;
1766     }
1767 }
1768 
1769 static int spapr_pre_load(void *opaque)
1770 {
1771     int rc;
1772 
1773     rc = spapr_caps_pre_load(opaque);
1774     if (rc) {
1775         return rc;
1776     }
1777 
1778     return 0;
1779 }
1780 
1781 static int spapr_post_load(void *opaque, int version_id)
1782 {
1783     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1784     int err = 0;
1785 
1786     err = spapr_caps_post_migration(spapr);
1787     if (err) {
1788         return err;
1789     }
1790 
1791     /*
1792      * In earlier versions, there was no separate qdev for the PAPR
1793      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1794      * So when migrating from those versions, poke the incoming offset
1795      * value into the RTC device
1796      */
1797     if (version_id < 3) {
1798         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1799         if (err) {
1800             return err;
1801         }
1802     }
1803 
1804     if (kvm_enabled() && spapr->patb_entry) {
1805         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1806         bool radix = !!(spapr->patb_entry & PATE1_GR);
1807         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1808 
1809         /*
1810          * Update LPCR:HR and UPRT as they may not be set properly in
1811          * the stream
1812          */
1813         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1814                             LPCR_HR | LPCR_UPRT);
1815 
1816         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1817         if (err) {
1818             error_report("Process table config unsupported by the host");
1819             return -EINVAL;
1820         }
1821     }
1822 
1823     err = spapr_irq_post_load(spapr, version_id);
1824     if (err) {
1825         return err;
1826     }
1827 
1828     return err;
1829 }
1830 
1831 static int spapr_pre_save(void *opaque)
1832 {
1833     int rc;
1834 
1835     rc = spapr_caps_pre_save(opaque);
1836     if (rc) {
1837         return rc;
1838     }
1839 
1840     return 0;
1841 }
1842 
1843 static bool version_before_3(void *opaque, int version_id)
1844 {
1845     return version_id < 3;
1846 }
1847 
1848 static bool spapr_pending_events_needed(void *opaque)
1849 {
1850     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1851     return !QTAILQ_EMPTY(&spapr->pending_events);
1852 }
1853 
1854 static const VMStateDescription vmstate_spapr_event_entry = {
1855     .name = "spapr_event_log_entry",
1856     .version_id = 1,
1857     .minimum_version_id = 1,
1858     .fields = (VMStateField[]) {
1859         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1860         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1861         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1862                                      NULL, extended_length),
1863         VMSTATE_END_OF_LIST()
1864     },
1865 };
1866 
1867 static const VMStateDescription vmstate_spapr_pending_events = {
1868     .name = "spapr_pending_events",
1869     .version_id = 1,
1870     .minimum_version_id = 1,
1871     .needed = spapr_pending_events_needed,
1872     .fields = (VMStateField[]) {
1873         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1874                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1875         VMSTATE_END_OF_LIST()
1876     },
1877 };
1878 
1879 static bool spapr_ov5_cas_needed(void *opaque)
1880 {
1881     SpaprMachineState *spapr = opaque;
1882     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1883     bool cas_needed;
1884 
1885     /* Prior to the introduction of SpaprOptionVector, we had two option
1886      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1887      * Both of these options encode machine topology into the device-tree
1888      * in such a way that the now-booted OS should still be able to interact
1889      * appropriately with QEMU regardless of what options were actually
1890      * negotiatied on the source side.
1891      *
1892      * As such, we can avoid migrating the CAS-negotiated options if these
1893      * are the only options available on the current machine/platform.
1894      * Since these are the only options available for pseries-2.7 and
1895      * earlier, this allows us to maintain old->new/new->old migration
1896      * compatibility.
1897      *
1898      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1899      * via default pseries-2.8 machines and explicit command-line parameters.
1900      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1901      * of the actual CAS-negotiated values to continue working properly. For
1902      * example, availability of memory unplug depends on knowing whether
1903      * OV5_HP_EVT was negotiated via CAS.
1904      *
1905      * Thus, for any cases where the set of available CAS-negotiatable
1906      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1907      * include the CAS-negotiated options in the migration stream, unless
1908      * if they affect boot time behaviour only.
1909      */
1910     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1911     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1912     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1913 
1914     /* We need extra information if we have any bits outside the mask
1915      * defined above */
1916     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1917 
1918     spapr_ovec_cleanup(ov5_mask);
1919 
1920     return cas_needed;
1921 }
1922 
1923 static const VMStateDescription vmstate_spapr_ov5_cas = {
1924     .name = "spapr_option_vector_ov5_cas",
1925     .version_id = 1,
1926     .minimum_version_id = 1,
1927     .needed = spapr_ov5_cas_needed,
1928     .fields = (VMStateField[]) {
1929         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1930                                  vmstate_spapr_ovec, SpaprOptionVector),
1931         VMSTATE_END_OF_LIST()
1932     },
1933 };
1934 
1935 static bool spapr_patb_entry_needed(void *opaque)
1936 {
1937     SpaprMachineState *spapr = opaque;
1938 
1939     return !!spapr->patb_entry;
1940 }
1941 
1942 static const VMStateDescription vmstate_spapr_patb_entry = {
1943     .name = "spapr_patb_entry",
1944     .version_id = 1,
1945     .minimum_version_id = 1,
1946     .needed = spapr_patb_entry_needed,
1947     .fields = (VMStateField[]) {
1948         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1949         VMSTATE_END_OF_LIST()
1950     },
1951 };
1952 
1953 static bool spapr_irq_map_needed(void *opaque)
1954 {
1955     SpaprMachineState *spapr = opaque;
1956 
1957     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1958 }
1959 
1960 static const VMStateDescription vmstate_spapr_irq_map = {
1961     .name = "spapr_irq_map",
1962     .version_id = 1,
1963     .minimum_version_id = 1,
1964     .needed = spapr_irq_map_needed,
1965     .fields = (VMStateField[]) {
1966         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1967         VMSTATE_END_OF_LIST()
1968     },
1969 };
1970 
1971 static bool spapr_dtb_needed(void *opaque)
1972 {
1973     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1974 
1975     return smc->update_dt_enabled;
1976 }
1977 
1978 static int spapr_dtb_pre_load(void *opaque)
1979 {
1980     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1981 
1982     g_free(spapr->fdt_blob);
1983     spapr->fdt_blob = NULL;
1984     spapr->fdt_size = 0;
1985 
1986     return 0;
1987 }
1988 
1989 static const VMStateDescription vmstate_spapr_dtb = {
1990     .name = "spapr_dtb",
1991     .version_id = 1,
1992     .minimum_version_id = 1,
1993     .needed = spapr_dtb_needed,
1994     .pre_load = spapr_dtb_pre_load,
1995     .fields = (VMStateField[]) {
1996         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1997         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1998         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1999                                      fdt_size),
2000         VMSTATE_END_OF_LIST()
2001     },
2002 };
2003 
2004 static bool spapr_fwnmi_needed(void *opaque)
2005 {
2006     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2007 
2008     return spapr->fwnmi_machine_check_addr != -1;
2009 }
2010 
2011 static int spapr_fwnmi_pre_save(void *opaque)
2012 {
2013     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2014 
2015     /*
2016      * Check if machine check handling is in progress and print a
2017      * warning message.
2018      */
2019     if (spapr->fwnmi_machine_check_interlock != -1) {
2020         warn_report("A machine check is being handled during migration. The"
2021                 "handler may run and log hardware error on the destination");
2022     }
2023 
2024     return 0;
2025 }
2026 
2027 static const VMStateDescription vmstate_spapr_fwnmi = {
2028     .name = "spapr_fwnmi",
2029     .version_id = 1,
2030     .minimum_version_id = 1,
2031     .needed = spapr_fwnmi_needed,
2032     .pre_save = spapr_fwnmi_pre_save,
2033     .fields = (VMStateField[]) {
2034         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2035         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2036         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2037         VMSTATE_END_OF_LIST()
2038     },
2039 };
2040 
2041 static const VMStateDescription vmstate_spapr = {
2042     .name = "spapr",
2043     .version_id = 3,
2044     .minimum_version_id = 1,
2045     .pre_load = spapr_pre_load,
2046     .post_load = spapr_post_load,
2047     .pre_save = spapr_pre_save,
2048     .fields = (VMStateField[]) {
2049         /* used to be @next_irq */
2050         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2051 
2052         /* RTC offset */
2053         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2054 
2055         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2056         VMSTATE_END_OF_LIST()
2057     },
2058     .subsections = (const VMStateDescription*[]) {
2059         &vmstate_spapr_ov5_cas,
2060         &vmstate_spapr_patb_entry,
2061         &vmstate_spapr_pending_events,
2062         &vmstate_spapr_cap_htm,
2063         &vmstate_spapr_cap_vsx,
2064         &vmstate_spapr_cap_dfp,
2065         &vmstate_spapr_cap_cfpc,
2066         &vmstate_spapr_cap_sbbc,
2067         &vmstate_spapr_cap_ibs,
2068         &vmstate_spapr_cap_hpt_maxpagesize,
2069         &vmstate_spapr_irq_map,
2070         &vmstate_spapr_cap_nested_kvm_hv,
2071         &vmstate_spapr_dtb,
2072         &vmstate_spapr_cap_large_decr,
2073         &vmstate_spapr_cap_ccf_assist,
2074         &vmstate_spapr_cap_fwnmi,
2075         &vmstate_spapr_fwnmi,
2076         NULL
2077     }
2078 };
2079 
2080 static int htab_save_setup(QEMUFile *f, void *opaque)
2081 {
2082     SpaprMachineState *spapr = opaque;
2083 
2084     /* "Iteration" header */
2085     if (!spapr->htab_shift) {
2086         qemu_put_be32(f, -1);
2087     } else {
2088         qemu_put_be32(f, spapr->htab_shift);
2089     }
2090 
2091     if (spapr->htab) {
2092         spapr->htab_save_index = 0;
2093         spapr->htab_first_pass = true;
2094     } else {
2095         if (spapr->htab_shift) {
2096             assert(kvm_enabled());
2097         }
2098     }
2099 
2100 
2101     return 0;
2102 }
2103 
2104 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2105                             int chunkstart, int n_valid, int n_invalid)
2106 {
2107     qemu_put_be32(f, chunkstart);
2108     qemu_put_be16(f, n_valid);
2109     qemu_put_be16(f, n_invalid);
2110     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2111                     HASH_PTE_SIZE_64 * n_valid);
2112 }
2113 
2114 static void htab_save_end_marker(QEMUFile *f)
2115 {
2116     qemu_put_be32(f, 0);
2117     qemu_put_be16(f, 0);
2118     qemu_put_be16(f, 0);
2119 }
2120 
2121 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2122                                  int64_t max_ns)
2123 {
2124     bool has_timeout = max_ns != -1;
2125     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2126     int index = spapr->htab_save_index;
2127     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2128 
2129     assert(spapr->htab_first_pass);
2130 
2131     do {
2132         int chunkstart;
2133 
2134         /* Consume invalid HPTEs */
2135         while ((index < htabslots)
2136                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2137             CLEAN_HPTE(HPTE(spapr->htab, index));
2138             index++;
2139         }
2140 
2141         /* Consume valid HPTEs */
2142         chunkstart = index;
2143         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2144                && HPTE_VALID(HPTE(spapr->htab, index))) {
2145             CLEAN_HPTE(HPTE(spapr->htab, index));
2146             index++;
2147         }
2148 
2149         if (index > chunkstart) {
2150             int n_valid = index - chunkstart;
2151 
2152             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2153 
2154             if (has_timeout &&
2155                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2156                 break;
2157             }
2158         }
2159     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2160 
2161     if (index >= htabslots) {
2162         assert(index == htabslots);
2163         index = 0;
2164         spapr->htab_first_pass = false;
2165     }
2166     spapr->htab_save_index = index;
2167 }
2168 
2169 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2170                                 int64_t max_ns)
2171 {
2172     bool final = max_ns < 0;
2173     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2174     int examined = 0, sent = 0;
2175     int index = spapr->htab_save_index;
2176     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2177 
2178     assert(!spapr->htab_first_pass);
2179 
2180     do {
2181         int chunkstart, invalidstart;
2182 
2183         /* Consume non-dirty HPTEs */
2184         while ((index < htabslots)
2185                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2186             index++;
2187             examined++;
2188         }
2189 
2190         chunkstart = index;
2191         /* Consume valid dirty HPTEs */
2192         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2193                && HPTE_DIRTY(HPTE(spapr->htab, index))
2194                && HPTE_VALID(HPTE(spapr->htab, index))) {
2195             CLEAN_HPTE(HPTE(spapr->htab, index));
2196             index++;
2197             examined++;
2198         }
2199 
2200         invalidstart = index;
2201         /* Consume invalid dirty HPTEs */
2202         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2203                && HPTE_DIRTY(HPTE(spapr->htab, index))
2204                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2205             CLEAN_HPTE(HPTE(spapr->htab, index));
2206             index++;
2207             examined++;
2208         }
2209 
2210         if (index > chunkstart) {
2211             int n_valid = invalidstart - chunkstart;
2212             int n_invalid = index - invalidstart;
2213 
2214             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2215             sent += index - chunkstart;
2216 
2217             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2218                 break;
2219             }
2220         }
2221 
2222         if (examined >= htabslots) {
2223             break;
2224         }
2225 
2226         if (index >= htabslots) {
2227             assert(index == htabslots);
2228             index = 0;
2229         }
2230     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2231 
2232     if (index >= htabslots) {
2233         assert(index == htabslots);
2234         index = 0;
2235     }
2236 
2237     spapr->htab_save_index = index;
2238 
2239     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2240 }
2241 
2242 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2243 #define MAX_KVM_BUF_SIZE    2048
2244 
2245 static int htab_save_iterate(QEMUFile *f, void *opaque)
2246 {
2247     SpaprMachineState *spapr = opaque;
2248     int fd;
2249     int rc = 0;
2250 
2251     /* Iteration header */
2252     if (!spapr->htab_shift) {
2253         qemu_put_be32(f, -1);
2254         return 1;
2255     } else {
2256         qemu_put_be32(f, 0);
2257     }
2258 
2259     if (!spapr->htab) {
2260         assert(kvm_enabled());
2261 
2262         fd = get_htab_fd(spapr);
2263         if (fd < 0) {
2264             return fd;
2265         }
2266 
2267         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2268         if (rc < 0) {
2269             return rc;
2270         }
2271     } else  if (spapr->htab_first_pass) {
2272         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2273     } else {
2274         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2275     }
2276 
2277     htab_save_end_marker(f);
2278 
2279     return rc;
2280 }
2281 
2282 static int htab_save_complete(QEMUFile *f, void *opaque)
2283 {
2284     SpaprMachineState *spapr = opaque;
2285     int fd;
2286 
2287     /* Iteration header */
2288     if (!spapr->htab_shift) {
2289         qemu_put_be32(f, -1);
2290         return 0;
2291     } else {
2292         qemu_put_be32(f, 0);
2293     }
2294 
2295     if (!spapr->htab) {
2296         int rc;
2297 
2298         assert(kvm_enabled());
2299 
2300         fd = get_htab_fd(spapr);
2301         if (fd < 0) {
2302             return fd;
2303         }
2304 
2305         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2306         if (rc < 0) {
2307             return rc;
2308         }
2309     } else {
2310         if (spapr->htab_first_pass) {
2311             htab_save_first_pass(f, spapr, -1);
2312         }
2313         htab_save_later_pass(f, spapr, -1);
2314     }
2315 
2316     /* End marker */
2317     htab_save_end_marker(f);
2318 
2319     return 0;
2320 }
2321 
2322 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2323 {
2324     SpaprMachineState *spapr = opaque;
2325     uint32_t section_hdr;
2326     int fd = -1;
2327     Error *local_err = NULL;
2328 
2329     if (version_id < 1 || version_id > 1) {
2330         error_report("htab_load() bad version");
2331         return -EINVAL;
2332     }
2333 
2334     section_hdr = qemu_get_be32(f);
2335 
2336     if (section_hdr == -1) {
2337         spapr_free_hpt(spapr);
2338         return 0;
2339     }
2340 
2341     if (section_hdr) {
2342         /* First section gives the htab size */
2343         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2344         if (local_err) {
2345             error_report_err(local_err);
2346             return -EINVAL;
2347         }
2348         return 0;
2349     }
2350 
2351     if (!spapr->htab) {
2352         assert(kvm_enabled());
2353 
2354         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2355         if (fd < 0) {
2356             error_report_err(local_err);
2357             return fd;
2358         }
2359     }
2360 
2361     while (true) {
2362         uint32_t index;
2363         uint16_t n_valid, n_invalid;
2364 
2365         index = qemu_get_be32(f);
2366         n_valid = qemu_get_be16(f);
2367         n_invalid = qemu_get_be16(f);
2368 
2369         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2370             /* End of Stream */
2371             break;
2372         }
2373 
2374         if ((index + n_valid + n_invalid) >
2375             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2376             /* Bad index in stream */
2377             error_report(
2378                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2379                 index, n_valid, n_invalid, spapr->htab_shift);
2380             return -EINVAL;
2381         }
2382 
2383         if (spapr->htab) {
2384             if (n_valid) {
2385                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2386                                 HASH_PTE_SIZE_64 * n_valid);
2387             }
2388             if (n_invalid) {
2389                 memset(HPTE(spapr->htab, index + n_valid), 0,
2390                        HASH_PTE_SIZE_64 * n_invalid);
2391             }
2392         } else {
2393             int rc;
2394 
2395             assert(fd >= 0);
2396 
2397             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2398             if (rc < 0) {
2399                 return rc;
2400             }
2401         }
2402     }
2403 
2404     if (!spapr->htab) {
2405         assert(fd >= 0);
2406         close(fd);
2407     }
2408 
2409     return 0;
2410 }
2411 
2412 static void htab_save_cleanup(void *opaque)
2413 {
2414     SpaprMachineState *spapr = opaque;
2415 
2416     close_htab_fd(spapr);
2417 }
2418 
2419 static SaveVMHandlers savevm_htab_handlers = {
2420     .save_setup = htab_save_setup,
2421     .save_live_iterate = htab_save_iterate,
2422     .save_live_complete_precopy = htab_save_complete,
2423     .save_cleanup = htab_save_cleanup,
2424     .load_state = htab_load,
2425 };
2426 
2427 static void spapr_boot_set(void *opaque, const char *boot_device,
2428                            Error **errp)
2429 {
2430     MachineState *machine = MACHINE(opaque);
2431     machine->boot_order = g_strdup(boot_device);
2432 }
2433 
2434 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2435 {
2436     MachineState *machine = MACHINE(spapr);
2437     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2438     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2439     int i;
2440 
2441     for (i = 0; i < nr_lmbs; i++) {
2442         uint64_t addr;
2443 
2444         addr = i * lmb_size + machine->device_memory->base;
2445         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2446                                addr / lmb_size);
2447     }
2448 }
2449 
2450 /*
2451  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2452  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2453  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2454  */
2455 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2456 {
2457     int i;
2458 
2459     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2460         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2461                    " is not aligned to %" PRIu64 " MiB",
2462                    machine->ram_size,
2463                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2464         return;
2465     }
2466 
2467     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2468         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2469                    " is not aligned to %" PRIu64 " MiB",
2470                    machine->ram_size,
2471                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2472         return;
2473     }
2474 
2475     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2476         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2477             error_setg(errp,
2478                        "Node %d memory size 0x%" PRIx64
2479                        " is not aligned to %" PRIu64 " MiB",
2480                        i, machine->numa_state->nodes[i].node_mem,
2481                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2482             return;
2483         }
2484     }
2485 }
2486 
2487 /* find cpu slot in machine->possible_cpus by core_id */
2488 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2489 {
2490     int index = id / ms->smp.threads;
2491 
2492     if (index >= ms->possible_cpus->len) {
2493         return NULL;
2494     }
2495     if (idx) {
2496         *idx = index;
2497     }
2498     return &ms->possible_cpus->cpus[index];
2499 }
2500 
2501 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2502 {
2503     MachineState *ms = MACHINE(spapr);
2504     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2505     Error *local_err = NULL;
2506     bool vsmt_user = !!spapr->vsmt;
2507     int kvm_smt = kvmppc_smt_threads();
2508     int ret;
2509     unsigned int smp_threads = ms->smp.threads;
2510 
2511     if (!kvm_enabled() && (smp_threads > 1)) {
2512         error_setg(errp, "TCG cannot support more than 1 thread/core "
2513                    "on a pseries machine");
2514         return;
2515     }
2516     if (!is_power_of_2(smp_threads)) {
2517         error_setg(errp, "Cannot support %d threads/core on a pseries "
2518                    "machine because it must be a power of 2", smp_threads);
2519         return;
2520     }
2521 
2522     /* Detemine the VSMT mode to use: */
2523     if (vsmt_user) {
2524         if (spapr->vsmt < smp_threads) {
2525             error_setg(errp, "Cannot support VSMT mode %d"
2526                        " because it must be >= threads/core (%d)",
2527                        spapr->vsmt, smp_threads);
2528             return;
2529         }
2530         /* In this case, spapr->vsmt has been set by the command line */
2531     } else if (!smc->smp_threads_vsmt) {
2532         /*
2533          * Default VSMT value is tricky, because we need it to be as
2534          * consistent as possible (for migration), but this requires
2535          * changing it for at least some existing cases.  We pick 8 as
2536          * the value that we'd get with KVM on POWER8, the
2537          * overwhelmingly common case in production systems.
2538          */
2539         spapr->vsmt = MAX(8, smp_threads);
2540     } else {
2541         spapr->vsmt = smp_threads;
2542     }
2543 
2544     /* KVM: If necessary, set the SMT mode: */
2545     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2546         ret = kvmppc_set_smt_threads(spapr->vsmt);
2547         if (ret) {
2548             /* Looks like KVM isn't able to change VSMT mode */
2549             error_setg(&local_err,
2550                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2551                        spapr->vsmt, ret);
2552             /* We can live with that if the default one is big enough
2553              * for the number of threads, and a submultiple of the one
2554              * we want.  In this case we'll waste some vcpu ids, but
2555              * behaviour will be correct */
2556             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2557                 warn_report_err(local_err);
2558             } else {
2559                 if (!vsmt_user) {
2560                     error_append_hint(&local_err,
2561                                       "On PPC, a VM with %d threads/core"
2562                                       " on a host with %d threads/core"
2563                                       " requires the use of VSMT mode %d.\n",
2564                                       smp_threads, kvm_smt, spapr->vsmt);
2565                 }
2566                 kvmppc_error_append_smt_possible_hint(&local_err);
2567                 error_propagate(errp, local_err);
2568             }
2569         }
2570     }
2571     /* else TCG: nothing to do currently */
2572 }
2573 
2574 static void spapr_init_cpus(SpaprMachineState *spapr)
2575 {
2576     MachineState *machine = MACHINE(spapr);
2577     MachineClass *mc = MACHINE_GET_CLASS(machine);
2578     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2579     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2580     const CPUArchIdList *possible_cpus;
2581     unsigned int smp_cpus = machine->smp.cpus;
2582     unsigned int smp_threads = machine->smp.threads;
2583     unsigned int max_cpus = machine->smp.max_cpus;
2584     int boot_cores_nr = smp_cpus / smp_threads;
2585     int i;
2586 
2587     possible_cpus = mc->possible_cpu_arch_ids(machine);
2588     if (mc->has_hotpluggable_cpus) {
2589         if (smp_cpus % smp_threads) {
2590             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2591                          smp_cpus, smp_threads);
2592             exit(1);
2593         }
2594         if (max_cpus % smp_threads) {
2595             error_report("max_cpus (%u) must be multiple of threads (%u)",
2596                          max_cpus, smp_threads);
2597             exit(1);
2598         }
2599     } else {
2600         if (max_cpus != smp_cpus) {
2601             error_report("This machine version does not support CPU hotplug");
2602             exit(1);
2603         }
2604         boot_cores_nr = possible_cpus->len;
2605     }
2606 
2607     if (smc->pre_2_10_has_unused_icps) {
2608         int i;
2609 
2610         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2611             /* Dummy entries get deregistered when real ICPState objects
2612              * are registered during CPU core hotplug.
2613              */
2614             pre_2_10_vmstate_register_dummy_icp(i);
2615         }
2616     }
2617 
2618     for (i = 0; i < possible_cpus->len; i++) {
2619         int core_id = i * smp_threads;
2620 
2621         if (mc->has_hotpluggable_cpus) {
2622             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2623                                    spapr_vcpu_id(spapr, core_id));
2624         }
2625 
2626         if (i < boot_cores_nr) {
2627             Object *core  = object_new(type);
2628             int nr_threads = smp_threads;
2629 
2630             /* Handle the partially filled core for older machine types */
2631             if ((i + 1) * smp_threads >= smp_cpus) {
2632                 nr_threads = smp_cpus - i * smp_threads;
2633             }
2634 
2635             object_property_set_int(core, "nr-threads", nr_threads,
2636                                     &error_fatal);
2637             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2638                                     &error_fatal);
2639             qdev_realize(DEVICE(core), NULL, &error_fatal);
2640 
2641             object_unref(core);
2642         }
2643     }
2644 }
2645 
2646 static PCIHostState *spapr_create_default_phb(void)
2647 {
2648     DeviceState *dev;
2649 
2650     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2651     qdev_prop_set_uint32(dev, "index", 0);
2652     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2653 
2654     return PCI_HOST_BRIDGE(dev);
2655 }
2656 
2657 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2658 {
2659     MachineState *machine = MACHINE(spapr);
2660     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2661     hwaddr rma_size = machine->ram_size;
2662     hwaddr node0_size = spapr_node0_size(machine);
2663 
2664     /* RMA has to fit in the first NUMA node */
2665     rma_size = MIN(rma_size, node0_size);
2666 
2667     /*
2668      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2669      * never exceed that
2670      */
2671     rma_size = MIN(rma_size, 1 * TiB);
2672 
2673     /*
2674      * Clamp the RMA size based on machine type.  This is for
2675      * migration compatibility with older qemu versions, which limited
2676      * the RMA size for complicated and mostly bad reasons.
2677      */
2678     if (smc->rma_limit) {
2679         rma_size = MIN(rma_size, smc->rma_limit);
2680     }
2681 
2682     if (rma_size < MIN_RMA_SLOF) {
2683         error_setg(errp,
2684                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2685                    "ldMiB guest RMA (Real Mode Area memory)",
2686                    MIN_RMA_SLOF / MiB);
2687         return 0;
2688     }
2689 
2690     return rma_size;
2691 }
2692 
2693 /* pSeries LPAR / sPAPR hardware init */
2694 static void spapr_machine_init(MachineState *machine)
2695 {
2696     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2697     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2698     MachineClass *mc = MACHINE_GET_CLASS(machine);
2699     const char *kernel_filename = machine->kernel_filename;
2700     const char *initrd_filename = machine->initrd_filename;
2701     PCIHostState *phb;
2702     int i;
2703     MemoryRegion *sysmem = get_system_memory();
2704     long load_limit, fw_size;
2705     char *filename;
2706     Error *resize_hpt_err = NULL;
2707 
2708     msi_nonbroken = true;
2709 
2710     QLIST_INIT(&spapr->phbs);
2711     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2712 
2713     /* Determine capabilities to run with */
2714     spapr_caps_init(spapr);
2715 
2716     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2717     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2718         /*
2719          * If the user explicitly requested a mode we should either
2720          * supply it, or fail completely (which we do below).  But if
2721          * it's not set explicitly, we reset our mode to something
2722          * that works
2723          */
2724         if (resize_hpt_err) {
2725             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2726             error_free(resize_hpt_err);
2727             resize_hpt_err = NULL;
2728         } else {
2729             spapr->resize_hpt = smc->resize_hpt_default;
2730         }
2731     }
2732 
2733     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2734 
2735     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2736         /*
2737          * User requested HPT resize, but this host can't supply it.  Bail out
2738          */
2739         error_report_err(resize_hpt_err);
2740         exit(1);
2741     }
2742     error_free(resize_hpt_err);
2743 
2744     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2745 
2746     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2747     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2748 
2749     /*
2750      * VSMT must be set in order to be able to compute VCPU ids, ie to
2751      * call spapr_max_server_number() or spapr_vcpu_id().
2752      */
2753     spapr_set_vsmt_mode(spapr, &error_fatal);
2754 
2755     /* Set up Interrupt Controller before we create the VCPUs */
2756     spapr_irq_init(spapr, &error_fatal);
2757 
2758     /* Set up containers for ibm,client-architecture-support negotiated options
2759      */
2760     spapr->ov5 = spapr_ovec_new();
2761     spapr->ov5_cas = spapr_ovec_new();
2762 
2763     if (smc->dr_lmb_enabled) {
2764         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2765         spapr_validate_node_memory(machine, &error_fatal);
2766     }
2767 
2768     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2769 
2770     /* advertise support for dedicated HP event source to guests */
2771     if (spapr->use_hotplug_event_source) {
2772         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2773     }
2774 
2775     /* advertise support for HPT resizing */
2776     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2777         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2778     }
2779 
2780     /* advertise support for ibm,dyamic-memory-v2 */
2781     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2782 
2783     /* advertise XIVE on POWER9 machines */
2784     if (spapr->irq->xive) {
2785         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2786     }
2787 
2788     /* init CPUs */
2789     spapr_init_cpus(spapr);
2790 
2791     /*
2792      * check we don't have a memory-less/cpu-less NUMA node
2793      * Firmware relies on the existing memory/cpu topology to provide the
2794      * NUMA topology to the kernel.
2795      * And the linux kernel needs to know the NUMA topology at start
2796      * to be able to hotplug CPUs later.
2797      */
2798     if (machine->numa_state->num_nodes) {
2799         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2800             /* check for memory-less node */
2801             if (machine->numa_state->nodes[i].node_mem == 0) {
2802                 CPUState *cs;
2803                 int found = 0;
2804                 /* check for cpu-less node */
2805                 CPU_FOREACH(cs) {
2806                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2807                     if (cpu->node_id == i) {
2808                         found = 1;
2809                         break;
2810                     }
2811                 }
2812                 /* memory-less and cpu-less node */
2813                 if (!found) {
2814                     error_report(
2815                        "Memory-less/cpu-less nodes are not supported (node %d)",
2816                                  i);
2817                     exit(1);
2818                 }
2819             }
2820         }
2821 
2822     }
2823 
2824     /*
2825      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2826      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2827      * called from vPHB reset handler so we initialize the counter here.
2828      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2829      * must be equally distant from any other node.
2830      * The final value of spapr->gpu_numa_id is going to be written to
2831      * max-associativity-domains in spapr_build_fdt().
2832      */
2833     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2834 
2835     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2836         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2837                               spapr->max_compat_pvr)) {
2838         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2839         /* KVM and TCG always allow GTSE with radix... */
2840         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2841     }
2842     /* ... but not with hash (currently). */
2843 
2844     if (kvm_enabled()) {
2845         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2846         kvmppc_enable_logical_ci_hcalls();
2847         kvmppc_enable_set_mode_hcall();
2848 
2849         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2850         kvmppc_enable_clear_ref_mod_hcalls();
2851 
2852         /* Enable H_PAGE_INIT */
2853         kvmppc_enable_h_page_init();
2854     }
2855 
2856     /* map RAM */
2857     memory_region_add_subregion(sysmem, 0, machine->ram);
2858 
2859     /* always allocate the device memory information */
2860     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2861 
2862     /* initialize hotplug memory address space */
2863     if (machine->ram_size < machine->maxram_size) {
2864         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2865         /*
2866          * Limit the number of hotpluggable memory slots to half the number
2867          * slots that KVM supports, leaving the other half for PCI and other
2868          * devices. However ensure that number of slots doesn't drop below 32.
2869          */
2870         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2871                            SPAPR_MAX_RAM_SLOTS;
2872 
2873         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2874             max_memslots = SPAPR_MAX_RAM_SLOTS;
2875         }
2876         if (machine->ram_slots > max_memslots) {
2877             error_report("Specified number of memory slots %"
2878                          PRIu64" exceeds max supported %d",
2879                          machine->ram_slots, max_memslots);
2880             exit(1);
2881         }
2882 
2883         machine->device_memory->base = ROUND_UP(machine->ram_size,
2884                                                 SPAPR_DEVICE_MEM_ALIGN);
2885         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2886                            "device-memory", device_mem_size);
2887         memory_region_add_subregion(sysmem, machine->device_memory->base,
2888                                     &machine->device_memory->mr);
2889     }
2890 
2891     if (smc->dr_lmb_enabled) {
2892         spapr_create_lmb_dr_connectors(spapr);
2893     }
2894 
2895     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2896         /* Create the error string for live migration blocker */
2897         error_setg(&spapr->fwnmi_migration_blocker,
2898             "A machine check is being handled during migration. The handler"
2899             "may run and log hardware error on the destination");
2900     }
2901 
2902     if (mc->nvdimm_supported) {
2903         spapr_create_nvdimm_dr_connectors(spapr);
2904     }
2905 
2906     /* Set up RTAS event infrastructure */
2907     spapr_events_init(spapr);
2908 
2909     /* Set up the RTC RTAS interfaces */
2910     spapr_rtc_create(spapr);
2911 
2912     /* Set up VIO bus */
2913     spapr->vio_bus = spapr_vio_bus_init();
2914 
2915     for (i = 0; i < serial_max_hds(); i++) {
2916         if (serial_hd(i)) {
2917             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2918         }
2919     }
2920 
2921     /* We always have at least the nvram device on VIO */
2922     spapr_create_nvram(spapr);
2923 
2924     /*
2925      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2926      * connectors (described in root DT node's "ibm,drc-types" property)
2927      * are pre-initialized here. additional child connectors (such as
2928      * connectors for a PHBs PCI slots) are added as needed during their
2929      * parent's realization.
2930      */
2931     if (smc->dr_phb_enabled) {
2932         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2933             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2934         }
2935     }
2936 
2937     /* Set up PCI */
2938     spapr_pci_rtas_init();
2939 
2940     phb = spapr_create_default_phb();
2941 
2942     for (i = 0; i < nb_nics; i++) {
2943         NICInfo *nd = &nd_table[i];
2944 
2945         if (!nd->model) {
2946             nd->model = g_strdup("spapr-vlan");
2947         }
2948 
2949         if (g_str_equal(nd->model, "spapr-vlan") ||
2950             g_str_equal(nd->model, "ibmveth")) {
2951             spapr_vlan_create(spapr->vio_bus, nd);
2952         } else {
2953             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2954         }
2955     }
2956 
2957     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2958         spapr_vscsi_create(spapr->vio_bus);
2959     }
2960 
2961     /* Graphics */
2962     if (spapr_vga_init(phb->bus, &error_fatal)) {
2963         spapr->has_graphics = true;
2964         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2965     }
2966 
2967     if (machine->usb) {
2968         if (smc->use_ohci_by_default) {
2969             pci_create_simple(phb->bus, -1, "pci-ohci");
2970         } else {
2971             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2972         }
2973 
2974         if (spapr->has_graphics) {
2975             USBBus *usb_bus = usb_bus_find(-1);
2976 
2977             usb_create_simple(usb_bus, "usb-kbd");
2978             usb_create_simple(usb_bus, "usb-mouse");
2979         }
2980     }
2981 
2982     if (kernel_filename) {
2983         uint64_t lowaddr = 0;
2984 
2985         spapr->kernel_size = load_elf(kernel_filename, NULL,
2986                                       translate_kernel_address, spapr,
2987                                       NULL, &lowaddr, NULL, NULL, 1,
2988                                       PPC_ELF_MACHINE, 0, 0);
2989         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2990             spapr->kernel_size = load_elf(kernel_filename, NULL,
2991                                           translate_kernel_address, spapr, NULL,
2992                                           &lowaddr, NULL, NULL, 0,
2993                                           PPC_ELF_MACHINE,
2994                                           0, 0);
2995             spapr->kernel_le = spapr->kernel_size > 0;
2996         }
2997         if (spapr->kernel_size < 0) {
2998             error_report("error loading %s: %s", kernel_filename,
2999                          load_elf_strerror(spapr->kernel_size));
3000             exit(1);
3001         }
3002 
3003         /* load initrd */
3004         if (initrd_filename) {
3005             /* Try to locate the initrd in the gap between the kernel
3006              * and the firmware. Add a bit of space just in case
3007              */
3008             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3009                                   + 0x1ffff) & ~0xffff;
3010             spapr->initrd_size = load_image_targphys(initrd_filename,
3011                                                      spapr->initrd_base,
3012                                                      load_limit
3013                                                      - spapr->initrd_base);
3014             if (spapr->initrd_size < 0) {
3015                 error_report("could not load initial ram disk '%s'",
3016                              initrd_filename);
3017                 exit(1);
3018             }
3019         }
3020     }
3021 
3022     if (bios_name == NULL) {
3023         bios_name = FW_FILE_NAME;
3024     }
3025     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3026     if (!filename) {
3027         error_report("Could not find LPAR firmware '%s'", bios_name);
3028         exit(1);
3029     }
3030     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3031     if (fw_size <= 0) {
3032         error_report("Could not load LPAR firmware '%s'", filename);
3033         exit(1);
3034     }
3035     g_free(filename);
3036 
3037     /* FIXME: Should register things through the MachineState's qdev
3038      * interface, this is a legacy from the sPAPREnvironment structure
3039      * which predated MachineState but had a similar function */
3040     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3041     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3042                          &savevm_htab_handlers, spapr);
3043 
3044     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3045 
3046     qemu_register_boot_set(spapr_boot_set, spapr);
3047 
3048     /*
3049      * Nothing needs to be done to resume a suspended guest because
3050      * suspending does not change the machine state, so no need for
3051      * a ->wakeup method.
3052      */
3053     qemu_register_wakeup_support();
3054 
3055     if (kvm_enabled()) {
3056         /* to stop and start vmclock */
3057         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3058                                          &spapr->tb);
3059 
3060         kvmppc_spapr_enable_inkernel_multitce();
3061     }
3062 
3063     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3064 }
3065 
3066 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3067 {
3068     if (!vm_type) {
3069         return 0;
3070     }
3071 
3072     if (!strcmp(vm_type, "HV")) {
3073         return 1;
3074     }
3075 
3076     if (!strcmp(vm_type, "PR")) {
3077         return 2;
3078     }
3079 
3080     error_report("Unknown kvm-type specified '%s'", vm_type);
3081     exit(1);
3082 }
3083 
3084 /*
3085  * Implementation of an interface to adjust firmware path
3086  * for the bootindex property handling.
3087  */
3088 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3089                                    DeviceState *dev)
3090 {
3091 #define CAST(type, obj, name) \
3092     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3093     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3094     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3095     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3096 
3097     if (d) {
3098         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3099         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3100         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3101 
3102         if (spapr) {
3103             /*
3104              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3105              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3106              * 0x8000 | (target << 8) | (bus << 5) | lun
3107              * (see the "Logical unit addressing format" table in SAM5)
3108              */
3109             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3110             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3111                                    (uint64_t)id << 48);
3112         } else if (virtio) {
3113             /*
3114              * We use SRP luns of the form 01000000 | (target << 8) | lun
3115              * in the top 32 bits of the 64-bit LUN
3116              * Note: the quote above is from SLOF and it is wrong,
3117              * the actual binding is:
3118              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3119              */
3120             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3121             if (d->lun >= 256) {
3122                 /* Use the LUN "flat space addressing method" */
3123                 id |= 0x4000;
3124             }
3125             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3126                                    (uint64_t)id << 32);
3127         } else if (usb) {
3128             /*
3129              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3130              * in the top 32 bits of the 64-bit LUN
3131              */
3132             unsigned usb_port = atoi(usb->port->path);
3133             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3134             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3135                                    (uint64_t)id << 32);
3136         }
3137     }
3138 
3139     /*
3140      * SLOF probes the USB devices, and if it recognizes that the device is a
3141      * storage device, it changes its name to "storage" instead of "usb-host",
3142      * and additionally adds a child node for the SCSI LUN, so the correct
3143      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3144      */
3145     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3146         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3147         if (usb_host_dev_is_scsi_storage(usbdev)) {
3148             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3149         }
3150     }
3151 
3152     if (phb) {
3153         /* Replace "pci" with "pci@800000020000000" */
3154         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3155     }
3156 
3157     if (vsc) {
3158         /* Same logic as virtio above */
3159         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3160         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3161     }
3162 
3163     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3164         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3165         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3166         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3167     }
3168 
3169     return NULL;
3170 }
3171 
3172 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     return g_strdup(spapr->kvm_type);
3177 }
3178 
3179 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3180 {
3181     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3182 
3183     g_free(spapr->kvm_type);
3184     spapr->kvm_type = g_strdup(value);
3185 }
3186 
3187 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3188 {
3189     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3190 
3191     return spapr->use_hotplug_event_source;
3192 }
3193 
3194 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3195                                             Error **errp)
3196 {
3197     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3198 
3199     spapr->use_hotplug_event_source = value;
3200 }
3201 
3202 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3203 {
3204     return true;
3205 }
3206 
3207 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3208 {
3209     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3210 
3211     switch (spapr->resize_hpt) {
3212     case SPAPR_RESIZE_HPT_DEFAULT:
3213         return g_strdup("default");
3214     case SPAPR_RESIZE_HPT_DISABLED:
3215         return g_strdup("disabled");
3216     case SPAPR_RESIZE_HPT_ENABLED:
3217         return g_strdup("enabled");
3218     case SPAPR_RESIZE_HPT_REQUIRED:
3219         return g_strdup("required");
3220     }
3221     g_assert_not_reached();
3222 }
3223 
3224 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3225 {
3226     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3227 
3228     if (strcmp(value, "default") == 0) {
3229         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3230     } else if (strcmp(value, "disabled") == 0) {
3231         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3232     } else if (strcmp(value, "enabled") == 0) {
3233         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3234     } else if (strcmp(value, "required") == 0) {
3235         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3236     } else {
3237         error_setg(errp, "Bad value for \"resize-hpt\" property");
3238     }
3239 }
3240 
3241 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3242 {
3243     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3244 
3245     if (spapr->irq == &spapr_irq_xics_legacy) {
3246         return g_strdup("legacy");
3247     } else if (spapr->irq == &spapr_irq_xics) {
3248         return g_strdup("xics");
3249     } else if (spapr->irq == &spapr_irq_xive) {
3250         return g_strdup("xive");
3251     } else if (spapr->irq == &spapr_irq_dual) {
3252         return g_strdup("dual");
3253     }
3254     g_assert_not_reached();
3255 }
3256 
3257 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3258 {
3259     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3260 
3261     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3262         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3263         return;
3264     }
3265 
3266     /* The legacy IRQ backend can not be set */
3267     if (strcmp(value, "xics") == 0) {
3268         spapr->irq = &spapr_irq_xics;
3269     } else if (strcmp(value, "xive") == 0) {
3270         spapr->irq = &spapr_irq_xive;
3271     } else if (strcmp(value, "dual") == 0) {
3272         spapr->irq = &spapr_irq_dual;
3273     } else {
3274         error_setg(errp, "Bad value for \"ic-mode\" property");
3275     }
3276 }
3277 
3278 static char *spapr_get_host_model(Object *obj, Error **errp)
3279 {
3280     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3281 
3282     return g_strdup(spapr->host_model);
3283 }
3284 
3285 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3286 {
3287     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3288 
3289     g_free(spapr->host_model);
3290     spapr->host_model = g_strdup(value);
3291 }
3292 
3293 static char *spapr_get_host_serial(Object *obj, Error **errp)
3294 {
3295     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3296 
3297     return g_strdup(spapr->host_serial);
3298 }
3299 
3300 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3301 {
3302     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3303 
3304     g_free(spapr->host_serial);
3305     spapr->host_serial = g_strdup(value);
3306 }
3307 
3308 static void spapr_instance_init(Object *obj)
3309 {
3310     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3311     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3312 
3313     spapr->htab_fd = -1;
3314     spapr->use_hotplug_event_source = true;
3315     object_property_add_str(obj, "kvm-type",
3316                             spapr_get_kvm_type, spapr_set_kvm_type);
3317     object_property_set_description(obj, "kvm-type",
3318                                     "Specifies the KVM virtualization mode (HV, PR)");
3319     object_property_add_bool(obj, "modern-hotplug-events",
3320                             spapr_get_modern_hotplug_events,
3321                             spapr_set_modern_hotplug_events);
3322     object_property_set_description(obj, "modern-hotplug-events",
3323                                     "Use dedicated hotplug event mechanism in"
3324                                     " place of standard EPOW events when possible"
3325                                     " (required for memory hot-unplug support)");
3326     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3327                             "Maximum permitted CPU compatibility mode");
3328 
3329     object_property_add_str(obj, "resize-hpt",
3330                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3331     object_property_set_description(obj, "resize-hpt",
3332                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3333     object_property_add_uint32_ptr(obj, "vsmt",
3334                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3335     object_property_set_description(obj, "vsmt",
3336                                     "Virtual SMT: KVM behaves as if this were"
3337                                     " the host's SMT mode");
3338 
3339     object_property_add_bool(obj, "vfio-no-msix-emulation",
3340                              spapr_get_msix_emulation, NULL);
3341 
3342     object_property_add_uint64_ptr(obj, "kernel-addr",
3343                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3344     object_property_set_description(obj, "kernel-addr",
3345                                     stringify(KERNEL_LOAD_ADDR)
3346                                     " for -kernel is the default");
3347     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3348     /* The machine class defines the default interrupt controller mode */
3349     spapr->irq = smc->irq;
3350     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3351                             spapr_set_ic_mode);
3352     object_property_set_description(obj, "ic-mode",
3353                  "Specifies the interrupt controller mode (xics, xive, dual)");
3354 
3355     object_property_add_str(obj, "host-model",
3356         spapr_get_host_model, spapr_set_host_model);
3357     object_property_set_description(obj, "host-model",
3358         "Host model to advertise in guest device tree");
3359     object_property_add_str(obj, "host-serial",
3360         spapr_get_host_serial, spapr_set_host_serial);
3361     object_property_set_description(obj, "host-serial",
3362         "Host serial number to advertise in guest device tree");
3363 }
3364 
3365 static void spapr_machine_finalizefn(Object *obj)
3366 {
3367     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3368 
3369     g_free(spapr->kvm_type);
3370 }
3371 
3372 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3373 {
3374     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3375     PowerPCCPU *cpu = POWERPC_CPU(cs);
3376     CPUPPCState *env = &cpu->env;
3377 
3378     cpu_synchronize_state(cs);
3379     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3380     if (spapr->fwnmi_system_reset_addr != -1) {
3381         uint64_t rtas_addr, addr;
3382 
3383         /* get rtas addr from fdt */
3384         rtas_addr = spapr_get_rtas_addr();
3385         if (!rtas_addr) {
3386             qemu_system_guest_panicked(NULL);
3387             return;
3388         }
3389 
3390         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3391         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3392         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3393         env->gpr[3] = addr;
3394     }
3395     ppc_cpu_do_system_reset(cs);
3396     if (spapr->fwnmi_system_reset_addr != -1) {
3397         env->nip = spapr->fwnmi_system_reset_addr;
3398     }
3399 }
3400 
3401 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3402 {
3403     CPUState *cs;
3404 
3405     CPU_FOREACH(cs) {
3406         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3407     }
3408 }
3409 
3410 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3411                           void *fdt, int *fdt_start_offset, Error **errp)
3412 {
3413     uint64_t addr;
3414     uint32_t node;
3415 
3416     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3417     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3418                                     &error_abort);
3419     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3420                                              SPAPR_MEMORY_BLOCK_SIZE);
3421     return 0;
3422 }
3423 
3424 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3425                            bool dedicated_hp_event_source, Error **errp)
3426 {
3427     SpaprDrc *drc;
3428     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3429     int i;
3430     uint64_t addr = addr_start;
3431     bool hotplugged = spapr_drc_hotplugged(dev);
3432     Error *local_err = NULL;
3433 
3434     for (i = 0; i < nr_lmbs; i++) {
3435         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3436                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3437         g_assert(drc);
3438 
3439         spapr_drc_attach(drc, dev, &local_err);
3440         if (local_err) {
3441             while (addr > addr_start) {
3442                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3443                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3444                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3445                 spapr_drc_detach(drc);
3446             }
3447             error_propagate(errp, local_err);
3448             return;
3449         }
3450         if (!hotplugged) {
3451             spapr_drc_reset(drc);
3452         }
3453         addr += SPAPR_MEMORY_BLOCK_SIZE;
3454     }
3455     /* send hotplug notification to the
3456      * guest only in case of hotplugged memory
3457      */
3458     if (hotplugged) {
3459         if (dedicated_hp_event_source) {
3460             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3461                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3462             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3463                                                    nr_lmbs,
3464                                                    spapr_drc_index(drc));
3465         } else {
3466             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3467                                            nr_lmbs);
3468         }
3469     }
3470 }
3471 
3472 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3473                               Error **errp)
3474 {
3475     Error *local_err = NULL;
3476     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3477     PCDIMMDevice *dimm = PC_DIMM(dev);
3478     uint64_t size, addr, slot;
3479     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3480 
3481     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3482 
3483     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3484     if (local_err) {
3485         goto out;
3486     }
3487 
3488     if (!is_nvdimm) {
3489         addr = object_property_get_uint(OBJECT(dimm),
3490                                         PC_DIMM_ADDR_PROP, &local_err);
3491         if (local_err) {
3492             goto out_unplug;
3493         }
3494         spapr_add_lmbs(dev, addr, size,
3495                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3496                        &local_err);
3497     } else {
3498         slot = object_property_get_uint(OBJECT(dimm),
3499                                         PC_DIMM_SLOT_PROP, &local_err);
3500         if (local_err) {
3501             goto out_unplug;
3502         }
3503         spapr_add_nvdimm(dev, slot, &local_err);
3504     }
3505 
3506     if (local_err) {
3507         goto out_unplug;
3508     }
3509 
3510     return;
3511 
3512 out_unplug:
3513     pc_dimm_unplug(dimm, MACHINE(ms));
3514 out:
3515     error_propagate(errp, local_err);
3516 }
3517 
3518 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3519                                   Error **errp)
3520 {
3521     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3522     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3523     const MachineClass *mc = MACHINE_CLASS(smc);
3524     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3525     PCDIMMDevice *dimm = PC_DIMM(dev);
3526     Error *local_err = NULL;
3527     uint64_t size;
3528     Object *memdev;
3529     hwaddr pagesize;
3530 
3531     if (!smc->dr_lmb_enabled) {
3532         error_setg(errp, "Memory hotplug not supported for this machine");
3533         return;
3534     }
3535 
3536     if (is_nvdimm && !mc->nvdimm_supported) {
3537         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3538         return;
3539     }
3540 
3541     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3542     if (local_err) {
3543         error_propagate(errp, local_err);
3544         return;
3545     }
3546 
3547     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3548         error_setg(errp, "Hotplugged memory size must be a multiple of "
3549                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3550         return;
3551     } else if (is_nvdimm) {
3552         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3553         if (local_err) {
3554             error_propagate(errp, local_err);
3555             return;
3556         }
3557     }
3558 
3559     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3560                                       &error_abort);
3561     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3562     spapr_check_pagesize(spapr, pagesize, &local_err);
3563     if (local_err) {
3564         error_propagate(errp, local_err);
3565         return;
3566     }
3567 
3568     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3569 }
3570 
3571 struct SpaprDimmState {
3572     PCDIMMDevice *dimm;
3573     uint32_t nr_lmbs;
3574     QTAILQ_ENTRY(SpaprDimmState) next;
3575 };
3576 
3577 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3578                                                        PCDIMMDevice *dimm)
3579 {
3580     SpaprDimmState *dimm_state = NULL;
3581 
3582     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3583         if (dimm_state->dimm == dimm) {
3584             break;
3585         }
3586     }
3587     return dimm_state;
3588 }
3589 
3590 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3591                                                       uint32_t nr_lmbs,
3592                                                       PCDIMMDevice *dimm)
3593 {
3594     SpaprDimmState *ds = NULL;
3595 
3596     /*
3597      * If this request is for a DIMM whose removal had failed earlier
3598      * (due to guest's refusal to remove the LMBs), we would have this
3599      * dimm already in the pending_dimm_unplugs list. In that
3600      * case don't add again.
3601      */
3602     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3603     if (!ds) {
3604         ds = g_malloc0(sizeof(SpaprDimmState));
3605         ds->nr_lmbs = nr_lmbs;
3606         ds->dimm = dimm;
3607         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3608     }
3609     return ds;
3610 }
3611 
3612 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3613                                               SpaprDimmState *dimm_state)
3614 {
3615     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3616     g_free(dimm_state);
3617 }
3618 
3619 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3620                                                         PCDIMMDevice *dimm)
3621 {
3622     SpaprDrc *drc;
3623     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3624                                                   &error_abort);
3625     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3626     uint32_t avail_lmbs = 0;
3627     uint64_t addr_start, addr;
3628     int i;
3629 
3630     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3631                                          &error_abort);
3632 
3633     addr = addr_start;
3634     for (i = 0; i < nr_lmbs; i++) {
3635         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3636                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3637         g_assert(drc);
3638         if (drc->dev) {
3639             avail_lmbs++;
3640         }
3641         addr += SPAPR_MEMORY_BLOCK_SIZE;
3642     }
3643 
3644     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3645 }
3646 
3647 /* Callback to be called during DRC release. */
3648 void spapr_lmb_release(DeviceState *dev)
3649 {
3650     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3651     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3652     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3653 
3654     /* This information will get lost if a migration occurs
3655      * during the unplug process. In this case recover it. */
3656     if (ds == NULL) {
3657         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3658         g_assert(ds);
3659         /* The DRC being examined by the caller at least must be counted */
3660         g_assert(ds->nr_lmbs);
3661     }
3662 
3663     if (--ds->nr_lmbs) {
3664         return;
3665     }
3666 
3667     /*
3668      * Now that all the LMBs have been removed by the guest, call the
3669      * unplug handler chain. This can never fail.
3670      */
3671     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3672     object_unparent(OBJECT(dev));
3673 }
3674 
3675 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3676 {
3677     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3678     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3679 
3680     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3681     qdev_unrealize(dev);
3682     spapr_pending_dimm_unplugs_remove(spapr, ds);
3683 }
3684 
3685 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3686                                         DeviceState *dev, Error **errp)
3687 {
3688     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3689     Error *local_err = NULL;
3690     PCDIMMDevice *dimm = PC_DIMM(dev);
3691     uint32_t nr_lmbs;
3692     uint64_t size, addr_start, addr;
3693     int i;
3694     SpaprDrc *drc;
3695 
3696     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3697         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3698         return;
3699     }
3700 
3701     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3702     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3703 
3704     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3705                                          &local_err);
3706     if (local_err) {
3707         error_propagate(errp, local_err);
3708         return;
3709     }
3710 
3711     /*
3712      * An existing pending dimm state for this DIMM means that there is an
3713      * unplug operation in progress, waiting for the spapr_lmb_release
3714      * callback to complete the job (BQL can't cover that far). In this case,
3715      * bail out to avoid detaching DRCs that were already released.
3716      */
3717     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3718         error_setg(errp, "Memory unplug already in progress for device %s",
3719                    dev->id);
3720         return;
3721     }
3722 
3723     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3724 
3725     addr = addr_start;
3726     for (i = 0; i < nr_lmbs; i++) {
3727         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3728                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3729         g_assert(drc);
3730 
3731         spapr_drc_detach(drc);
3732         addr += SPAPR_MEMORY_BLOCK_SIZE;
3733     }
3734 
3735     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3736                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3737     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3738                                               nr_lmbs, spapr_drc_index(drc));
3739 }
3740 
3741 /* Callback to be called during DRC release. */
3742 void spapr_core_release(DeviceState *dev)
3743 {
3744     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3745 
3746     /* Call the unplug handler chain. This can never fail. */
3747     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3748     object_unparent(OBJECT(dev));
3749 }
3750 
3751 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3752 {
3753     MachineState *ms = MACHINE(hotplug_dev);
3754     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3755     CPUCore *cc = CPU_CORE(dev);
3756     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3757 
3758     if (smc->pre_2_10_has_unused_icps) {
3759         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3760         int i;
3761 
3762         for (i = 0; i < cc->nr_threads; i++) {
3763             CPUState *cs = CPU(sc->threads[i]);
3764 
3765             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3766         }
3767     }
3768 
3769     assert(core_slot);
3770     core_slot->cpu = NULL;
3771     qdev_unrealize(dev);
3772 }
3773 
3774 static
3775 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3776                                Error **errp)
3777 {
3778     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3779     int index;
3780     SpaprDrc *drc;
3781     CPUCore *cc = CPU_CORE(dev);
3782 
3783     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3784         error_setg(errp, "Unable to find CPU core with core-id: %d",
3785                    cc->core_id);
3786         return;
3787     }
3788     if (index == 0) {
3789         error_setg(errp, "Boot CPU core may not be unplugged");
3790         return;
3791     }
3792 
3793     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3794                           spapr_vcpu_id(spapr, cc->core_id));
3795     g_assert(drc);
3796 
3797     if (!spapr_drc_unplug_requested(drc)) {
3798         spapr_drc_detach(drc);
3799         spapr_hotplug_req_remove_by_index(drc);
3800     }
3801 }
3802 
3803 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3804                            void *fdt, int *fdt_start_offset, Error **errp)
3805 {
3806     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3807     CPUState *cs = CPU(core->threads[0]);
3808     PowerPCCPU *cpu = POWERPC_CPU(cs);
3809     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3810     int id = spapr_get_vcpu_id(cpu);
3811     char *nodename;
3812     int offset;
3813 
3814     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3815     offset = fdt_add_subnode(fdt, 0, nodename);
3816     g_free(nodename);
3817 
3818     spapr_dt_cpu(cs, fdt, offset, spapr);
3819 
3820     *fdt_start_offset = offset;
3821     return 0;
3822 }
3823 
3824 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3825                             Error **errp)
3826 {
3827     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3828     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3829     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3830     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3831     CPUCore *cc = CPU_CORE(dev);
3832     CPUState *cs;
3833     SpaprDrc *drc;
3834     Error *local_err = NULL;
3835     CPUArchId *core_slot;
3836     int index;
3837     bool hotplugged = spapr_drc_hotplugged(dev);
3838     int i;
3839 
3840     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3841     if (!core_slot) {
3842         error_setg(errp, "Unable to find CPU core with core-id: %d",
3843                    cc->core_id);
3844         return;
3845     }
3846     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3847                           spapr_vcpu_id(spapr, cc->core_id));
3848 
3849     g_assert(drc || !mc->has_hotpluggable_cpus);
3850 
3851     if (drc) {
3852         spapr_drc_attach(drc, dev, &local_err);
3853         if (local_err) {
3854             error_propagate(errp, local_err);
3855             return;
3856         }
3857 
3858         if (hotplugged) {
3859             /*
3860              * Send hotplug notification interrupt to the guest only
3861              * in case of hotplugged CPUs.
3862              */
3863             spapr_hotplug_req_add_by_index(drc);
3864         } else {
3865             spapr_drc_reset(drc);
3866         }
3867     }
3868 
3869     core_slot->cpu = OBJECT(dev);
3870 
3871     if (smc->pre_2_10_has_unused_icps) {
3872         for (i = 0; i < cc->nr_threads; i++) {
3873             cs = CPU(core->threads[i]);
3874             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3875         }
3876     }
3877 
3878     /*
3879      * Set compatibility mode to match the boot CPU, which was either set
3880      * by the machine reset code or by CAS.
3881      */
3882     if (hotplugged) {
3883         for (i = 0; i < cc->nr_threads; i++) {
3884             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3885                            &local_err);
3886             if (local_err) {
3887                 error_propagate(errp, local_err);
3888                 return;
3889             }
3890         }
3891     }
3892 }
3893 
3894 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3895                                 Error **errp)
3896 {
3897     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3898     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3899     CPUCore *cc = CPU_CORE(dev);
3900     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3901     const char *type = object_get_typename(OBJECT(dev));
3902     CPUArchId *core_slot;
3903     int index;
3904     unsigned int smp_threads = machine->smp.threads;
3905 
3906     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3907         error_setg(errp, "CPU hotplug not supported for this machine");
3908         return;
3909     }
3910 
3911     if (strcmp(base_core_type, type)) {
3912         error_setg(errp, "CPU core type should be %s", base_core_type);
3913         return;
3914     }
3915 
3916     if (cc->core_id % smp_threads) {
3917         error_setg(errp, "invalid core id %d", cc->core_id);
3918         return;
3919     }
3920 
3921     /*
3922      * In general we should have homogeneous threads-per-core, but old
3923      * (pre hotplug support) machine types allow the last core to have
3924      * reduced threads as a compatibility hack for when we allowed
3925      * total vcpus not a multiple of threads-per-core.
3926      */
3927     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3928         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3929                    smp_threads);
3930         return;
3931     }
3932 
3933     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3934     if (!core_slot) {
3935         error_setg(errp, "core id %d out of range", cc->core_id);
3936         return;
3937     }
3938 
3939     if (core_slot->cpu) {
3940         error_setg(errp, "core %d already populated", cc->core_id);
3941         return;
3942     }
3943 
3944     numa_cpu_pre_plug(core_slot, dev, errp);
3945 }
3946 
3947 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3948                           void *fdt, int *fdt_start_offset, Error **errp)
3949 {
3950     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3951     int intc_phandle;
3952 
3953     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3954     if (intc_phandle <= 0) {
3955         return -1;
3956     }
3957 
3958     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3959         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3960         return -1;
3961     }
3962 
3963     /* generally SLOF creates these, for hotplug it's up to QEMU */
3964     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3965 
3966     return 0;
3967 }
3968 
3969 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3970                                Error **errp)
3971 {
3972     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3973     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3974     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3975     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3976 
3977     if (dev->hotplugged && !smc->dr_phb_enabled) {
3978         error_setg(errp, "PHB hotplug not supported for this machine");
3979         return;
3980     }
3981 
3982     if (sphb->index == (uint32_t)-1) {
3983         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3984         return;
3985     }
3986 
3987     /*
3988      * This will check that sphb->index doesn't exceed the maximum number of
3989      * PHBs for the current machine type.
3990      */
3991     smc->phb_placement(spapr, sphb->index,
3992                        &sphb->buid, &sphb->io_win_addr,
3993                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3994                        windows_supported, sphb->dma_liobn,
3995                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3996                        errp);
3997 }
3998 
3999 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4000                            Error **errp)
4001 {
4002     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4003     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4004     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4005     SpaprDrc *drc;
4006     bool hotplugged = spapr_drc_hotplugged(dev);
4007     Error *local_err = NULL;
4008 
4009     if (!smc->dr_phb_enabled) {
4010         return;
4011     }
4012 
4013     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4014     /* hotplug hooks should check it's enabled before getting this far */
4015     assert(drc);
4016 
4017     spapr_drc_attach(drc, dev, &local_err);
4018     if (local_err) {
4019         error_propagate(errp, local_err);
4020         return;
4021     }
4022 
4023     if (hotplugged) {
4024         spapr_hotplug_req_add_by_index(drc);
4025     } else {
4026         spapr_drc_reset(drc);
4027     }
4028 }
4029 
4030 void spapr_phb_release(DeviceState *dev)
4031 {
4032     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4033 
4034     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4035     object_unparent(OBJECT(dev));
4036 }
4037 
4038 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4039 {
4040     qdev_unrealize(dev);
4041 }
4042 
4043 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4044                                      DeviceState *dev, Error **errp)
4045 {
4046     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4047     SpaprDrc *drc;
4048 
4049     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4050     assert(drc);
4051 
4052     if (!spapr_drc_unplug_requested(drc)) {
4053         spapr_drc_detach(drc);
4054         spapr_hotplug_req_remove_by_index(drc);
4055     }
4056 }
4057 
4058 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4059                                  Error **errp)
4060 {
4061     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4062     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4063 
4064     if (spapr->tpm_proxy != NULL) {
4065         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4066         return;
4067     }
4068 
4069     spapr->tpm_proxy = tpm_proxy;
4070 }
4071 
4072 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4073 {
4074     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4075 
4076     qdev_unrealize(dev);
4077     object_unparent(OBJECT(dev));
4078     spapr->tpm_proxy = NULL;
4079 }
4080 
4081 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4082                                       DeviceState *dev, Error **errp)
4083 {
4084     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4085         spapr_memory_plug(hotplug_dev, dev, errp);
4086     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4087         spapr_core_plug(hotplug_dev, dev, errp);
4088     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4089         spapr_phb_plug(hotplug_dev, dev, errp);
4090     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4091         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4092     }
4093 }
4094 
4095 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4096                                         DeviceState *dev, Error **errp)
4097 {
4098     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4099         spapr_memory_unplug(hotplug_dev, dev);
4100     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4101         spapr_core_unplug(hotplug_dev, dev);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4103         spapr_phb_unplug(hotplug_dev, dev);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4105         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4106     }
4107 }
4108 
4109 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4110                                                 DeviceState *dev, Error **errp)
4111 {
4112     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4113     MachineClass *mc = MACHINE_GET_CLASS(sms);
4114     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4115 
4116     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4117         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4118             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4119         } else {
4120             /* NOTE: this means there is a window after guest reset, prior to
4121              * CAS negotiation, where unplug requests will fail due to the
4122              * capability not being detected yet. This is a bit different than
4123              * the case with PCI unplug, where the events will be queued and
4124              * eventually handled by the guest after boot
4125              */
4126             error_setg(errp, "Memory hot unplug not supported for this guest");
4127         }
4128     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4129         if (!mc->has_hotpluggable_cpus) {
4130             error_setg(errp, "CPU hot unplug not supported on this machine");
4131             return;
4132         }
4133         spapr_core_unplug_request(hotplug_dev, dev, errp);
4134     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4135         if (!smc->dr_phb_enabled) {
4136             error_setg(errp, "PHB hot unplug not supported on this machine");
4137             return;
4138         }
4139         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4140     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4141         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4142     }
4143 }
4144 
4145 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4146                                           DeviceState *dev, Error **errp)
4147 {
4148     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4149         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4150     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4151         spapr_core_pre_plug(hotplug_dev, dev, errp);
4152     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4153         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4154     }
4155 }
4156 
4157 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4158                                                  DeviceState *dev)
4159 {
4160     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4161         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4162         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4163         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4164         return HOTPLUG_HANDLER(machine);
4165     }
4166     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4167         PCIDevice *pcidev = PCI_DEVICE(dev);
4168         PCIBus *root = pci_device_root_bus(pcidev);
4169         SpaprPhbState *phb =
4170             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4171                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4172 
4173         if (phb) {
4174             return HOTPLUG_HANDLER(phb);
4175         }
4176     }
4177     return NULL;
4178 }
4179 
4180 static CpuInstanceProperties
4181 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4182 {
4183     CPUArchId *core_slot;
4184     MachineClass *mc = MACHINE_GET_CLASS(machine);
4185 
4186     /* make sure possible_cpu are intialized */
4187     mc->possible_cpu_arch_ids(machine);
4188     /* get CPU core slot containing thread that matches cpu_index */
4189     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4190     assert(core_slot);
4191     return core_slot->props;
4192 }
4193 
4194 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4195 {
4196     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4197 }
4198 
4199 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4200 {
4201     int i;
4202     unsigned int smp_threads = machine->smp.threads;
4203     unsigned int smp_cpus = machine->smp.cpus;
4204     const char *core_type;
4205     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4206     MachineClass *mc = MACHINE_GET_CLASS(machine);
4207 
4208     if (!mc->has_hotpluggable_cpus) {
4209         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4210     }
4211     if (machine->possible_cpus) {
4212         assert(machine->possible_cpus->len == spapr_max_cores);
4213         return machine->possible_cpus;
4214     }
4215 
4216     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4217     if (!core_type) {
4218         error_report("Unable to find sPAPR CPU Core definition");
4219         exit(1);
4220     }
4221 
4222     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4223                              sizeof(CPUArchId) * spapr_max_cores);
4224     machine->possible_cpus->len = spapr_max_cores;
4225     for (i = 0; i < machine->possible_cpus->len; i++) {
4226         int core_id = i * smp_threads;
4227 
4228         machine->possible_cpus->cpus[i].type = core_type;
4229         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4230         machine->possible_cpus->cpus[i].arch_id = core_id;
4231         machine->possible_cpus->cpus[i].props.has_core_id = true;
4232         machine->possible_cpus->cpus[i].props.core_id = core_id;
4233     }
4234     return machine->possible_cpus;
4235 }
4236 
4237 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4238                                 uint64_t *buid, hwaddr *pio,
4239                                 hwaddr *mmio32, hwaddr *mmio64,
4240                                 unsigned n_dma, uint32_t *liobns,
4241                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4242 {
4243     /*
4244      * New-style PHB window placement.
4245      *
4246      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4247      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4248      * windows.
4249      *
4250      * Some guest kernels can't work with MMIO windows above 1<<46
4251      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4252      *
4253      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4254      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4255      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4256      * 1TiB 64-bit MMIO windows for each PHB.
4257      */
4258     const uint64_t base_buid = 0x800000020000000ULL;
4259     int i;
4260 
4261     /* Sanity check natural alignments */
4262     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4263     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4264     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4265     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4266     /* Sanity check bounds */
4267     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4268                       SPAPR_PCI_MEM32_WIN_SIZE);
4269     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4270                       SPAPR_PCI_MEM64_WIN_SIZE);
4271 
4272     if (index >= SPAPR_MAX_PHBS) {
4273         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4274                    SPAPR_MAX_PHBS - 1);
4275         return;
4276     }
4277 
4278     *buid = base_buid + index;
4279     for (i = 0; i < n_dma; ++i) {
4280         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4281     }
4282 
4283     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4284     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4285     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4286 
4287     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4288     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4289 }
4290 
4291 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4292 {
4293     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4294 
4295     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4296 }
4297 
4298 static void spapr_ics_resend(XICSFabric *dev)
4299 {
4300     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4301 
4302     ics_resend(spapr->ics);
4303 }
4304 
4305 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4306 {
4307     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4308 
4309     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4310 }
4311 
4312 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4313                                  Monitor *mon)
4314 {
4315     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4316 
4317     spapr_irq_print_info(spapr, mon);
4318     monitor_printf(mon, "irqchip: %s\n",
4319                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4320 }
4321 
4322 /*
4323  * This is a XIVE only operation
4324  */
4325 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4326                            uint8_t nvt_blk, uint32_t nvt_idx,
4327                            bool cam_ignore, uint8_t priority,
4328                            uint32_t logic_serv, XiveTCTXMatch *match)
4329 {
4330     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4331     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4332     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4333     int count;
4334 
4335     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4336                            priority, logic_serv, match);
4337     if (count < 0) {
4338         return count;
4339     }
4340 
4341     /*
4342      * When we implement the save and restore of the thread interrupt
4343      * contexts in the enter/exit CPU handlers of the machine and the
4344      * escalations in QEMU, we should be able to handle non dispatched
4345      * vCPUs.
4346      *
4347      * Until this is done, the sPAPR machine should find at least one
4348      * matching context always.
4349      */
4350     if (count == 0) {
4351         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4352                       nvt_blk, nvt_idx);
4353     }
4354 
4355     return count;
4356 }
4357 
4358 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4359 {
4360     return cpu->vcpu_id;
4361 }
4362 
4363 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4364 {
4365     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4366     MachineState *ms = MACHINE(spapr);
4367     int vcpu_id;
4368 
4369     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4370 
4371     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4372         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4373         error_append_hint(errp, "Adjust the number of cpus to %d "
4374                           "or try to raise the number of threads per core\n",
4375                           vcpu_id * ms->smp.threads / spapr->vsmt);
4376         return;
4377     }
4378 
4379     cpu->vcpu_id = vcpu_id;
4380 }
4381 
4382 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4383 {
4384     CPUState *cs;
4385 
4386     CPU_FOREACH(cs) {
4387         PowerPCCPU *cpu = POWERPC_CPU(cs);
4388 
4389         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4390             return cpu;
4391         }
4392     }
4393 
4394     return NULL;
4395 }
4396 
4397 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4398 {
4399     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4400 
4401     /* These are only called by TCG, KVM maintains dispatch state */
4402 
4403     spapr_cpu->prod = false;
4404     if (spapr_cpu->vpa_addr) {
4405         CPUState *cs = CPU(cpu);
4406         uint32_t dispatch;
4407 
4408         dispatch = ldl_be_phys(cs->as,
4409                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4410         dispatch++;
4411         if ((dispatch & 1) != 0) {
4412             qemu_log_mask(LOG_GUEST_ERROR,
4413                           "VPA: incorrect dispatch counter value for "
4414                           "dispatched partition %u, correcting.\n", dispatch);
4415             dispatch++;
4416         }
4417         stl_be_phys(cs->as,
4418                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4419     }
4420 }
4421 
4422 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4423 {
4424     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4425 
4426     if (spapr_cpu->vpa_addr) {
4427         CPUState *cs = CPU(cpu);
4428         uint32_t dispatch;
4429 
4430         dispatch = ldl_be_phys(cs->as,
4431                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4432         dispatch++;
4433         if ((dispatch & 1) != 1) {
4434             qemu_log_mask(LOG_GUEST_ERROR,
4435                           "VPA: incorrect dispatch counter value for "
4436                           "preempted partition %u, correcting.\n", dispatch);
4437             dispatch++;
4438         }
4439         stl_be_phys(cs->as,
4440                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4441     }
4442 }
4443 
4444 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4445 {
4446     MachineClass *mc = MACHINE_CLASS(oc);
4447     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4448     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4449     NMIClass *nc = NMI_CLASS(oc);
4450     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4451     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4452     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4453     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4454     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4455 
4456     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4457     mc->ignore_boot_device_suffixes = true;
4458 
4459     /*
4460      * We set up the default / latest behaviour here.  The class_init
4461      * functions for the specific versioned machine types can override
4462      * these details for backwards compatibility
4463      */
4464     mc->init = spapr_machine_init;
4465     mc->reset = spapr_machine_reset;
4466     mc->block_default_type = IF_SCSI;
4467     mc->max_cpus = 1024;
4468     mc->no_parallel = 1;
4469     mc->default_boot_order = "";
4470     mc->default_ram_size = 512 * MiB;
4471     mc->default_ram_id = "ppc_spapr.ram";
4472     mc->default_display = "std";
4473     mc->kvm_type = spapr_kvm_type;
4474     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4475     mc->pci_allow_0_address = true;
4476     assert(!mc->get_hotplug_handler);
4477     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4478     hc->pre_plug = spapr_machine_device_pre_plug;
4479     hc->plug = spapr_machine_device_plug;
4480     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4481     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4482     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4483     hc->unplug_request = spapr_machine_device_unplug_request;
4484     hc->unplug = spapr_machine_device_unplug;
4485 
4486     smc->dr_lmb_enabled = true;
4487     smc->update_dt_enabled = true;
4488     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4489     mc->has_hotpluggable_cpus = true;
4490     mc->nvdimm_supported = true;
4491     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4492     fwc->get_dev_path = spapr_get_fw_dev_path;
4493     nc->nmi_monitor_handler = spapr_nmi;
4494     smc->phb_placement = spapr_phb_placement;
4495     vhc->hypercall = emulate_spapr_hypercall;
4496     vhc->hpt_mask = spapr_hpt_mask;
4497     vhc->map_hptes = spapr_map_hptes;
4498     vhc->unmap_hptes = spapr_unmap_hptes;
4499     vhc->hpte_set_c = spapr_hpte_set_c;
4500     vhc->hpte_set_r = spapr_hpte_set_r;
4501     vhc->get_pate = spapr_get_pate;
4502     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4503     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4504     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4505     xic->ics_get = spapr_ics_get;
4506     xic->ics_resend = spapr_ics_resend;
4507     xic->icp_get = spapr_icp_get;
4508     ispc->print_info = spapr_pic_print_info;
4509     /* Force NUMA node memory size to be a multiple of
4510      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4511      * in which LMBs are represented and hot-added
4512      */
4513     mc->numa_mem_align_shift = 28;
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] = SPAPR_CAP_ON;
4527     spapr_caps_add_properties(smc);
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-5.2
4586  */
4587 static void spapr_machine_5_2_class_options(MachineClass *mc)
4588 {
4589     /* Defaults for the latest behaviour inherited from the base class */
4590 }
4591 
4592 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4593 
4594 /*
4595  * pseries-5.1
4596  */
4597 static void spapr_machine_5_1_class_options(MachineClass *mc)
4598 {
4599     spapr_machine_5_2_class_options(mc);
4600     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4601 }
4602 
4603 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4604 
4605 /*
4606  * pseries-5.0
4607  */
4608 static void spapr_machine_5_0_class_options(MachineClass *mc)
4609 {
4610     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4611     static GlobalProperty compat[] = {
4612         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4613     };
4614 
4615     spapr_machine_5_1_class_options(mc);
4616     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4617     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4618     mc->numa_mem_supported = true;
4619     smc->pre_5_1_assoc_refpoints = true;
4620 }
4621 
4622 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4623 
4624 /*
4625  * pseries-4.2
4626  */
4627 static void spapr_machine_4_2_class_options(MachineClass *mc)
4628 {
4629     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4630 
4631     spapr_machine_5_0_class_options(mc);
4632     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4633     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4634     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4635     smc->rma_limit = 16 * GiB;
4636     mc->nvdimm_supported = false;
4637 }
4638 
4639 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4640 
4641 /*
4642  * pseries-4.1
4643  */
4644 static void spapr_machine_4_1_class_options(MachineClass *mc)
4645 {
4646     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4647     static GlobalProperty compat[] = {
4648         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4649         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4650     };
4651 
4652     spapr_machine_4_2_class_options(mc);
4653     smc->linux_pci_probe = false;
4654     smc->smp_threads_vsmt = false;
4655     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4656     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4657 }
4658 
4659 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4660 
4661 /*
4662  * pseries-4.0
4663  */
4664 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4665                               uint64_t *buid, hwaddr *pio,
4666                               hwaddr *mmio32, hwaddr *mmio64,
4667                               unsigned n_dma, uint32_t *liobns,
4668                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4669 {
4670     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4671                         nv2gpa, nv2atsd, errp);
4672     *nv2gpa = 0;
4673     *nv2atsd = 0;
4674 }
4675 
4676 static void spapr_machine_4_0_class_options(MachineClass *mc)
4677 {
4678     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4679 
4680     spapr_machine_4_1_class_options(mc);
4681     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4682     smc->phb_placement = phb_placement_4_0;
4683     smc->irq = &spapr_irq_xics;
4684     smc->pre_4_1_migration = true;
4685 }
4686 
4687 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4688 
4689 /*
4690  * pseries-3.1
4691  */
4692 static void spapr_machine_3_1_class_options(MachineClass *mc)
4693 {
4694     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4695 
4696     spapr_machine_4_0_class_options(mc);
4697     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4698 
4699     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4700     smc->update_dt_enabled = false;
4701     smc->dr_phb_enabled = false;
4702     smc->broken_host_serial_model = true;
4703     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4704     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4705     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4706     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4707 }
4708 
4709 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4710 
4711 /*
4712  * pseries-3.0
4713  */
4714 
4715 static void spapr_machine_3_0_class_options(MachineClass *mc)
4716 {
4717     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4718 
4719     spapr_machine_3_1_class_options(mc);
4720     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4721 
4722     smc->legacy_irq_allocation = true;
4723     smc->nr_xirqs = 0x400;
4724     smc->irq = &spapr_irq_xics_legacy;
4725 }
4726 
4727 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4728 
4729 /*
4730  * pseries-2.12
4731  */
4732 static void spapr_machine_2_12_class_options(MachineClass *mc)
4733 {
4734     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4735     static GlobalProperty compat[] = {
4736         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4737         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4738     };
4739 
4740     spapr_machine_3_0_class_options(mc);
4741     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4742     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4743 
4744     /* We depend on kvm_enabled() to choose a default value for the
4745      * hpt-max-page-size capability. Of course we can't do it here
4746      * because this is too early and the HW accelerator isn't initialzed
4747      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4748      */
4749     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4753 
4754 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4755 {
4756     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4757 
4758     spapr_machine_2_12_class_options(mc);
4759     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4760     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4761     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4765 
4766 /*
4767  * pseries-2.11
4768  */
4769 
4770 static void spapr_machine_2_11_class_options(MachineClass *mc)
4771 {
4772     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4773 
4774     spapr_machine_2_12_class_options(mc);
4775     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4776     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4777 }
4778 
4779 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4780 
4781 /*
4782  * pseries-2.10
4783  */
4784 
4785 static void spapr_machine_2_10_class_options(MachineClass *mc)
4786 {
4787     spapr_machine_2_11_class_options(mc);
4788     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4789 }
4790 
4791 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4792 
4793 /*
4794  * pseries-2.9
4795  */
4796 
4797 static void spapr_machine_2_9_class_options(MachineClass *mc)
4798 {
4799     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4800     static GlobalProperty compat[] = {
4801         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4802     };
4803 
4804     spapr_machine_2_10_class_options(mc);
4805     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4806     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4807     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4808     smc->pre_2_10_has_unused_icps = true;
4809     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4810 }
4811 
4812 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4813 
4814 /*
4815  * pseries-2.8
4816  */
4817 
4818 static void spapr_machine_2_8_class_options(MachineClass *mc)
4819 {
4820     static GlobalProperty compat[] = {
4821         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4822     };
4823 
4824     spapr_machine_2_9_class_options(mc);
4825     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4826     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4827     mc->numa_mem_align_shift = 23;
4828 }
4829 
4830 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4831 
4832 /*
4833  * pseries-2.7
4834  */
4835 
4836 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4837                               uint64_t *buid, hwaddr *pio,
4838                               hwaddr *mmio32, hwaddr *mmio64,
4839                               unsigned n_dma, uint32_t *liobns,
4840                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4841 {
4842     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4843     const uint64_t base_buid = 0x800000020000000ULL;
4844     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4845     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4846     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4847     const uint32_t max_index = 255;
4848     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4849 
4850     uint64_t ram_top = MACHINE(spapr)->ram_size;
4851     hwaddr phb0_base, phb_base;
4852     int i;
4853 
4854     /* Do we have device memory? */
4855     if (MACHINE(spapr)->maxram_size > ram_top) {
4856         /* Can't just use maxram_size, because there may be an
4857          * alignment gap between normal and device memory regions
4858          */
4859         ram_top = MACHINE(spapr)->device_memory->base +
4860             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4861     }
4862 
4863     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4864 
4865     if (index > max_index) {
4866         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4867                    max_index);
4868         return;
4869     }
4870 
4871     *buid = base_buid + index;
4872     for (i = 0; i < n_dma; ++i) {
4873         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4874     }
4875 
4876     phb_base = phb0_base + index * phb_spacing;
4877     *pio = phb_base + pio_offset;
4878     *mmio32 = phb_base + mmio_offset;
4879     /*
4880      * We don't set the 64-bit MMIO window, relying on the PHB's
4881      * fallback behaviour of automatically splitting a large "32-bit"
4882      * window into contiguous 32-bit and 64-bit windows
4883      */
4884 
4885     *nv2gpa = 0;
4886     *nv2atsd = 0;
4887 }
4888 
4889 static void spapr_machine_2_7_class_options(MachineClass *mc)
4890 {
4891     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4892     static GlobalProperty compat[] = {
4893         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4894         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4895         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4896         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4897     };
4898 
4899     spapr_machine_2_8_class_options(mc);
4900     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4901     mc->default_machine_opts = "modern-hotplug-events=off";
4902     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4903     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4904     smc->phb_placement = phb_placement_2_7;
4905 }
4906 
4907 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4908 
4909 /*
4910  * pseries-2.6
4911  */
4912 
4913 static void spapr_machine_2_6_class_options(MachineClass *mc)
4914 {
4915     static GlobalProperty compat[] = {
4916         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4917     };
4918 
4919     spapr_machine_2_7_class_options(mc);
4920     mc->has_hotpluggable_cpus = false;
4921     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4922     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4923 }
4924 
4925 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4926 
4927 /*
4928  * pseries-2.5
4929  */
4930 
4931 static void spapr_machine_2_5_class_options(MachineClass *mc)
4932 {
4933     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4934     static GlobalProperty compat[] = {
4935         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4936     };
4937 
4938     spapr_machine_2_6_class_options(mc);
4939     smc->use_ohci_by_default = true;
4940     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4941     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4942 }
4943 
4944 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4945 
4946 /*
4947  * pseries-2.4
4948  */
4949 
4950 static void spapr_machine_2_4_class_options(MachineClass *mc)
4951 {
4952     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4953 
4954     spapr_machine_2_5_class_options(mc);
4955     smc->dr_lmb_enabled = false;
4956     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4957 }
4958 
4959 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4960 
4961 /*
4962  * pseries-2.3
4963  */
4964 
4965 static void spapr_machine_2_3_class_options(MachineClass *mc)
4966 {
4967     static GlobalProperty compat[] = {
4968         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4969     };
4970     spapr_machine_2_4_class_options(mc);
4971     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4972     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4973 }
4974 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4975 
4976 /*
4977  * pseries-2.2
4978  */
4979 
4980 static void spapr_machine_2_2_class_options(MachineClass *mc)
4981 {
4982     static GlobalProperty compat[] = {
4983         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4984     };
4985 
4986     spapr_machine_2_3_class_options(mc);
4987     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4988     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4989     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4990 }
4991 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4992 
4993 /*
4994  * pseries-2.1
4995  */
4996 
4997 static void spapr_machine_2_1_class_options(MachineClass *mc)
4998 {
4999     spapr_machine_2_2_class_options(mc);
5000     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5001 }
5002 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5003 
5004 static void spapr_machine_register_types(void)
5005 {
5006     type_register_static(&spapr_machine_info);
5007 }
5008 
5009 type_init(spapr_machine_register_types)
5010