xref: /openbmc/qemu/hw/ppc/spapr.c (revision ac12b601)
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 "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/visitor.h"
33 #include "sysemu/sysemu.h"
34 #include "sysemu/hostmem.h"
35 #include "sysemu/numa.h"
36 #include "sysemu/qtest.h"
37 #include "sysemu/reset.h"
38 #include "sysemu/runstate.h"
39 #include "qemu/log.h"
40 #include "hw/fw-path-provider.h"
41 #include "elf.h"
42 #include "net/net.h"
43 #include "sysemu/device_tree.h"
44 #include "sysemu/cpus.h"
45 #include "sysemu/hw_accel.h"
46 #include "kvm_ppc.h"
47 #include "migration/misc.h"
48 #include "migration/qemu-file-types.h"
49 #include "migration/global_state.h"
50 #include "migration/register.h"
51 #include "migration/blocker.h"
52 #include "mmu-hash64.h"
53 #include "mmu-book3s-v3.h"
54 #include "cpu-models.h"
55 #include "hw/core/cpu.h"
56 
57 #include "hw/ppc/ppc.h"
58 #include "hw/loader.h"
59 
60 #include "hw/ppc/fdt.h"
61 #include "hw/ppc/spapr.h"
62 #include "hw/ppc/spapr_vio.h"
63 #include "hw/qdev-properties.h"
64 #include "hw/pci-host/spapr.h"
65 #include "hw/pci/msi.h"
66 
67 #include "hw/pci/pci.h"
68 #include "hw/scsi/scsi.h"
69 #include "hw/virtio/virtio-scsi.h"
70 #include "hw/virtio/vhost-scsi-common.h"
71 
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 #include "hw/ppc/spapr_numa.h"
85 #include "hw/ppc/pef.h"
86 
87 #include "monitor/monitor.h"
88 
89 #include <libfdt.h>
90 
91 /* SLOF memory layout:
92  *
93  * SLOF raw image loaded at 0, copies its romfs right below the flat
94  * device-tree, then position SLOF itself 31M below that
95  *
96  * So we set FW_OVERHEAD to 40MB which should account for all of that
97  * and more
98  *
99  * We load our kernel at 4M, leaving space for SLOF initial image
100  */
101 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
102 #define FW_MAX_SIZE             0x400000
103 #define FW_FILE_NAME            "slof.bin"
104 #define FW_OVERHEAD             0x2800000
105 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
106 
107 #define MIN_RMA_SLOF            (128 * MiB)
108 
109 #define PHANDLE_INTC            0x00001111
110 
111 /* These two functions implement the VCPU id numbering: one to compute them
112  * all and one to identify thread 0 of a VCORE. Any change to the first one
113  * is likely to have an impact on the second one, so let's keep them close.
114  */
115 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
116 {
117     MachineState *ms = MACHINE(spapr);
118     unsigned int smp_threads = ms->smp.threads;
119 
120     assert(spapr->vsmt);
121     return
122         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
123 }
124 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
125                                       PowerPCCPU *cpu)
126 {
127     assert(spapr->vsmt);
128     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
129 }
130 
131 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
132 {
133     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
134      * and newer QEMUs don't even have them. In both cases, we don't want
135      * to send anything on the wire.
136      */
137     return false;
138 }
139 
140 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
141     .name = "icp/server",
142     .version_id = 1,
143     .minimum_version_id = 1,
144     .needed = pre_2_10_vmstate_dummy_icp_needed,
145     .fields = (VMStateField[]) {
146         VMSTATE_UNUSED(4), /* uint32_t xirr */
147         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
148         VMSTATE_UNUSED(1), /* uint8_t mfrr */
149         VMSTATE_END_OF_LIST()
150     },
151 };
152 
153 static void pre_2_10_vmstate_register_dummy_icp(int i)
154 {
155     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
156                      (void *)(uintptr_t) i);
157 }
158 
159 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
160 {
161     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
162                        (void *)(uintptr_t) i);
163 }
164 
165 int spapr_max_server_number(SpaprMachineState *spapr)
166 {
167     MachineState *ms = MACHINE(spapr);
168 
169     assert(spapr->vsmt);
170     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
171 }
172 
173 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
174                                   int smt_threads)
175 {
176     int i, ret = 0;
177     uint32_t servers_prop[smt_threads];
178     uint32_t gservers_prop[smt_threads * 2];
179     int index = spapr_get_vcpu_id(cpu);
180 
181     if (cpu->compat_pvr) {
182         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
183         if (ret < 0) {
184             return ret;
185         }
186     }
187 
188     /* Build interrupt servers and gservers properties */
189     for (i = 0; i < smt_threads; i++) {
190         servers_prop[i] = cpu_to_be32(index + i);
191         /* Hack, direct the group queues back to cpu 0 */
192         gservers_prop[i*2] = cpu_to_be32(index + i);
193         gservers_prop[i*2 + 1] = 0;
194     }
195     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
196                       servers_prop, sizeof(servers_prop));
197     if (ret < 0) {
198         return ret;
199     }
200     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
201                       gservers_prop, sizeof(gservers_prop));
202 
203     return ret;
204 }
205 
206 static void spapr_dt_pa_features(SpaprMachineState *spapr,
207                                  PowerPCCPU *cpu,
208                                  void *fdt, int offset)
209 {
210     uint8_t pa_features_206[] = { 6, 0,
211         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
212     uint8_t pa_features_207[] = { 24, 0,
213         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
214         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
215         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
216         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
217     uint8_t pa_features_300[] = { 66, 0,
218         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
219         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
220         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
221         /* 6: DS207 */
222         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
223         /* 16: Vector */
224         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
225         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
226         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
227         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
228         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
229         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
230         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
231         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
233         /* 42: PM, 44: PC RA, 46: SC vec'd */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
235         /* 48: SIMD, 50: QP BFP, 52: String */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
237         /* 54: DecFP, 56: DecI, 58: SHA */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
239         /* 60: NM atomic, 62: RNG */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
241     };
242     uint8_t *pa_features = NULL;
243     size_t pa_size;
244 
245     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
246         pa_features = pa_features_206;
247         pa_size = sizeof(pa_features_206);
248     }
249     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
250         pa_features = pa_features_207;
251         pa_size = sizeof(pa_features_207);
252     }
253     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
254         pa_features = pa_features_300;
255         pa_size = sizeof(pa_features_300);
256     }
257     if (!pa_features) {
258         return;
259     }
260 
261     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
262         /*
263          * Note: we keep CI large pages off by default because a 64K capable
264          * guest provisioned with large pages might otherwise try to map a qemu
265          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
266          * even if that qemu runs on a 4k host.
267          * We dd this bit back here if we are confident this is not an issue
268          */
269         pa_features[3] |= 0x20;
270     }
271     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
272         pa_features[24] |= 0x80;    /* Transactional memory support */
273     }
274     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
275         /* Workaround for broken kernels that attempt (guest) radix
276          * mode when they can't handle it, if they see the radix bit set
277          * in pa-features. So hide it from them. */
278         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
279     }
280 
281     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
282 }
283 
284 static hwaddr spapr_node0_size(MachineState *machine)
285 {
286     if (machine->numa_state->num_nodes) {
287         int i;
288         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
289             if (machine->numa_state->nodes[i].node_mem) {
290                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
291                            machine->ram_size);
292             }
293         }
294     }
295     return machine->ram_size;
296 }
297 
298 static void add_str(GString *s, const gchar *s1)
299 {
300     g_string_append_len(s, s1, strlen(s1) + 1);
301 }
302 
303 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
304                                 hwaddr start, hwaddr size)
305 {
306     char mem_name[32];
307     uint64_t mem_reg_property[2];
308     int off;
309 
310     mem_reg_property[0] = cpu_to_be64(start);
311     mem_reg_property[1] = cpu_to_be64(size);
312 
313     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
314     off = fdt_add_subnode(fdt, 0, mem_name);
315     _FDT(off);
316     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
317     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
318                       sizeof(mem_reg_property))));
319     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
320     return off;
321 }
322 
323 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
324 {
325     MemoryDeviceInfoList *info;
326 
327     for (info = list; info; info = info->next) {
328         MemoryDeviceInfo *value = info->value;
329 
330         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
331             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
332 
333             if (addr >= pcdimm_info->addr &&
334                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
335                 return pcdimm_info->node;
336             }
337         }
338     }
339 
340     return -1;
341 }
342 
343 struct sPAPRDrconfCellV2 {
344      uint32_t seq_lmbs;
345      uint64_t base_addr;
346      uint32_t drc_index;
347      uint32_t aa_index;
348      uint32_t flags;
349 } QEMU_PACKED;
350 
351 typedef struct DrconfCellQueue {
352     struct sPAPRDrconfCellV2 cell;
353     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
354 } DrconfCellQueue;
355 
356 static DrconfCellQueue *
357 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
358                       uint32_t drc_index, uint32_t aa_index,
359                       uint32_t flags)
360 {
361     DrconfCellQueue *elem;
362 
363     elem = g_malloc0(sizeof(*elem));
364     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
365     elem->cell.base_addr = cpu_to_be64(base_addr);
366     elem->cell.drc_index = cpu_to_be32(drc_index);
367     elem->cell.aa_index = cpu_to_be32(aa_index);
368     elem->cell.flags = cpu_to_be32(flags);
369 
370     return elem;
371 }
372 
373 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
374                                       int offset, MemoryDeviceInfoList *dimms)
375 {
376     MachineState *machine = MACHINE(spapr);
377     uint8_t *int_buf, *cur_index;
378     int ret;
379     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
380     uint64_t addr, cur_addr, size;
381     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
382     uint64_t mem_end = machine->device_memory->base +
383                        memory_region_size(&machine->device_memory->mr);
384     uint32_t node, buf_len, nr_entries = 0;
385     SpaprDrc *drc;
386     DrconfCellQueue *elem, *next;
387     MemoryDeviceInfoList *info;
388     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
389         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
390 
391     /* Entry to cover RAM and the gap area */
392     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
393                                  SPAPR_LMB_FLAGS_RESERVED |
394                                  SPAPR_LMB_FLAGS_DRC_INVALID);
395     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
396     nr_entries++;
397 
398     cur_addr = machine->device_memory->base;
399     for (info = dimms; info; info = info->next) {
400         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
401 
402         addr = di->addr;
403         size = di->size;
404         node = di->node;
405 
406         /*
407          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
408          * area is marked hotpluggable in the next iteration for the bigger
409          * chunk including the NVDIMM occupied area.
410          */
411         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
412             continue;
413 
414         /* Entry for hot-pluggable area */
415         if (cur_addr < addr) {
416             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
417             g_assert(drc);
418             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
419                                          cur_addr, spapr_drc_index(drc), -1, 0);
420             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
421             nr_entries++;
422         }
423 
424         /* Entry for DIMM */
425         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
426         g_assert(drc);
427         elem = spapr_get_drconf_cell(size / lmb_size, addr,
428                                      spapr_drc_index(drc), node,
429                                      (SPAPR_LMB_FLAGS_ASSIGNED |
430                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
431         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
432         nr_entries++;
433         cur_addr = addr + size;
434     }
435 
436     /* Entry for remaining hotpluggable area */
437     if (cur_addr < mem_end) {
438         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
439         g_assert(drc);
440         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
441                                      cur_addr, spapr_drc_index(drc), -1, 0);
442         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
443         nr_entries++;
444     }
445 
446     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
447     int_buf = cur_index = g_malloc0(buf_len);
448     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
449     cur_index += sizeof(nr_entries);
450 
451     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
452         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
453         cur_index += sizeof(elem->cell);
454         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
455         g_free(elem);
456     }
457 
458     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
459     g_free(int_buf);
460     if (ret < 0) {
461         return -1;
462     }
463     return 0;
464 }
465 
466 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
467                                    int offset, MemoryDeviceInfoList *dimms)
468 {
469     MachineState *machine = MACHINE(spapr);
470     int i, ret;
471     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
472     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
473     uint32_t nr_lmbs = (machine->device_memory->base +
474                        memory_region_size(&machine->device_memory->mr)) /
475                        lmb_size;
476     uint32_t *int_buf, *cur_index, buf_len;
477 
478     /*
479      * Allocate enough buffer size to fit in ibm,dynamic-memory
480      */
481     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
482     cur_index = int_buf = g_malloc0(buf_len);
483     int_buf[0] = cpu_to_be32(nr_lmbs);
484     cur_index++;
485     for (i = 0; i < nr_lmbs; i++) {
486         uint64_t addr = i * lmb_size;
487         uint32_t *dynamic_memory = cur_index;
488 
489         if (i >= device_lmb_start) {
490             SpaprDrc *drc;
491 
492             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
493             g_assert(drc);
494 
495             dynamic_memory[0] = cpu_to_be32(addr >> 32);
496             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
497             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
498             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
499             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
500             if (memory_region_present(get_system_memory(), addr)) {
501                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
502             } else {
503                 dynamic_memory[5] = cpu_to_be32(0);
504             }
505         } else {
506             /*
507              * LMB information for RMA, boot time RAM and gap b/n RAM and
508              * device memory region -- all these are marked as reserved
509              * and as having no valid DRC.
510              */
511             dynamic_memory[0] = cpu_to_be32(addr >> 32);
512             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
513             dynamic_memory[2] = cpu_to_be32(0);
514             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
515             dynamic_memory[4] = cpu_to_be32(-1);
516             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
517                                             SPAPR_LMB_FLAGS_DRC_INVALID);
518         }
519 
520         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
521     }
522     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
523     g_free(int_buf);
524     if (ret < 0) {
525         return -1;
526     }
527     return 0;
528 }
529 
530 /*
531  * Adds ibm,dynamic-reconfiguration-memory node.
532  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
533  * of this device tree node.
534  */
535 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
536                                                    void *fdt)
537 {
538     MachineState *machine = MACHINE(spapr);
539     int ret, offset;
540     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
541     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
542                                 cpu_to_be32(lmb_size & 0xffffffff)};
543     MemoryDeviceInfoList *dimms = NULL;
544 
545     /*
546      * Don't create the node if there is no device memory
547      */
548     if (machine->ram_size == machine->maxram_size) {
549         return 0;
550     }
551 
552     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
553 
554     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
555                     sizeof(prop_lmb_size));
556     if (ret < 0) {
557         return ret;
558     }
559 
560     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
561     if (ret < 0) {
562         return ret;
563     }
564 
565     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
566     if (ret < 0) {
567         return ret;
568     }
569 
570     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
571     dimms = qmp_memory_device_list();
572     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
573         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
574     } else {
575         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
576     }
577     qapi_free_MemoryDeviceInfoList(dimms);
578 
579     if (ret < 0) {
580         return ret;
581     }
582 
583     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
584 
585     return ret;
586 }
587 
588 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
589 {
590     MachineState *machine = MACHINE(spapr);
591     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
592     hwaddr mem_start, node_size;
593     int i, nb_nodes = machine->numa_state->num_nodes;
594     NodeInfo *nodes = machine->numa_state->nodes;
595 
596     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
597         if (!nodes[i].node_mem) {
598             continue;
599         }
600         if (mem_start >= machine->ram_size) {
601             node_size = 0;
602         } else {
603             node_size = nodes[i].node_mem;
604             if (node_size > machine->ram_size - mem_start) {
605                 node_size = machine->ram_size - mem_start;
606             }
607         }
608         if (!mem_start) {
609             /* spapr_machine_init() checks for rma_size <= node0_size
610              * already */
611             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
612             mem_start += spapr->rma_size;
613             node_size -= spapr->rma_size;
614         }
615         for ( ; node_size; ) {
616             hwaddr sizetmp = pow2floor(node_size);
617 
618             /* mem_start != 0 here */
619             if (ctzl(mem_start) < ctzl(sizetmp)) {
620                 sizetmp = 1ULL << ctzl(mem_start);
621             }
622 
623             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
624             node_size -= sizetmp;
625             mem_start += sizetmp;
626         }
627     }
628 
629     /* Generate ibm,dynamic-reconfiguration-memory node if required */
630     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
631         int ret;
632 
633         g_assert(smc->dr_lmb_enabled);
634         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
635         if (ret) {
636             return ret;
637         }
638     }
639 
640     return 0;
641 }
642 
643 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
644                          SpaprMachineState *spapr)
645 {
646     MachineState *ms = MACHINE(spapr);
647     PowerPCCPU *cpu = POWERPC_CPU(cs);
648     CPUPPCState *env = &cpu->env;
649     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
650     int index = spapr_get_vcpu_id(cpu);
651     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
652                        0xffffffff, 0xffffffff};
653     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
654         : SPAPR_TIMEBASE_FREQ;
655     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
656     uint32_t page_sizes_prop[64];
657     size_t page_sizes_prop_size;
658     unsigned int smp_threads = ms->smp.threads;
659     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
660     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
661     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
662     SpaprDrc *drc;
663     int drc_index;
664     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
665     int i;
666 
667     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
668     if (drc) {
669         drc_index = spapr_drc_index(drc);
670         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
671     }
672 
673     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
674     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
675 
676     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
677     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
678                            env->dcache_line_size)));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
682                            env->icache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
684                            env->icache_line_size)));
685 
686     if (pcc->l1_dcache_size) {
687         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
688                                pcc->l1_dcache_size)));
689     } else {
690         warn_report("Unknown L1 dcache size for cpu");
691     }
692     if (pcc->l1_icache_size) {
693         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
694                                pcc->l1_icache_size)));
695     } else {
696         warn_report("Unknown L1 icache size for cpu");
697     }
698 
699     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
700     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
701     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
702     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
703     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
704     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
705 
706     if (env->spr_cb[SPR_PURR].oea_read) {
707         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
708     }
709     if (env->spr_cb[SPR_SPURR].oea_read) {
710         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
711     }
712 
713     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
714         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
715                           segs, sizeof(segs))));
716     }
717 
718     /* Advertise VSX (vector extensions) if available
719      *   1               == VMX / Altivec available
720      *   2               == VSX available
721      *
722      * Only CPUs for which we create core types in spapr_cpu_core.c
723      * are possible, and all of those have VMX */
724     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
725         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
726     } else {
727         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
728     }
729 
730     /* Advertise DFP (Decimal Floating Point) if available
731      *   0 / no property == no DFP
732      *   1               == DFP available */
733     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
734         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
735     }
736 
737     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
738                                                       sizeof(page_sizes_prop));
739     if (page_sizes_prop_size) {
740         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
741                           page_sizes_prop, page_sizes_prop_size)));
742     }
743 
744     spapr_dt_pa_features(spapr, cpu, fdt, offset);
745 
746     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
747                            cs->cpu_index / vcpus_per_socket)));
748 
749     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
750                       pft_size_prop, sizeof(pft_size_prop))));
751 
752     if (ms->numa_state->num_nodes > 1) {
753         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
754     }
755 
756     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
757 
758     if (pcc->radix_page_info) {
759         for (i = 0; i < pcc->radix_page_info->count; i++) {
760             radix_AP_encodings[i] =
761                 cpu_to_be32(pcc->radix_page_info->entries[i]);
762         }
763         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
764                           radix_AP_encodings,
765                           pcc->radix_page_info->count *
766                           sizeof(radix_AP_encodings[0]))));
767     }
768 
769     /*
770      * We set this property to let the guest know that it can use the large
771      * decrementer and its width in bits.
772      */
773     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
774         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
775                               pcc->lrg_decr_bits)));
776 }
777 
778 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
779 {
780     CPUState **rev;
781     CPUState *cs;
782     int n_cpus;
783     int cpus_offset;
784     int i;
785 
786     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
787     _FDT(cpus_offset);
788     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
789     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
790 
791     /*
792      * We walk the CPUs in reverse order to ensure that CPU DT nodes
793      * created by fdt_add_subnode() end up in the right order in FDT
794      * for the guest kernel the enumerate the CPUs correctly.
795      *
796      * The CPU list cannot be traversed in reverse order, so we need
797      * to do extra work.
798      */
799     n_cpus = 0;
800     rev = NULL;
801     CPU_FOREACH(cs) {
802         rev = g_renew(CPUState *, rev, n_cpus + 1);
803         rev[n_cpus++] = cs;
804     }
805 
806     for (i = n_cpus - 1; i >= 0; i--) {
807         CPUState *cs = rev[i];
808         PowerPCCPU *cpu = POWERPC_CPU(cs);
809         int index = spapr_get_vcpu_id(cpu);
810         DeviceClass *dc = DEVICE_GET_CLASS(cs);
811         g_autofree char *nodename = NULL;
812         int offset;
813 
814         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
815             continue;
816         }
817 
818         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
819         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
820         _FDT(offset);
821         spapr_dt_cpu(cs, fdt, offset, spapr);
822     }
823 
824     g_free(rev);
825 }
826 
827 static int spapr_dt_rng(void *fdt)
828 {
829     int node;
830     int ret;
831 
832     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
833     if (node <= 0) {
834         return -1;
835     }
836     ret = fdt_setprop_string(fdt, node, "device_type",
837                              "ibm,platform-facilities");
838     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
839     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
840 
841     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
842     if (node <= 0) {
843         return -1;
844     }
845     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
846 
847     return ret ? -1 : 0;
848 }
849 
850 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
851 {
852     MachineState *ms = MACHINE(spapr);
853     int rtas;
854     GString *hypertas = g_string_sized_new(256);
855     GString *qemu_hypertas = g_string_sized_new(256);
856     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
857         memory_region_size(&MACHINE(spapr)->device_memory->mr);
858     uint32_t lrdr_capacity[] = {
859         cpu_to_be32(max_device_addr >> 32),
860         cpu_to_be32(max_device_addr & 0xffffffff),
861         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
862         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
863         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
864     };
865 
866     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
867 
868     /* hypertas */
869     add_str(hypertas, "hcall-pft");
870     add_str(hypertas, "hcall-term");
871     add_str(hypertas, "hcall-dabr");
872     add_str(hypertas, "hcall-interrupt");
873     add_str(hypertas, "hcall-tce");
874     add_str(hypertas, "hcall-vio");
875     add_str(hypertas, "hcall-splpar");
876     add_str(hypertas, "hcall-join");
877     add_str(hypertas, "hcall-bulk");
878     add_str(hypertas, "hcall-set-mode");
879     add_str(hypertas, "hcall-sprg0");
880     add_str(hypertas, "hcall-copy");
881     add_str(hypertas, "hcall-debug");
882     add_str(hypertas, "hcall-vphn");
883     add_str(qemu_hypertas, "hcall-memop1");
884 
885     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
886         add_str(hypertas, "hcall-multi-tce");
887     }
888 
889     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
890         add_str(hypertas, "hcall-hpt-resize");
891     }
892 
893     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
894                      hypertas->str, hypertas->len));
895     g_string_free(hypertas, TRUE);
896     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
897                      qemu_hypertas->str, qemu_hypertas->len));
898     g_string_free(qemu_hypertas, TRUE);
899 
900     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
901 
902     /*
903      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
904      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
905      *
906      * The system reset requirements are driven by existing Linux and PowerVM
907      * implementation which (contrary to PAPR) saves r3 in the error log
908      * structure like machine check, so Linux expects to find the saved r3
909      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
910      * does not look at the error value).
911      *
912      * System reset interrupts are not subject to interlock like machine
913      * check, so this memory area could be corrupted if the sreset is
914      * interrupted by a machine check (or vice versa) if it was shared. To
915      * prevent this, system reset uses per-CPU areas for the sreset save
916      * area. A system reset that interrupts a system reset handler could
917      * still overwrite this area, but Linux doesn't try to recover in that
918      * case anyway.
919      *
920      * The extra 8 bytes is required because Linux's FWNMI error log check
921      * is off-by-one.
922      */
923     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
924 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
925     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
926                           RTAS_ERROR_LOG_MAX));
927     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
928                           RTAS_EVENT_SCAN_RATE));
929 
930     g_assert(msi_nonbroken);
931     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
932 
933     /*
934      * According to PAPR, rtas ibm,os-term does not guarantee a return
935      * back to the guest cpu.
936      *
937      * While an additional ibm,extended-os-term property indicates
938      * that rtas call return will always occur. Set this property.
939      */
940     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
941 
942     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
943                      lrdr_capacity, sizeof(lrdr_capacity)));
944 
945     spapr_dt_rtas_tokens(fdt, rtas);
946 }
947 
948 /*
949  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
950  * and the XIVE features that the guest may request and thus the valid
951  * values for bytes 23..26 of option vector 5:
952  */
953 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
954                                           int chosen)
955 {
956     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
957 
958     char val[2 * 4] = {
959         23, 0x00, /* XICS / XIVE mode */
960         24, 0x00, /* Hash/Radix, filled in below. */
961         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
962         26, 0x40, /* Radix options: GTSE == yes. */
963     };
964 
965     if (spapr->irq->xics && spapr->irq->xive) {
966         val[1] = SPAPR_OV5_XIVE_BOTH;
967     } else if (spapr->irq->xive) {
968         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
969     } else {
970         assert(spapr->irq->xics);
971         val[1] = SPAPR_OV5_XIVE_LEGACY;
972     }
973 
974     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
975                           first_ppc_cpu->compat_pvr)) {
976         /*
977          * If we're in a pre POWER9 compat mode then the guest should
978          * do hash and use the legacy interrupt mode
979          */
980         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
981         val[3] = 0x00; /* Hash */
982     } else if (kvm_enabled()) {
983         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
984             val[3] = 0x80; /* OV5_MMU_BOTH */
985         } else if (kvmppc_has_cap_mmu_radix()) {
986             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
987         } else {
988             val[3] = 0x00; /* Hash */
989         }
990     } else {
991         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
992         val[3] = 0xC0;
993     }
994     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
995                      val, sizeof(val)));
996 }
997 
998 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
999 {
1000     MachineState *machine = MACHINE(spapr);
1001     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1002     int chosen;
1003 
1004     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1005 
1006     if (reset) {
1007         const char *boot_device = machine->boot_order;
1008         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1009         size_t cb = 0;
1010         char *bootlist = get_boot_devices_list(&cb);
1011 
1012         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1013             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1014                                     machine->kernel_cmdline));
1015         }
1016 
1017         if (spapr->initrd_size) {
1018             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1019                                   spapr->initrd_base));
1020             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1021                                   spapr->initrd_base + spapr->initrd_size));
1022         }
1023 
1024         if (spapr->kernel_size) {
1025             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1026                                   cpu_to_be64(spapr->kernel_size) };
1027 
1028             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1029                          &kprop, sizeof(kprop)));
1030             if (spapr->kernel_le) {
1031                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1032             }
1033         }
1034         if (boot_menu) {
1035             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1036         }
1037         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1038         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1039         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1040 
1041         if (cb && bootlist) {
1042             int i;
1043 
1044             for (i = 0; i < cb; i++) {
1045                 if (bootlist[i] == '\n') {
1046                     bootlist[i] = ' ';
1047                 }
1048             }
1049             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1050         }
1051 
1052         if (boot_device && strlen(boot_device)) {
1053             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1054         }
1055 
1056         if (!spapr->has_graphics && stdout_path) {
1057             /*
1058              * "linux,stdout-path" and "stdout" properties are
1059              * deprecated by linux kernel. New platforms should only
1060              * use the "stdout-path" property. Set the new property
1061              * and continue using older property to remain compatible
1062              * with the existing firmware.
1063              */
1064             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1065             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1066         }
1067 
1068         /*
1069          * We can deal with BAR reallocation just fine, advertise it
1070          * to the guest
1071          */
1072         if (smc->linux_pci_probe) {
1073             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1074         }
1075 
1076         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1077 
1078         g_free(stdout_path);
1079         g_free(bootlist);
1080     }
1081 
1082     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1083 }
1084 
1085 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1086 {
1087     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1088      * KVM to work under pHyp with some guest co-operation */
1089     int hypervisor;
1090     uint8_t hypercall[16];
1091 
1092     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1093     /* indicate KVM hypercall interface */
1094     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1095     if (kvmppc_has_cap_fixup_hcalls()) {
1096         /*
1097          * Older KVM versions with older guest kernels were broken
1098          * with the magic page, don't allow the guest to map it.
1099          */
1100         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1101                                   sizeof(hypercall))) {
1102             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1103                              hypercall, sizeof(hypercall)));
1104         }
1105     }
1106 }
1107 
1108 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1109 {
1110     MachineState *machine = MACHINE(spapr);
1111     MachineClass *mc = MACHINE_GET_CLASS(machine);
1112     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1113     uint32_t root_drc_type_mask = 0;
1114     int ret;
1115     void *fdt;
1116     SpaprPhbState *phb;
1117     char *buf;
1118 
1119     fdt = g_malloc0(space);
1120     _FDT((fdt_create_empty_tree(fdt, space)));
1121 
1122     /* Root node */
1123     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1124     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1125     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1126 
1127     /* Guest UUID & Name*/
1128     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1129     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1130     if (qemu_uuid_set) {
1131         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1132     }
1133     g_free(buf);
1134 
1135     if (qemu_get_vm_name()) {
1136         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1137                                 qemu_get_vm_name()));
1138     }
1139 
1140     /* Host Model & Serial Number */
1141     if (spapr->host_model) {
1142         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1143     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1144         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1145         g_free(buf);
1146     }
1147 
1148     if (spapr->host_serial) {
1149         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1150     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1151         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1152         g_free(buf);
1153     }
1154 
1155     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1156     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1157 
1158     /* /interrupt controller */
1159     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1160 
1161     ret = spapr_dt_memory(spapr, fdt);
1162     if (ret < 0) {
1163         error_report("couldn't setup memory nodes in fdt");
1164         exit(1);
1165     }
1166 
1167     /* /vdevice */
1168     spapr_dt_vdevice(spapr->vio_bus, fdt);
1169 
1170     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1171         ret = spapr_dt_rng(fdt);
1172         if (ret < 0) {
1173             error_report("could not set up rng device in the fdt");
1174             exit(1);
1175         }
1176     }
1177 
1178     QLIST_FOREACH(phb, &spapr->phbs, list) {
1179         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1180         if (ret < 0) {
1181             error_report("couldn't setup PCI devices in fdt");
1182             exit(1);
1183         }
1184     }
1185 
1186     spapr_dt_cpus(fdt, spapr);
1187 
1188     /* ibm,drc-indexes and friends */
1189     if (smc->dr_lmb_enabled) {
1190         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1191     }
1192     if (smc->dr_phb_enabled) {
1193         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1194     }
1195     if (mc->nvdimm_supported) {
1196         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1197     }
1198     if (root_drc_type_mask) {
1199         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1200     }
1201 
1202     if (mc->has_hotpluggable_cpus) {
1203         int offset = fdt_path_offset(fdt, "/cpus");
1204         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1205         if (ret < 0) {
1206             error_report("Couldn't set up CPU DR device tree properties");
1207             exit(1);
1208         }
1209     }
1210 
1211     /* /event-sources */
1212     spapr_dt_events(spapr, fdt);
1213 
1214     /* /rtas */
1215     spapr_dt_rtas(spapr, fdt);
1216 
1217     /* /chosen */
1218     spapr_dt_chosen(spapr, fdt, reset);
1219 
1220     /* /hypervisor */
1221     if (kvm_enabled()) {
1222         spapr_dt_hypervisor(spapr, fdt);
1223     }
1224 
1225     /* Build memory reserve map */
1226     if (reset) {
1227         if (spapr->kernel_size) {
1228             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1229                                   spapr->kernel_size)));
1230         }
1231         if (spapr->initrd_size) {
1232             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1233                                   spapr->initrd_size)));
1234         }
1235     }
1236 
1237     /* NVDIMM devices */
1238     if (mc->nvdimm_supported) {
1239         spapr_dt_persistent_memory(spapr, fdt);
1240     }
1241 
1242     return fdt;
1243 }
1244 
1245 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1246 {
1247     SpaprMachineState *spapr = opaque;
1248 
1249     return (addr & 0x0fffffff) + spapr->kernel_addr;
1250 }
1251 
1252 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1253                                     PowerPCCPU *cpu)
1254 {
1255     CPUPPCState *env = &cpu->env;
1256 
1257     /* The TCG path should also be holding the BQL at this point */
1258     g_assert(qemu_mutex_iothread_locked());
1259 
1260     if (msr_pr) {
1261         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1262         env->gpr[3] = H_PRIVILEGE;
1263     } else {
1264         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1265     }
1266 }
1267 
1268 struct LPCRSyncState {
1269     target_ulong value;
1270     target_ulong mask;
1271 };
1272 
1273 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1274 {
1275     struct LPCRSyncState *s = arg.host_ptr;
1276     PowerPCCPU *cpu = POWERPC_CPU(cs);
1277     CPUPPCState *env = &cpu->env;
1278     target_ulong lpcr;
1279 
1280     cpu_synchronize_state(cs);
1281     lpcr = env->spr[SPR_LPCR];
1282     lpcr &= ~s->mask;
1283     lpcr |= s->value;
1284     ppc_store_lpcr(cpu, lpcr);
1285 }
1286 
1287 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1288 {
1289     CPUState *cs;
1290     struct LPCRSyncState s = {
1291         .value = value,
1292         .mask = mask
1293     };
1294     CPU_FOREACH(cs) {
1295         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1296     }
1297 }
1298 
1299 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1300 {
1301     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1302 
1303     /* Copy PATE1:GR into PATE0:HR */
1304     entry->dw0 = spapr->patb_entry & PATE0_HR;
1305     entry->dw1 = spapr->patb_entry;
1306 }
1307 
1308 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1309 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1310 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1311 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1312 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1313 
1314 /*
1315  * Get the fd to access the kernel htab, re-opening it if necessary
1316  */
1317 static int get_htab_fd(SpaprMachineState *spapr)
1318 {
1319     Error *local_err = NULL;
1320 
1321     if (spapr->htab_fd >= 0) {
1322         return spapr->htab_fd;
1323     }
1324 
1325     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1326     if (spapr->htab_fd < 0) {
1327         error_report_err(local_err);
1328     }
1329 
1330     return spapr->htab_fd;
1331 }
1332 
1333 void close_htab_fd(SpaprMachineState *spapr)
1334 {
1335     if (spapr->htab_fd >= 0) {
1336         close(spapr->htab_fd);
1337     }
1338     spapr->htab_fd = -1;
1339 }
1340 
1341 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1342 {
1343     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1344 
1345     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1346 }
1347 
1348 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1349 {
1350     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1351 
1352     assert(kvm_enabled());
1353 
1354     if (!spapr->htab) {
1355         return 0;
1356     }
1357 
1358     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1359 }
1360 
1361 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1362                                                 hwaddr ptex, int n)
1363 {
1364     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1365     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1366 
1367     if (!spapr->htab) {
1368         /*
1369          * HTAB is controlled by KVM. Fetch into temporary buffer
1370          */
1371         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1372         kvmppc_read_hptes(hptes, ptex, n);
1373         return hptes;
1374     }
1375 
1376     /*
1377      * HTAB is controlled by QEMU. Just point to the internally
1378      * accessible PTEG.
1379      */
1380     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1381 }
1382 
1383 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1384                               const ppc_hash_pte64_t *hptes,
1385                               hwaddr ptex, int n)
1386 {
1387     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1388 
1389     if (!spapr->htab) {
1390         g_free((void *)hptes);
1391     }
1392 
1393     /* Nothing to do for qemu managed HPT */
1394 }
1395 
1396 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1397                       uint64_t pte0, uint64_t pte1)
1398 {
1399     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1400     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1401 
1402     if (!spapr->htab) {
1403         kvmppc_write_hpte(ptex, pte0, pte1);
1404     } else {
1405         if (pte0 & HPTE64_V_VALID) {
1406             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1407             /*
1408              * When setting valid, we write PTE1 first. This ensures
1409              * proper synchronization with the reading code in
1410              * ppc_hash64_pteg_search()
1411              */
1412             smp_wmb();
1413             stq_p(spapr->htab + offset, pte0);
1414         } else {
1415             stq_p(spapr->htab + offset, pte0);
1416             /*
1417              * When clearing it we set PTE0 first. This ensures proper
1418              * synchronization with the reading code in
1419              * ppc_hash64_pteg_search()
1420              */
1421             smp_wmb();
1422             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1423         }
1424     }
1425 }
1426 
1427 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1428                              uint64_t pte1)
1429 {
1430     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1431     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1432 
1433     if (!spapr->htab) {
1434         /* There should always be a hash table when this is called */
1435         error_report("spapr_hpte_set_c called with no hash table !");
1436         return;
1437     }
1438 
1439     /* The HW performs a non-atomic byte update */
1440     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1441 }
1442 
1443 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1444                              uint64_t pte1)
1445 {
1446     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1447     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1448 
1449     if (!spapr->htab) {
1450         /* There should always be a hash table when this is called */
1451         error_report("spapr_hpte_set_r called with no hash table !");
1452         return;
1453     }
1454 
1455     /* The HW performs a non-atomic byte update */
1456     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1457 }
1458 
1459 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1460 {
1461     int shift;
1462 
1463     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1464      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1465      * that's much more than is needed for Linux guests */
1466     shift = ctz64(pow2ceil(ramsize)) - 7;
1467     shift = MAX(shift, 18); /* Minimum architected size */
1468     shift = MIN(shift, 46); /* Maximum architected size */
1469     return shift;
1470 }
1471 
1472 void spapr_free_hpt(SpaprMachineState *spapr)
1473 {
1474     g_free(spapr->htab);
1475     spapr->htab = NULL;
1476     spapr->htab_shift = 0;
1477     close_htab_fd(spapr);
1478 }
1479 
1480 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1481 {
1482     ERRP_GUARD();
1483     long rc;
1484 
1485     /* Clean up any HPT info from a previous boot */
1486     spapr_free_hpt(spapr);
1487 
1488     rc = kvmppc_reset_htab(shift);
1489 
1490     if (rc == -EOPNOTSUPP) {
1491         error_setg(errp, "HPT not supported in nested guests");
1492         return -EOPNOTSUPP;
1493     }
1494 
1495     if (rc < 0) {
1496         /* kernel-side HPT needed, but couldn't allocate one */
1497         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1498                          shift);
1499         error_append_hint(errp, "Try smaller maxmem?\n");
1500         return -errno;
1501     } else if (rc > 0) {
1502         /* kernel-side HPT allocated */
1503         if (rc != shift) {
1504             error_setg(errp,
1505                        "Requested order %d HPT, but kernel allocated order %ld",
1506                        shift, rc);
1507             error_append_hint(errp, "Try smaller maxmem?\n");
1508             return -ENOSPC;
1509         }
1510 
1511         spapr->htab_shift = shift;
1512         spapr->htab = NULL;
1513     } else {
1514         /* kernel-side HPT not needed, allocate in userspace instead */
1515         size_t size = 1ULL << shift;
1516         int i;
1517 
1518         spapr->htab = qemu_memalign(size, size);
1519         memset(spapr->htab, 0, size);
1520         spapr->htab_shift = shift;
1521 
1522         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1523             DIRTY_HPTE(HPTE(spapr->htab, i));
1524         }
1525     }
1526     /* We're setting up a hash table, so that means we're not radix */
1527     spapr->patb_entry = 0;
1528     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1529     return 0;
1530 }
1531 
1532 void spapr_setup_hpt(SpaprMachineState *spapr)
1533 {
1534     int hpt_shift;
1535 
1536     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1537         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1538     } else {
1539         uint64_t current_ram_size;
1540 
1541         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1542         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1543     }
1544     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1545 
1546     if (kvm_enabled()) {
1547         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1548 
1549         /* Check our RMA fits in the possible VRMA */
1550         if (vrma_limit < spapr->rma_size) {
1551             error_report("Unable to create %" HWADDR_PRIu
1552                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1553                          spapr->rma_size / MiB, vrma_limit / MiB);
1554             exit(EXIT_FAILURE);
1555         }
1556     }
1557 }
1558 
1559 static void spapr_machine_reset(MachineState *machine)
1560 {
1561     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1562     PowerPCCPU *first_ppc_cpu;
1563     hwaddr fdt_addr;
1564     void *fdt;
1565     int rc;
1566 
1567     pef_kvm_reset(machine->cgs, &error_fatal);
1568     spapr_caps_apply(spapr);
1569 
1570     first_ppc_cpu = POWERPC_CPU(first_cpu);
1571     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1572         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1573                               spapr->max_compat_pvr)) {
1574         /*
1575          * If using KVM with radix mode available, VCPUs can be started
1576          * without a HPT because KVM will start them in radix mode.
1577          * Set the GR bit in PATE so that we know there is no HPT.
1578          */
1579         spapr->patb_entry = PATE1_GR;
1580         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1581     } else {
1582         spapr_setup_hpt(spapr);
1583     }
1584 
1585     qemu_devices_reset();
1586 
1587     spapr_ovec_cleanup(spapr->ov5_cas);
1588     spapr->ov5_cas = spapr_ovec_new();
1589 
1590     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1591 
1592     /*
1593      * This is fixing some of the default configuration of the XIVE
1594      * devices. To be called after the reset of the machine devices.
1595      */
1596     spapr_irq_reset(spapr, &error_fatal);
1597 
1598     /*
1599      * There is no CAS under qtest. Simulate one to please the code that
1600      * depends on spapr->ov5_cas. This is especially needed to test device
1601      * unplug, so we do that before resetting the DRCs.
1602      */
1603     if (qtest_enabled()) {
1604         spapr_ovec_cleanup(spapr->ov5_cas);
1605         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1606     }
1607 
1608     /* DRC reset may cause a device to be unplugged. This will cause troubles
1609      * if this device is used by another device (eg, a running vhost backend
1610      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1611      * situations, we reset DRCs after all devices have been reset.
1612      */
1613     spapr_drc_reset_all(spapr);
1614 
1615     spapr_clear_pending_events(spapr);
1616 
1617     /*
1618      * We place the device tree just below either the top of the RMA,
1619      * or just below 2GB, whichever is lower, so that it can be
1620      * processed with 32-bit real mode code if necessary
1621      */
1622     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1623 
1624     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1625 
1626     rc = fdt_pack(fdt);
1627 
1628     /* Should only fail if we've built a corrupted tree */
1629     assert(rc == 0);
1630 
1631     /* Load the fdt */
1632     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1633     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1634     g_free(spapr->fdt_blob);
1635     spapr->fdt_size = fdt_totalsize(fdt);
1636     spapr->fdt_initial_size = spapr->fdt_size;
1637     spapr->fdt_blob = fdt;
1638 
1639     /* Set up the entry state */
1640     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1641     first_ppc_cpu->env.gpr[5] = 0;
1642 
1643     spapr->fwnmi_system_reset_addr = -1;
1644     spapr->fwnmi_machine_check_addr = -1;
1645     spapr->fwnmi_machine_check_interlock = -1;
1646 
1647     /* Signal all vCPUs waiting on this condition */
1648     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1649 
1650     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1651 }
1652 
1653 static void spapr_create_nvram(SpaprMachineState *spapr)
1654 {
1655     DeviceState *dev = qdev_new("spapr-nvram");
1656     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1657 
1658     if (dinfo) {
1659         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1660                                 &error_fatal);
1661     }
1662 
1663     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1664 
1665     spapr->nvram = (struct SpaprNvram *)dev;
1666 }
1667 
1668 static void spapr_rtc_create(SpaprMachineState *spapr)
1669 {
1670     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1671                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1672                                        &error_fatal, NULL);
1673     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1674     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1675                               "date");
1676 }
1677 
1678 /* Returns whether we want to use VGA or not */
1679 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1680 {
1681     switch (vga_interface_type) {
1682     case VGA_NONE:
1683         return false;
1684     case VGA_DEVICE:
1685         return true;
1686     case VGA_STD:
1687     case VGA_VIRTIO:
1688     case VGA_CIRRUS:
1689         return pci_vga_init(pci_bus) != NULL;
1690     default:
1691         error_setg(errp,
1692                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1693         return false;
1694     }
1695 }
1696 
1697 static int spapr_pre_load(void *opaque)
1698 {
1699     int rc;
1700 
1701     rc = spapr_caps_pre_load(opaque);
1702     if (rc) {
1703         return rc;
1704     }
1705 
1706     return 0;
1707 }
1708 
1709 static int spapr_post_load(void *opaque, int version_id)
1710 {
1711     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1712     int err = 0;
1713 
1714     err = spapr_caps_post_migration(spapr);
1715     if (err) {
1716         return err;
1717     }
1718 
1719     /*
1720      * In earlier versions, there was no separate qdev for the PAPR
1721      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1722      * So when migrating from those versions, poke the incoming offset
1723      * value into the RTC device
1724      */
1725     if (version_id < 3) {
1726         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1727         if (err) {
1728             return err;
1729         }
1730     }
1731 
1732     if (kvm_enabled() && spapr->patb_entry) {
1733         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1734         bool radix = !!(spapr->patb_entry & PATE1_GR);
1735         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1736 
1737         /*
1738          * Update LPCR:HR and UPRT as they may not be set properly in
1739          * the stream
1740          */
1741         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1742                             LPCR_HR | LPCR_UPRT);
1743 
1744         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1745         if (err) {
1746             error_report("Process table config unsupported by the host");
1747             return -EINVAL;
1748         }
1749     }
1750 
1751     err = spapr_irq_post_load(spapr, version_id);
1752     if (err) {
1753         return err;
1754     }
1755 
1756     return err;
1757 }
1758 
1759 static int spapr_pre_save(void *opaque)
1760 {
1761     int rc;
1762 
1763     rc = spapr_caps_pre_save(opaque);
1764     if (rc) {
1765         return rc;
1766     }
1767 
1768     return 0;
1769 }
1770 
1771 static bool version_before_3(void *opaque, int version_id)
1772 {
1773     return version_id < 3;
1774 }
1775 
1776 static bool spapr_pending_events_needed(void *opaque)
1777 {
1778     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1779     return !QTAILQ_EMPTY(&spapr->pending_events);
1780 }
1781 
1782 static const VMStateDescription vmstate_spapr_event_entry = {
1783     .name = "spapr_event_log_entry",
1784     .version_id = 1,
1785     .minimum_version_id = 1,
1786     .fields = (VMStateField[]) {
1787         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1788         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1789         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1790                                      NULL, extended_length),
1791         VMSTATE_END_OF_LIST()
1792     },
1793 };
1794 
1795 static const VMStateDescription vmstate_spapr_pending_events = {
1796     .name = "spapr_pending_events",
1797     .version_id = 1,
1798     .minimum_version_id = 1,
1799     .needed = spapr_pending_events_needed,
1800     .fields = (VMStateField[]) {
1801         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1802                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1803         VMSTATE_END_OF_LIST()
1804     },
1805 };
1806 
1807 static bool spapr_ov5_cas_needed(void *opaque)
1808 {
1809     SpaprMachineState *spapr = opaque;
1810     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1811     bool cas_needed;
1812 
1813     /* Prior to the introduction of SpaprOptionVector, we had two option
1814      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1815      * Both of these options encode machine topology into the device-tree
1816      * in such a way that the now-booted OS should still be able to interact
1817      * appropriately with QEMU regardless of what options were actually
1818      * negotiatied on the source side.
1819      *
1820      * As such, we can avoid migrating the CAS-negotiated options if these
1821      * are the only options available on the current machine/platform.
1822      * Since these are the only options available for pseries-2.7 and
1823      * earlier, this allows us to maintain old->new/new->old migration
1824      * compatibility.
1825      *
1826      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1827      * via default pseries-2.8 machines and explicit command-line parameters.
1828      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1829      * of the actual CAS-negotiated values to continue working properly. For
1830      * example, availability of memory unplug depends on knowing whether
1831      * OV5_HP_EVT was negotiated via CAS.
1832      *
1833      * Thus, for any cases where the set of available CAS-negotiatable
1834      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1835      * include the CAS-negotiated options in the migration stream, unless
1836      * if they affect boot time behaviour only.
1837      */
1838     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1839     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1840     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1841 
1842     /* We need extra information if we have any bits outside the mask
1843      * defined above */
1844     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1845 
1846     spapr_ovec_cleanup(ov5_mask);
1847 
1848     return cas_needed;
1849 }
1850 
1851 static const VMStateDescription vmstate_spapr_ov5_cas = {
1852     .name = "spapr_option_vector_ov5_cas",
1853     .version_id = 1,
1854     .minimum_version_id = 1,
1855     .needed = spapr_ov5_cas_needed,
1856     .fields = (VMStateField[]) {
1857         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1858                                  vmstate_spapr_ovec, SpaprOptionVector),
1859         VMSTATE_END_OF_LIST()
1860     },
1861 };
1862 
1863 static bool spapr_patb_entry_needed(void *opaque)
1864 {
1865     SpaprMachineState *spapr = opaque;
1866 
1867     return !!spapr->patb_entry;
1868 }
1869 
1870 static const VMStateDescription vmstate_spapr_patb_entry = {
1871     .name = "spapr_patb_entry",
1872     .version_id = 1,
1873     .minimum_version_id = 1,
1874     .needed = spapr_patb_entry_needed,
1875     .fields = (VMStateField[]) {
1876         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1877         VMSTATE_END_OF_LIST()
1878     },
1879 };
1880 
1881 static bool spapr_irq_map_needed(void *opaque)
1882 {
1883     SpaprMachineState *spapr = opaque;
1884 
1885     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1886 }
1887 
1888 static const VMStateDescription vmstate_spapr_irq_map = {
1889     .name = "spapr_irq_map",
1890     .version_id = 1,
1891     .minimum_version_id = 1,
1892     .needed = spapr_irq_map_needed,
1893     .fields = (VMStateField[]) {
1894         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1895         VMSTATE_END_OF_LIST()
1896     },
1897 };
1898 
1899 static bool spapr_dtb_needed(void *opaque)
1900 {
1901     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1902 
1903     return smc->update_dt_enabled;
1904 }
1905 
1906 static int spapr_dtb_pre_load(void *opaque)
1907 {
1908     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1909 
1910     g_free(spapr->fdt_blob);
1911     spapr->fdt_blob = NULL;
1912     spapr->fdt_size = 0;
1913 
1914     return 0;
1915 }
1916 
1917 static const VMStateDescription vmstate_spapr_dtb = {
1918     .name = "spapr_dtb",
1919     .version_id = 1,
1920     .minimum_version_id = 1,
1921     .needed = spapr_dtb_needed,
1922     .pre_load = spapr_dtb_pre_load,
1923     .fields = (VMStateField[]) {
1924         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1925         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1926         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1927                                      fdt_size),
1928         VMSTATE_END_OF_LIST()
1929     },
1930 };
1931 
1932 static bool spapr_fwnmi_needed(void *opaque)
1933 {
1934     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1935 
1936     return spapr->fwnmi_machine_check_addr != -1;
1937 }
1938 
1939 static int spapr_fwnmi_pre_save(void *opaque)
1940 {
1941     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1942 
1943     /*
1944      * Check if machine check handling is in progress and print a
1945      * warning message.
1946      */
1947     if (spapr->fwnmi_machine_check_interlock != -1) {
1948         warn_report("A machine check is being handled during migration. The"
1949                 "handler may run and log hardware error on the destination");
1950     }
1951 
1952     return 0;
1953 }
1954 
1955 static const VMStateDescription vmstate_spapr_fwnmi = {
1956     .name = "spapr_fwnmi",
1957     .version_id = 1,
1958     .minimum_version_id = 1,
1959     .needed = spapr_fwnmi_needed,
1960     .pre_save = spapr_fwnmi_pre_save,
1961     .fields = (VMStateField[]) {
1962         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1963         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1964         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1965         VMSTATE_END_OF_LIST()
1966     },
1967 };
1968 
1969 static const VMStateDescription vmstate_spapr = {
1970     .name = "spapr",
1971     .version_id = 3,
1972     .minimum_version_id = 1,
1973     .pre_load = spapr_pre_load,
1974     .post_load = spapr_post_load,
1975     .pre_save = spapr_pre_save,
1976     .fields = (VMStateField[]) {
1977         /* used to be @next_irq */
1978         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1979 
1980         /* RTC offset */
1981         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1982 
1983         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1984         VMSTATE_END_OF_LIST()
1985     },
1986     .subsections = (const VMStateDescription*[]) {
1987         &vmstate_spapr_ov5_cas,
1988         &vmstate_spapr_patb_entry,
1989         &vmstate_spapr_pending_events,
1990         &vmstate_spapr_cap_htm,
1991         &vmstate_spapr_cap_vsx,
1992         &vmstate_spapr_cap_dfp,
1993         &vmstate_spapr_cap_cfpc,
1994         &vmstate_spapr_cap_sbbc,
1995         &vmstate_spapr_cap_ibs,
1996         &vmstate_spapr_cap_hpt_maxpagesize,
1997         &vmstate_spapr_irq_map,
1998         &vmstate_spapr_cap_nested_kvm_hv,
1999         &vmstate_spapr_dtb,
2000         &vmstate_spapr_cap_large_decr,
2001         &vmstate_spapr_cap_ccf_assist,
2002         &vmstate_spapr_cap_fwnmi,
2003         &vmstate_spapr_fwnmi,
2004         NULL
2005     }
2006 };
2007 
2008 static int htab_save_setup(QEMUFile *f, void *opaque)
2009 {
2010     SpaprMachineState *spapr = opaque;
2011 
2012     /* "Iteration" header */
2013     if (!spapr->htab_shift) {
2014         qemu_put_be32(f, -1);
2015     } else {
2016         qemu_put_be32(f, spapr->htab_shift);
2017     }
2018 
2019     if (spapr->htab) {
2020         spapr->htab_save_index = 0;
2021         spapr->htab_first_pass = true;
2022     } else {
2023         if (spapr->htab_shift) {
2024             assert(kvm_enabled());
2025         }
2026     }
2027 
2028 
2029     return 0;
2030 }
2031 
2032 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2033                             int chunkstart, int n_valid, int n_invalid)
2034 {
2035     qemu_put_be32(f, chunkstart);
2036     qemu_put_be16(f, n_valid);
2037     qemu_put_be16(f, n_invalid);
2038     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2039                     HASH_PTE_SIZE_64 * n_valid);
2040 }
2041 
2042 static void htab_save_end_marker(QEMUFile *f)
2043 {
2044     qemu_put_be32(f, 0);
2045     qemu_put_be16(f, 0);
2046     qemu_put_be16(f, 0);
2047 }
2048 
2049 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2050                                  int64_t max_ns)
2051 {
2052     bool has_timeout = max_ns != -1;
2053     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2054     int index = spapr->htab_save_index;
2055     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2056 
2057     assert(spapr->htab_first_pass);
2058 
2059     do {
2060         int chunkstart;
2061 
2062         /* Consume invalid HPTEs */
2063         while ((index < htabslots)
2064                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2065             CLEAN_HPTE(HPTE(spapr->htab, index));
2066             index++;
2067         }
2068 
2069         /* Consume valid HPTEs */
2070         chunkstart = index;
2071         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2072                && HPTE_VALID(HPTE(spapr->htab, index))) {
2073             CLEAN_HPTE(HPTE(spapr->htab, index));
2074             index++;
2075         }
2076 
2077         if (index > chunkstart) {
2078             int n_valid = index - chunkstart;
2079 
2080             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2081 
2082             if (has_timeout &&
2083                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2084                 break;
2085             }
2086         }
2087     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2088 
2089     if (index >= htabslots) {
2090         assert(index == htabslots);
2091         index = 0;
2092         spapr->htab_first_pass = false;
2093     }
2094     spapr->htab_save_index = index;
2095 }
2096 
2097 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2098                                 int64_t max_ns)
2099 {
2100     bool final = max_ns < 0;
2101     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2102     int examined = 0, sent = 0;
2103     int index = spapr->htab_save_index;
2104     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2105 
2106     assert(!spapr->htab_first_pass);
2107 
2108     do {
2109         int chunkstart, invalidstart;
2110 
2111         /* Consume non-dirty HPTEs */
2112         while ((index < htabslots)
2113                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2114             index++;
2115             examined++;
2116         }
2117 
2118         chunkstart = index;
2119         /* Consume valid dirty HPTEs */
2120         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2121                && HPTE_DIRTY(HPTE(spapr->htab, index))
2122                && HPTE_VALID(HPTE(spapr->htab, index))) {
2123             CLEAN_HPTE(HPTE(spapr->htab, index));
2124             index++;
2125             examined++;
2126         }
2127 
2128         invalidstart = index;
2129         /* Consume invalid dirty HPTEs */
2130         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2131                && HPTE_DIRTY(HPTE(spapr->htab, index))
2132                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2133             CLEAN_HPTE(HPTE(spapr->htab, index));
2134             index++;
2135             examined++;
2136         }
2137 
2138         if (index > chunkstart) {
2139             int n_valid = invalidstart - chunkstart;
2140             int n_invalid = index - invalidstart;
2141 
2142             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2143             sent += index - chunkstart;
2144 
2145             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2146                 break;
2147             }
2148         }
2149 
2150         if (examined >= htabslots) {
2151             break;
2152         }
2153 
2154         if (index >= htabslots) {
2155             assert(index == htabslots);
2156             index = 0;
2157         }
2158     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2159 
2160     if (index >= htabslots) {
2161         assert(index == htabslots);
2162         index = 0;
2163     }
2164 
2165     spapr->htab_save_index = index;
2166 
2167     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2168 }
2169 
2170 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2171 #define MAX_KVM_BUF_SIZE    2048
2172 
2173 static int htab_save_iterate(QEMUFile *f, void *opaque)
2174 {
2175     SpaprMachineState *spapr = opaque;
2176     int fd;
2177     int rc = 0;
2178 
2179     /* Iteration header */
2180     if (!spapr->htab_shift) {
2181         qemu_put_be32(f, -1);
2182         return 1;
2183     } else {
2184         qemu_put_be32(f, 0);
2185     }
2186 
2187     if (!spapr->htab) {
2188         assert(kvm_enabled());
2189 
2190         fd = get_htab_fd(spapr);
2191         if (fd < 0) {
2192             return fd;
2193         }
2194 
2195         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2196         if (rc < 0) {
2197             return rc;
2198         }
2199     } else  if (spapr->htab_first_pass) {
2200         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2201     } else {
2202         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2203     }
2204 
2205     htab_save_end_marker(f);
2206 
2207     return rc;
2208 }
2209 
2210 static int htab_save_complete(QEMUFile *f, void *opaque)
2211 {
2212     SpaprMachineState *spapr = opaque;
2213     int fd;
2214 
2215     /* Iteration header */
2216     if (!spapr->htab_shift) {
2217         qemu_put_be32(f, -1);
2218         return 0;
2219     } else {
2220         qemu_put_be32(f, 0);
2221     }
2222 
2223     if (!spapr->htab) {
2224         int rc;
2225 
2226         assert(kvm_enabled());
2227 
2228         fd = get_htab_fd(spapr);
2229         if (fd < 0) {
2230             return fd;
2231         }
2232 
2233         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2234         if (rc < 0) {
2235             return rc;
2236         }
2237     } else {
2238         if (spapr->htab_first_pass) {
2239             htab_save_first_pass(f, spapr, -1);
2240         }
2241         htab_save_later_pass(f, spapr, -1);
2242     }
2243 
2244     /* End marker */
2245     htab_save_end_marker(f);
2246 
2247     return 0;
2248 }
2249 
2250 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2251 {
2252     SpaprMachineState *spapr = opaque;
2253     uint32_t section_hdr;
2254     int fd = -1;
2255     Error *local_err = NULL;
2256 
2257     if (version_id < 1 || version_id > 1) {
2258         error_report("htab_load() bad version");
2259         return -EINVAL;
2260     }
2261 
2262     section_hdr = qemu_get_be32(f);
2263 
2264     if (section_hdr == -1) {
2265         spapr_free_hpt(spapr);
2266         return 0;
2267     }
2268 
2269     if (section_hdr) {
2270         int ret;
2271 
2272         /* First section gives the htab size */
2273         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2274         if (ret < 0) {
2275             error_report_err(local_err);
2276             return ret;
2277         }
2278         return 0;
2279     }
2280 
2281     if (!spapr->htab) {
2282         assert(kvm_enabled());
2283 
2284         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2285         if (fd < 0) {
2286             error_report_err(local_err);
2287             return fd;
2288         }
2289     }
2290 
2291     while (true) {
2292         uint32_t index;
2293         uint16_t n_valid, n_invalid;
2294 
2295         index = qemu_get_be32(f);
2296         n_valid = qemu_get_be16(f);
2297         n_invalid = qemu_get_be16(f);
2298 
2299         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2300             /* End of Stream */
2301             break;
2302         }
2303 
2304         if ((index + n_valid + n_invalid) >
2305             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2306             /* Bad index in stream */
2307             error_report(
2308                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2309                 index, n_valid, n_invalid, spapr->htab_shift);
2310             return -EINVAL;
2311         }
2312 
2313         if (spapr->htab) {
2314             if (n_valid) {
2315                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2316                                 HASH_PTE_SIZE_64 * n_valid);
2317             }
2318             if (n_invalid) {
2319                 memset(HPTE(spapr->htab, index + n_valid), 0,
2320                        HASH_PTE_SIZE_64 * n_invalid);
2321             }
2322         } else {
2323             int rc;
2324 
2325             assert(fd >= 0);
2326 
2327             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2328                                         &local_err);
2329             if (rc < 0) {
2330                 error_report_err(local_err);
2331                 return rc;
2332             }
2333         }
2334     }
2335 
2336     if (!spapr->htab) {
2337         assert(fd >= 0);
2338         close(fd);
2339     }
2340 
2341     return 0;
2342 }
2343 
2344 static void htab_save_cleanup(void *opaque)
2345 {
2346     SpaprMachineState *spapr = opaque;
2347 
2348     close_htab_fd(spapr);
2349 }
2350 
2351 static SaveVMHandlers savevm_htab_handlers = {
2352     .save_setup = htab_save_setup,
2353     .save_live_iterate = htab_save_iterate,
2354     .save_live_complete_precopy = htab_save_complete,
2355     .save_cleanup = htab_save_cleanup,
2356     .load_state = htab_load,
2357 };
2358 
2359 static void spapr_boot_set(void *opaque, const char *boot_device,
2360                            Error **errp)
2361 {
2362     MachineState *machine = MACHINE(opaque);
2363     machine->boot_order = g_strdup(boot_device);
2364 }
2365 
2366 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2367 {
2368     MachineState *machine = MACHINE(spapr);
2369     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2370     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2371     int i;
2372 
2373     for (i = 0; i < nr_lmbs; i++) {
2374         uint64_t addr;
2375 
2376         addr = i * lmb_size + machine->device_memory->base;
2377         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2378                                addr / lmb_size);
2379     }
2380 }
2381 
2382 /*
2383  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2384  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2385  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2386  */
2387 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2388 {
2389     int i;
2390 
2391     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2392         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2393                    " is not aligned to %" PRIu64 " MiB",
2394                    machine->ram_size,
2395                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2396         return;
2397     }
2398 
2399     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2400         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2401                    " is not aligned to %" PRIu64 " MiB",
2402                    machine->ram_size,
2403                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2404         return;
2405     }
2406 
2407     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2408         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2409             error_setg(errp,
2410                        "Node %d memory size 0x%" PRIx64
2411                        " is not aligned to %" PRIu64 " MiB",
2412                        i, machine->numa_state->nodes[i].node_mem,
2413                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2414             return;
2415         }
2416     }
2417 }
2418 
2419 /* find cpu slot in machine->possible_cpus by core_id */
2420 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2421 {
2422     int index = id / ms->smp.threads;
2423 
2424     if (index >= ms->possible_cpus->len) {
2425         return NULL;
2426     }
2427     if (idx) {
2428         *idx = index;
2429     }
2430     return &ms->possible_cpus->cpus[index];
2431 }
2432 
2433 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2434 {
2435     MachineState *ms = MACHINE(spapr);
2436     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2437     Error *local_err = NULL;
2438     bool vsmt_user = !!spapr->vsmt;
2439     int kvm_smt = kvmppc_smt_threads();
2440     int ret;
2441     unsigned int smp_threads = ms->smp.threads;
2442 
2443     if (!kvm_enabled() && (smp_threads > 1)) {
2444         error_setg(errp, "TCG cannot support more than 1 thread/core "
2445                    "on a pseries machine");
2446         return;
2447     }
2448     if (!is_power_of_2(smp_threads)) {
2449         error_setg(errp, "Cannot support %d threads/core on a pseries "
2450                    "machine because it must be a power of 2", smp_threads);
2451         return;
2452     }
2453 
2454     /* Detemine the VSMT mode to use: */
2455     if (vsmt_user) {
2456         if (spapr->vsmt < smp_threads) {
2457             error_setg(errp, "Cannot support VSMT mode %d"
2458                        " because it must be >= threads/core (%d)",
2459                        spapr->vsmt, smp_threads);
2460             return;
2461         }
2462         /* In this case, spapr->vsmt has been set by the command line */
2463     } else if (!smc->smp_threads_vsmt) {
2464         /*
2465          * Default VSMT value is tricky, because we need it to be as
2466          * consistent as possible (for migration), but this requires
2467          * changing it for at least some existing cases.  We pick 8 as
2468          * the value that we'd get with KVM on POWER8, the
2469          * overwhelmingly common case in production systems.
2470          */
2471         spapr->vsmt = MAX(8, smp_threads);
2472     } else {
2473         spapr->vsmt = smp_threads;
2474     }
2475 
2476     /* KVM: If necessary, set the SMT mode: */
2477     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2478         ret = kvmppc_set_smt_threads(spapr->vsmt);
2479         if (ret) {
2480             /* Looks like KVM isn't able to change VSMT mode */
2481             error_setg(&local_err,
2482                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2483                        spapr->vsmt, ret);
2484             /* We can live with that if the default one is big enough
2485              * for the number of threads, and a submultiple of the one
2486              * we want.  In this case we'll waste some vcpu ids, but
2487              * behaviour will be correct */
2488             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2489                 warn_report_err(local_err);
2490             } else {
2491                 if (!vsmt_user) {
2492                     error_append_hint(&local_err,
2493                                       "On PPC, a VM with %d threads/core"
2494                                       " on a host with %d threads/core"
2495                                       " requires the use of VSMT mode %d.\n",
2496                                       smp_threads, kvm_smt, spapr->vsmt);
2497                 }
2498                 kvmppc_error_append_smt_possible_hint(&local_err);
2499                 error_propagate(errp, local_err);
2500             }
2501         }
2502     }
2503     /* else TCG: nothing to do currently */
2504 }
2505 
2506 static void spapr_init_cpus(SpaprMachineState *spapr)
2507 {
2508     MachineState *machine = MACHINE(spapr);
2509     MachineClass *mc = MACHINE_GET_CLASS(machine);
2510     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2511     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2512     const CPUArchIdList *possible_cpus;
2513     unsigned int smp_cpus = machine->smp.cpus;
2514     unsigned int smp_threads = machine->smp.threads;
2515     unsigned int max_cpus = machine->smp.max_cpus;
2516     int boot_cores_nr = smp_cpus / smp_threads;
2517     int i;
2518 
2519     possible_cpus = mc->possible_cpu_arch_ids(machine);
2520     if (mc->has_hotpluggable_cpus) {
2521         if (smp_cpus % smp_threads) {
2522             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2523                          smp_cpus, smp_threads);
2524             exit(1);
2525         }
2526         if (max_cpus % smp_threads) {
2527             error_report("max_cpus (%u) must be multiple of threads (%u)",
2528                          max_cpus, smp_threads);
2529             exit(1);
2530         }
2531     } else {
2532         if (max_cpus != smp_cpus) {
2533             error_report("This machine version does not support CPU hotplug");
2534             exit(1);
2535         }
2536         boot_cores_nr = possible_cpus->len;
2537     }
2538 
2539     if (smc->pre_2_10_has_unused_icps) {
2540         int i;
2541 
2542         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2543             /* Dummy entries get deregistered when real ICPState objects
2544              * are registered during CPU core hotplug.
2545              */
2546             pre_2_10_vmstate_register_dummy_icp(i);
2547         }
2548     }
2549 
2550     for (i = 0; i < possible_cpus->len; i++) {
2551         int core_id = i * smp_threads;
2552 
2553         if (mc->has_hotpluggable_cpus) {
2554             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2555                                    spapr_vcpu_id(spapr, core_id));
2556         }
2557 
2558         if (i < boot_cores_nr) {
2559             Object *core  = object_new(type);
2560             int nr_threads = smp_threads;
2561 
2562             /* Handle the partially filled core for older machine types */
2563             if ((i + 1) * smp_threads >= smp_cpus) {
2564                 nr_threads = smp_cpus - i * smp_threads;
2565             }
2566 
2567             object_property_set_int(core, "nr-threads", nr_threads,
2568                                     &error_fatal);
2569             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2570                                     &error_fatal);
2571             qdev_realize(DEVICE(core), NULL, &error_fatal);
2572 
2573             object_unref(core);
2574         }
2575     }
2576 }
2577 
2578 static PCIHostState *spapr_create_default_phb(void)
2579 {
2580     DeviceState *dev;
2581 
2582     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2583     qdev_prop_set_uint32(dev, "index", 0);
2584     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2585 
2586     return PCI_HOST_BRIDGE(dev);
2587 }
2588 
2589 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2590 {
2591     MachineState *machine = MACHINE(spapr);
2592     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2593     hwaddr rma_size = machine->ram_size;
2594     hwaddr node0_size = spapr_node0_size(machine);
2595 
2596     /* RMA has to fit in the first NUMA node */
2597     rma_size = MIN(rma_size, node0_size);
2598 
2599     /*
2600      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2601      * never exceed that
2602      */
2603     rma_size = MIN(rma_size, 1 * TiB);
2604 
2605     /*
2606      * Clamp the RMA size based on machine type.  This is for
2607      * migration compatibility with older qemu versions, which limited
2608      * the RMA size for complicated and mostly bad reasons.
2609      */
2610     if (smc->rma_limit) {
2611         rma_size = MIN(rma_size, smc->rma_limit);
2612     }
2613 
2614     if (rma_size < MIN_RMA_SLOF) {
2615         error_setg(errp,
2616                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2617                    "ldMiB guest RMA (Real Mode Area memory)",
2618                    MIN_RMA_SLOF / MiB);
2619         return 0;
2620     }
2621 
2622     return rma_size;
2623 }
2624 
2625 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2626 {
2627     MachineState *machine = MACHINE(spapr);
2628     int i;
2629 
2630     for (i = 0; i < machine->ram_slots; i++) {
2631         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2632     }
2633 }
2634 
2635 /* pSeries LPAR / sPAPR hardware init */
2636 static void spapr_machine_init(MachineState *machine)
2637 {
2638     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2639     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2640     MachineClass *mc = MACHINE_GET_CLASS(machine);
2641     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2642     const char *kernel_filename = machine->kernel_filename;
2643     const char *initrd_filename = machine->initrd_filename;
2644     PCIHostState *phb;
2645     int i;
2646     MemoryRegion *sysmem = get_system_memory();
2647     long load_limit, fw_size;
2648     char *filename;
2649     Error *resize_hpt_err = NULL;
2650 
2651     /*
2652      * if Secure VM (PEF) support is configured, then initialize it
2653      */
2654     pef_kvm_init(machine->cgs, &error_fatal);
2655 
2656     msi_nonbroken = true;
2657 
2658     QLIST_INIT(&spapr->phbs);
2659     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2660 
2661     /* Determine capabilities to run with */
2662     spapr_caps_init(spapr);
2663 
2664     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2665     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2666         /*
2667          * If the user explicitly requested a mode we should either
2668          * supply it, or fail completely (which we do below).  But if
2669          * it's not set explicitly, we reset our mode to something
2670          * that works
2671          */
2672         if (resize_hpt_err) {
2673             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2674             error_free(resize_hpt_err);
2675             resize_hpt_err = NULL;
2676         } else {
2677             spapr->resize_hpt = smc->resize_hpt_default;
2678         }
2679     }
2680 
2681     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2682 
2683     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2684         /*
2685          * User requested HPT resize, but this host can't supply it.  Bail out
2686          */
2687         error_report_err(resize_hpt_err);
2688         exit(1);
2689     }
2690     error_free(resize_hpt_err);
2691 
2692     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2693 
2694     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2695     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2696 
2697     /*
2698      * VSMT must be set in order to be able to compute VCPU ids, ie to
2699      * call spapr_max_server_number() or spapr_vcpu_id().
2700      */
2701     spapr_set_vsmt_mode(spapr, &error_fatal);
2702 
2703     /* Set up Interrupt Controller before we create the VCPUs */
2704     spapr_irq_init(spapr, &error_fatal);
2705 
2706     /* Set up containers for ibm,client-architecture-support negotiated options
2707      */
2708     spapr->ov5 = spapr_ovec_new();
2709     spapr->ov5_cas = spapr_ovec_new();
2710 
2711     if (smc->dr_lmb_enabled) {
2712         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2713         spapr_validate_node_memory(machine, &error_fatal);
2714     }
2715 
2716     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2717 
2718     /* advertise support for dedicated HP event source to guests */
2719     if (spapr->use_hotplug_event_source) {
2720         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2721     }
2722 
2723     /* advertise support for HPT resizing */
2724     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2725         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2726     }
2727 
2728     /* advertise support for ibm,dyamic-memory-v2 */
2729     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2730 
2731     /* advertise XIVE on POWER9 machines */
2732     if (spapr->irq->xive) {
2733         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2734     }
2735 
2736     /* init CPUs */
2737     spapr_init_cpus(spapr);
2738 
2739     /*
2740      * check we don't have a memory-less/cpu-less NUMA node
2741      * Firmware relies on the existing memory/cpu topology to provide the
2742      * NUMA topology to the kernel.
2743      * And the linux kernel needs to know the NUMA topology at start
2744      * to be able to hotplug CPUs later.
2745      */
2746     if (machine->numa_state->num_nodes) {
2747         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2748             /* check for memory-less node */
2749             if (machine->numa_state->nodes[i].node_mem == 0) {
2750                 CPUState *cs;
2751                 int found = 0;
2752                 /* check for cpu-less node */
2753                 CPU_FOREACH(cs) {
2754                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2755                     if (cpu->node_id == i) {
2756                         found = 1;
2757                         break;
2758                     }
2759                 }
2760                 /* memory-less and cpu-less node */
2761                 if (!found) {
2762                     error_report(
2763                        "Memory-less/cpu-less nodes are not supported (node %d)",
2764                                  i);
2765                     exit(1);
2766                 }
2767             }
2768         }
2769 
2770     }
2771 
2772     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2773 
2774     /* Init numa_assoc_array */
2775     spapr_numa_associativity_init(spapr, machine);
2776 
2777     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2778         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2779                               spapr->max_compat_pvr)) {
2780         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2781         /* KVM and TCG always allow GTSE with radix... */
2782         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2783     }
2784     /* ... but not with hash (currently). */
2785 
2786     if (kvm_enabled()) {
2787         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2788         kvmppc_enable_logical_ci_hcalls();
2789         kvmppc_enable_set_mode_hcall();
2790 
2791         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2792         kvmppc_enable_clear_ref_mod_hcalls();
2793 
2794         /* Enable H_PAGE_INIT */
2795         kvmppc_enable_h_page_init();
2796     }
2797 
2798     /* map RAM */
2799     memory_region_add_subregion(sysmem, 0, machine->ram);
2800 
2801     /* always allocate the device memory information */
2802     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2803 
2804     /* initialize hotplug memory address space */
2805     if (machine->ram_size < machine->maxram_size) {
2806         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2807         /*
2808          * Limit the number of hotpluggable memory slots to half the number
2809          * slots that KVM supports, leaving the other half for PCI and other
2810          * devices. However ensure that number of slots doesn't drop below 32.
2811          */
2812         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2813                            SPAPR_MAX_RAM_SLOTS;
2814 
2815         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2816             max_memslots = SPAPR_MAX_RAM_SLOTS;
2817         }
2818         if (machine->ram_slots > max_memslots) {
2819             error_report("Specified number of memory slots %"
2820                          PRIu64" exceeds max supported %d",
2821                          machine->ram_slots, max_memslots);
2822             exit(1);
2823         }
2824 
2825         machine->device_memory->base = ROUND_UP(machine->ram_size,
2826                                                 SPAPR_DEVICE_MEM_ALIGN);
2827         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2828                            "device-memory", device_mem_size);
2829         memory_region_add_subregion(sysmem, machine->device_memory->base,
2830                                     &machine->device_memory->mr);
2831     }
2832 
2833     if (smc->dr_lmb_enabled) {
2834         spapr_create_lmb_dr_connectors(spapr);
2835     }
2836 
2837     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2838         /* Create the error string for live migration blocker */
2839         error_setg(&spapr->fwnmi_migration_blocker,
2840             "A machine check is being handled during migration. The handler"
2841             "may run and log hardware error on the destination");
2842     }
2843 
2844     if (mc->nvdimm_supported) {
2845         spapr_create_nvdimm_dr_connectors(spapr);
2846     }
2847 
2848     /* Set up RTAS event infrastructure */
2849     spapr_events_init(spapr);
2850 
2851     /* Set up the RTC RTAS interfaces */
2852     spapr_rtc_create(spapr);
2853 
2854     /* Set up VIO bus */
2855     spapr->vio_bus = spapr_vio_bus_init();
2856 
2857     for (i = 0; serial_hd(i); i++) {
2858         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2859     }
2860 
2861     /* We always have at least the nvram device on VIO */
2862     spapr_create_nvram(spapr);
2863 
2864     /*
2865      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2866      * connectors (described in root DT node's "ibm,drc-types" property)
2867      * are pre-initialized here. additional child connectors (such as
2868      * connectors for a PHBs PCI slots) are added as needed during their
2869      * parent's realization.
2870      */
2871     if (smc->dr_phb_enabled) {
2872         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2873             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2874         }
2875     }
2876 
2877     /* Set up PCI */
2878     spapr_pci_rtas_init();
2879 
2880     phb = spapr_create_default_phb();
2881 
2882     for (i = 0; i < nb_nics; i++) {
2883         NICInfo *nd = &nd_table[i];
2884 
2885         if (!nd->model) {
2886             nd->model = g_strdup("spapr-vlan");
2887         }
2888 
2889         if (g_str_equal(nd->model, "spapr-vlan") ||
2890             g_str_equal(nd->model, "ibmveth")) {
2891             spapr_vlan_create(spapr->vio_bus, nd);
2892         } else {
2893             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2894         }
2895     }
2896 
2897     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2898         spapr_vscsi_create(spapr->vio_bus);
2899     }
2900 
2901     /* Graphics */
2902     if (spapr_vga_init(phb->bus, &error_fatal)) {
2903         spapr->has_graphics = true;
2904         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2905     }
2906 
2907     if (machine->usb) {
2908         if (smc->use_ohci_by_default) {
2909             pci_create_simple(phb->bus, -1, "pci-ohci");
2910         } else {
2911             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2912         }
2913 
2914         if (spapr->has_graphics) {
2915             USBBus *usb_bus = usb_bus_find(-1);
2916 
2917             usb_create_simple(usb_bus, "usb-kbd");
2918             usb_create_simple(usb_bus, "usb-mouse");
2919         }
2920     }
2921 
2922     if (kernel_filename) {
2923         spapr->kernel_size = load_elf(kernel_filename, NULL,
2924                                       translate_kernel_address, spapr,
2925                                       NULL, NULL, NULL, NULL, 1,
2926                                       PPC_ELF_MACHINE, 0, 0);
2927         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2928             spapr->kernel_size = load_elf(kernel_filename, NULL,
2929                                           translate_kernel_address, spapr,
2930                                           NULL, NULL, NULL, NULL, 0,
2931                                           PPC_ELF_MACHINE, 0, 0);
2932             spapr->kernel_le = spapr->kernel_size > 0;
2933         }
2934         if (spapr->kernel_size < 0) {
2935             error_report("error loading %s: %s", kernel_filename,
2936                          load_elf_strerror(spapr->kernel_size));
2937             exit(1);
2938         }
2939 
2940         /* load initrd */
2941         if (initrd_filename) {
2942             /* Try to locate the initrd in the gap between the kernel
2943              * and the firmware. Add a bit of space just in case
2944              */
2945             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2946                                   + 0x1ffff) & ~0xffff;
2947             spapr->initrd_size = load_image_targphys(initrd_filename,
2948                                                      spapr->initrd_base,
2949                                                      load_limit
2950                                                      - spapr->initrd_base);
2951             if (spapr->initrd_size < 0) {
2952                 error_report("could not load initial ram disk '%s'",
2953                              initrd_filename);
2954                 exit(1);
2955             }
2956         }
2957     }
2958 
2959     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2960     if (!filename) {
2961         error_report("Could not find LPAR firmware '%s'", bios_name);
2962         exit(1);
2963     }
2964     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2965     if (fw_size <= 0) {
2966         error_report("Could not load LPAR firmware '%s'", filename);
2967         exit(1);
2968     }
2969     g_free(filename);
2970 
2971     /* FIXME: Should register things through the MachineState's qdev
2972      * interface, this is a legacy from the sPAPREnvironment structure
2973      * which predated MachineState but had a similar function */
2974     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2975     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2976                          &savevm_htab_handlers, spapr);
2977 
2978     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2979 
2980     qemu_register_boot_set(spapr_boot_set, spapr);
2981 
2982     /*
2983      * Nothing needs to be done to resume a suspended guest because
2984      * suspending does not change the machine state, so no need for
2985      * a ->wakeup method.
2986      */
2987     qemu_register_wakeup_support();
2988 
2989     if (kvm_enabled()) {
2990         /* to stop and start vmclock */
2991         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2992                                          &spapr->tb);
2993 
2994         kvmppc_spapr_enable_inkernel_multitce();
2995     }
2996 
2997     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
2998 }
2999 
3000 #define DEFAULT_KVM_TYPE "auto"
3001 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3002 {
3003     /*
3004      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3005      * accomodate the 'HV' and 'PV' formats that exists in the
3006      * wild. The 'auto' mode is being introduced already as
3007      * lower-case, thus we don't need to bother checking for
3008      * "AUTO".
3009      */
3010     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3011         return 0;
3012     }
3013 
3014     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3015         return 1;
3016     }
3017 
3018     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3019         return 2;
3020     }
3021 
3022     error_report("Unknown kvm-type specified '%s'", vm_type);
3023     exit(1);
3024 }
3025 
3026 /*
3027  * Implementation of an interface to adjust firmware path
3028  * for the bootindex property handling.
3029  */
3030 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3031                                    DeviceState *dev)
3032 {
3033 #define CAST(type, obj, name) \
3034     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3035     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3036     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3037     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3038     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3039 
3040     if (d) {
3041         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3042         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3043         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3044 
3045         if (spapr) {
3046             /*
3047              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3048              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3049              * 0x8000 | (target << 8) | (bus << 5) | lun
3050              * (see the "Logical unit addressing format" table in SAM5)
3051              */
3052             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3053             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3054                                    (uint64_t)id << 48);
3055         } else if (virtio) {
3056             /*
3057              * We use SRP luns of the form 01000000 | (target << 8) | lun
3058              * in the top 32 bits of the 64-bit LUN
3059              * Note: the quote above is from SLOF and it is wrong,
3060              * the actual binding is:
3061              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3062              */
3063             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3064             if (d->lun >= 256) {
3065                 /* Use the LUN "flat space addressing method" */
3066                 id |= 0x4000;
3067             }
3068             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3069                                    (uint64_t)id << 32);
3070         } else if (usb) {
3071             /*
3072              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3073              * in the top 32 bits of the 64-bit LUN
3074              */
3075             unsigned usb_port = atoi(usb->port->path);
3076             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3077             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3078                                    (uint64_t)id << 32);
3079         }
3080     }
3081 
3082     /*
3083      * SLOF probes the USB devices, and if it recognizes that the device is a
3084      * storage device, it changes its name to "storage" instead of "usb-host",
3085      * and additionally adds a child node for the SCSI LUN, so the correct
3086      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3087      */
3088     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3089         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3090         if (usb_host_dev_is_scsi_storage(usbdev)) {
3091             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3092         }
3093     }
3094 
3095     if (phb) {
3096         /* Replace "pci" with "pci@800000020000000" */
3097         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3098     }
3099 
3100     if (vsc) {
3101         /* Same logic as virtio above */
3102         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3103         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3104     }
3105 
3106     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3107         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3108         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3109         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3110     }
3111 
3112     if (pcidev) {
3113         return spapr_pci_fw_dev_name(pcidev);
3114     }
3115 
3116     return NULL;
3117 }
3118 
3119 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3120 {
3121     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3122 
3123     return g_strdup(spapr->kvm_type);
3124 }
3125 
3126 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3127 {
3128     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3129 
3130     g_free(spapr->kvm_type);
3131     spapr->kvm_type = g_strdup(value);
3132 }
3133 
3134 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3135 {
3136     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3137 
3138     return spapr->use_hotplug_event_source;
3139 }
3140 
3141 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3142                                             Error **errp)
3143 {
3144     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3145 
3146     spapr->use_hotplug_event_source = value;
3147 }
3148 
3149 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3150 {
3151     return true;
3152 }
3153 
3154 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3155 {
3156     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3157 
3158     switch (spapr->resize_hpt) {
3159     case SPAPR_RESIZE_HPT_DEFAULT:
3160         return g_strdup("default");
3161     case SPAPR_RESIZE_HPT_DISABLED:
3162         return g_strdup("disabled");
3163     case SPAPR_RESIZE_HPT_ENABLED:
3164         return g_strdup("enabled");
3165     case SPAPR_RESIZE_HPT_REQUIRED:
3166         return g_strdup("required");
3167     }
3168     g_assert_not_reached();
3169 }
3170 
3171 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3172 {
3173     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3174 
3175     if (strcmp(value, "default") == 0) {
3176         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3177     } else if (strcmp(value, "disabled") == 0) {
3178         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3179     } else if (strcmp(value, "enabled") == 0) {
3180         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3181     } else if (strcmp(value, "required") == 0) {
3182         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3183     } else {
3184         error_setg(errp, "Bad value for \"resize-hpt\" property");
3185     }
3186 }
3187 
3188 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3189 {
3190     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3191 
3192     if (spapr->irq == &spapr_irq_xics_legacy) {
3193         return g_strdup("legacy");
3194     } else if (spapr->irq == &spapr_irq_xics) {
3195         return g_strdup("xics");
3196     } else if (spapr->irq == &spapr_irq_xive) {
3197         return g_strdup("xive");
3198     } else if (spapr->irq == &spapr_irq_dual) {
3199         return g_strdup("dual");
3200     }
3201     g_assert_not_reached();
3202 }
3203 
3204 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3205 {
3206     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3207 
3208     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3209         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3210         return;
3211     }
3212 
3213     /* The legacy IRQ backend can not be set */
3214     if (strcmp(value, "xics") == 0) {
3215         spapr->irq = &spapr_irq_xics;
3216     } else if (strcmp(value, "xive") == 0) {
3217         spapr->irq = &spapr_irq_xive;
3218     } else if (strcmp(value, "dual") == 0) {
3219         spapr->irq = &spapr_irq_dual;
3220     } else {
3221         error_setg(errp, "Bad value for \"ic-mode\" property");
3222     }
3223 }
3224 
3225 static char *spapr_get_host_model(Object *obj, Error **errp)
3226 {
3227     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3228 
3229     return g_strdup(spapr->host_model);
3230 }
3231 
3232 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3233 {
3234     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3235 
3236     g_free(spapr->host_model);
3237     spapr->host_model = g_strdup(value);
3238 }
3239 
3240 static char *spapr_get_host_serial(Object *obj, Error **errp)
3241 {
3242     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3243 
3244     return g_strdup(spapr->host_serial);
3245 }
3246 
3247 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3248 {
3249     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3250 
3251     g_free(spapr->host_serial);
3252     spapr->host_serial = g_strdup(value);
3253 }
3254 
3255 static void spapr_instance_init(Object *obj)
3256 {
3257     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3258     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3259     MachineState *ms = MACHINE(spapr);
3260     MachineClass *mc = MACHINE_GET_CLASS(ms);
3261 
3262     /*
3263      * NVDIMM support went live in 5.1 without considering that, in
3264      * other archs, the user needs to enable NVDIMM support with the
3265      * 'nvdimm' machine option and the default behavior is NVDIMM
3266      * support disabled. It is too late to roll back to the standard
3267      * behavior without breaking 5.1 guests.
3268      */
3269     if (mc->nvdimm_supported) {
3270         ms->nvdimms_state->is_enabled = true;
3271     }
3272 
3273     spapr->htab_fd = -1;
3274     spapr->use_hotplug_event_source = true;
3275     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3276     object_property_add_str(obj, "kvm-type",
3277                             spapr_get_kvm_type, spapr_set_kvm_type);
3278     object_property_set_description(obj, "kvm-type",
3279                                     "Specifies the KVM virtualization mode (auto,"
3280                                     " hv, pr). Defaults to 'auto'. This mode will use"
3281                                     " any available KVM module loaded in the host,"
3282                                     " where kvm_hv takes precedence if both kvm_hv and"
3283                                     " kvm_pr are loaded.");
3284     object_property_add_bool(obj, "modern-hotplug-events",
3285                             spapr_get_modern_hotplug_events,
3286                             spapr_set_modern_hotplug_events);
3287     object_property_set_description(obj, "modern-hotplug-events",
3288                                     "Use dedicated hotplug event mechanism in"
3289                                     " place of standard EPOW events when possible"
3290                                     " (required for memory hot-unplug support)");
3291     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3292                             "Maximum permitted CPU compatibility mode");
3293 
3294     object_property_add_str(obj, "resize-hpt",
3295                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3296     object_property_set_description(obj, "resize-hpt",
3297                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3298     object_property_add_uint32_ptr(obj, "vsmt",
3299                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3300     object_property_set_description(obj, "vsmt",
3301                                     "Virtual SMT: KVM behaves as if this were"
3302                                     " the host's SMT mode");
3303 
3304     object_property_add_bool(obj, "vfio-no-msix-emulation",
3305                              spapr_get_msix_emulation, NULL);
3306 
3307     object_property_add_uint64_ptr(obj, "kernel-addr",
3308                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3309     object_property_set_description(obj, "kernel-addr",
3310                                     stringify(KERNEL_LOAD_ADDR)
3311                                     " for -kernel is the default");
3312     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3313     /* The machine class defines the default interrupt controller mode */
3314     spapr->irq = smc->irq;
3315     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3316                             spapr_set_ic_mode);
3317     object_property_set_description(obj, "ic-mode",
3318                  "Specifies the interrupt controller mode (xics, xive, dual)");
3319 
3320     object_property_add_str(obj, "host-model",
3321         spapr_get_host_model, spapr_set_host_model);
3322     object_property_set_description(obj, "host-model",
3323         "Host model to advertise in guest device tree");
3324     object_property_add_str(obj, "host-serial",
3325         spapr_get_host_serial, spapr_set_host_serial);
3326     object_property_set_description(obj, "host-serial",
3327         "Host serial number to advertise in guest device tree");
3328 }
3329 
3330 static void spapr_machine_finalizefn(Object *obj)
3331 {
3332     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3333 
3334     g_free(spapr->kvm_type);
3335 }
3336 
3337 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3338 {
3339     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3340     PowerPCCPU *cpu = POWERPC_CPU(cs);
3341     CPUPPCState *env = &cpu->env;
3342 
3343     cpu_synchronize_state(cs);
3344     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3345     if (spapr->fwnmi_system_reset_addr != -1) {
3346         uint64_t rtas_addr, addr;
3347 
3348         /* get rtas addr from fdt */
3349         rtas_addr = spapr_get_rtas_addr();
3350         if (!rtas_addr) {
3351             qemu_system_guest_panicked(NULL);
3352             return;
3353         }
3354 
3355         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3356         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3357         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3358         env->gpr[3] = addr;
3359     }
3360     ppc_cpu_do_system_reset(cs);
3361     if (spapr->fwnmi_system_reset_addr != -1) {
3362         env->nip = spapr->fwnmi_system_reset_addr;
3363     }
3364 }
3365 
3366 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3367 {
3368     CPUState *cs;
3369 
3370     CPU_FOREACH(cs) {
3371         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3372     }
3373 }
3374 
3375 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3376                           void *fdt, int *fdt_start_offset, Error **errp)
3377 {
3378     uint64_t addr;
3379     uint32_t node;
3380 
3381     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3382     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3383                                     &error_abort);
3384     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3385                                              SPAPR_MEMORY_BLOCK_SIZE);
3386     return 0;
3387 }
3388 
3389 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3390                            bool dedicated_hp_event_source)
3391 {
3392     SpaprDrc *drc;
3393     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3394     int i;
3395     uint64_t addr = addr_start;
3396     bool hotplugged = spapr_drc_hotplugged(dev);
3397 
3398     for (i = 0; i < nr_lmbs; i++) {
3399         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3400                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3401         g_assert(drc);
3402 
3403         /*
3404          * memory_device_get_free_addr() provided a range of free addresses
3405          * that doesn't overlap with any existing mapping at pre-plug. The
3406          * corresponding LMB DRCs are thus assumed to be all attachable.
3407          */
3408         spapr_drc_attach(drc, dev);
3409         if (!hotplugged) {
3410             spapr_drc_reset(drc);
3411         }
3412         addr += SPAPR_MEMORY_BLOCK_SIZE;
3413     }
3414     /* send hotplug notification to the
3415      * guest only in case of hotplugged memory
3416      */
3417     if (hotplugged) {
3418         if (dedicated_hp_event_source) {
3419             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3420                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3421             g_assert(drc);
3422             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3423                                                    nr_lmbs,
3424                                                    spapr_drc_index(drc));
3425         } else {
3426             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3427                                            nr_lmbs);
3428         }
3429     }
3430 }
3431 
3432 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3433 {
3434     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3435     PCDIMMDevice *dimm = PC_DIMM(dev);
3436     uint64_t size, addr;
3437     int64_t slot;
3438     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3439 
3440     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3441 
3442     pc_dimm_plug(dimm, MACHINE(ms));
3443 
3444     if (!is_nvdimm) {
3445         addr = object_property_get_uint(OBJECT(dimm),
3446                                         PC_DIMM_ADDR_PROP, &error_abort);
3447         spapr_add_lmbs(dev, addr, size,
3448                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3449     } else {
3450         slot = object_property_get_int(OBJECT(dimm),
3451                                        PC_DIMM_SLOT_PROP, &error_abort);
3452         /* We should have valid slot number at this point */
3453         g_assert(slot >= 0);
3454         spapr_add_nvdimm(dev, slot);
3455     }
3456 }
3457 
3458 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3459                                   Error **errp)
3460 {
3461     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3462     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3463     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3464     PCDIMMDevice *dimm = PC_DIMM(dev);
3465     Error *local_err = NULL;
3466     uint64_t size;
3467     Object *memdev;
3468     hwaddr pagesize;
3469 
3470     if (!smc->dr_lmb_enabled) {
3471         error_setg(errp, "Memory hotplug not supported for this machine");
3472         return;
3473     }
3474 
3475     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3476     if (local_err) {
3477         error_propagate(errp, local_err);
3478         return;
3479     }
3480 
3481     if (is_nvdimm) {
3482         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3483             return;
3484         }
3485     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3486         error_setg(errp, "Hotplugged memory size must be a multiple of "
3487                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3488         return;
3489     }
3490 
3491     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3492                                       &error_abort);
3493     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3494     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3495         return;
3496     }
3497 
3498     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3499 }
3500 
3501 struct SpaprDimmState {
3502     PCDIMMDevice *dimm;
3503     uint32_t nr_lmbs;
3504     QTAILQ_ENTRY(SpaprDimmState) next;
3505 };
3506 
3507 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3508                                                        PCDIMMDevice *dimm)
3509 {
3510     SpaprDimmState *dimm_state = NULL;
3511 
3512     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3513         if (dimm_state->dimm == dimm) {
3514             break;
3515         }
3516     }
3517     return dimm_state;
3518 }
3519 
3520 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3521                                                       uint32_t nr_lmbs,
3522                                                       PCDIMMDevice *dimm)
3523 {
3524     SpaprDimmState *ds = NULL;
3525 
3526     /*
3527      * If this request is for a DIMM whose removal had failed earlier
3528      * (due to guest's refusal to remove the LMBs), we would have this
3529      * dimm already in the pending_dimm_unplugs list. In that
3530      * case don't add again.
3531      */
3532     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3533     if (!ds) {
3534         ds = g_malloc0(sizeof(SpaprDimmState));
3535         ds->nr_lmbs = nr_lmbs;
3536         ds->dimm = dimm;
3537         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3538     }
3539     return ds;
3540 }
3541 
3542 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3543                                               SpaprDimmState *dimm_state)
3544 {
3545     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3546     g_free(dimm_state);
3547 }
3548 
3549 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3550                                                         PCDIMMDevice *dimm)
3551 {
3552     SpaprDrc *drc;
3553     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3554                                                   &error_abort);
3555     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3556     uint32_t avail_lmbs = 0;
3557     uint64_t addr_start, addr;
3558     int i;
3559 
3560     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3561                                           &error_abort);
3562 
3563     addr = addr_start;
3564     for (i = 0; i < nr_lmbs; i++) {
3565         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3566                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3567         g_assert(drc);
3568         if (drc->dev) {
3569             avail_lmbs++;
3570         }
3571         addr += SPAPR_MEMORY_BLOCK_SIZE;
3572     }
3573 
3574     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3575 }
3576 
3577 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3578 {
3579     SpaprDimmState *ds;
3580     PCDIMMDevice *dimm;
3581     SpaprDrc *drc;
3582     uint32_t nr_lmbs;
3583     uint64_t size, addr_start, addr;
3584     g_autofree char *qapi_error = NULL;
3585     int i;
3586 
3587     if (!dev) {
3588         return;
3589     }
3590 
3591     dimm = PC_DIMM(dev);
3592     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3593 
3594     /*
3595      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3596      * unplug state, but one of its DRC is marked as unplug_requested.
3597      * This is bad and weird enough to g_assert() out.
3598      */
3599     g_assert(ds);
3600 
3601     spapr_pending_dimm_unplugs_remove(spapr, ds);
3602 
3603     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3604     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3605 
3606     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3607                                           &error_abort);
3608 
3609     addr = addr_start;
3610     for (i = 0; i < nr_lmbs; i++) {
3611         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3612                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3613         g_assert(drc);
3614 
3615         drc->unplug_requested = false;
3616         addr += SPAPR_MEMORY_BLOCK_SIZE;
3617     }
3618 
3619     /*
3620      * Tell QAPI that something happened and the memory
3621      * hotunplug wasn't successful.
3622      */
3623     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3624                                  "for device %s", dev->id);
3625     qapi_event_send_mem_unplug_error(dev->id, qapi_error);
3626 }
3627 
3628 /* Callback to be called during DRC release. */
3629 void spapr_lmb_release(DeviceState *dev)
3630 {
3631     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3632     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3633     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3634 
3635     /* This information will get lost if a migration occurs
3636      * during the unplug process. In this case recover it. */
3637     if (ds == NULL) {
3638         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3639         g_assert(ds);
3640         /* The DRC being examined by the caller at least must be counted */
3641         g_assert(ds->nr_lmbs);
3642     }
3643 
3644     if (--ds->nr_lmbs) {
3645         return;
3646     }
3647 
3648     /*
3649      * Now that all the LMBs have been removed by the guest, call the
3650      * unplug handler chain. This can never fail.
3651      */
3652     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3653     object_unparent(OBJECT(dev));
3654 }
3655 
3656 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3657 {
3658     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3659     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3660 
3661     /* We really shouldn't get this far without anything to unplug */
3662     g_assert(ds);
3663 
3664     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3665     qdev_unrealize(dev);
3666     spapr_pending_dimm_unplugs_remove(spapr, ds);
3667 }
3668 
3669 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3670                                         DeviceState *dev, Error **errp)
3671 {
3672     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3673     PCDIMMDevice *dimm = PC_DIMM(dev);
3674     uint32_t nr_lmbs;
3675     uint64_t size, addr_start, addr;
3676     int i;
3677     SpaprDrc *drc;
3678 
3679     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3680         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3681         return;
3682     }
3683 
3684     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3685     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3686 
3687     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3688                                           &error_abort);
3689 
3690     /*
3691      * An existing pending dimm state for this DIMM means that there is an
3692      * unplug operation in progress, waiting for the spapr_lmb_release
3693      * callback to complete the job (BQL can't cover that far). In this case,
3694      * bail out to avoid detaching DRCs that were already released.
3695      */
3696     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3697         error_setg(errp, "Memory unplug already in progress for device %s",
3698                    dev->id);
3699         return;
3700     }
3701 
3702     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3703 
3704     addr = addr_start;
3705     for (i = 0; i < nr_lmbs; i++) {
3706         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3707                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3708         g_assert(drc);
3709 
3710         spapr_drc_unplug_request(drc);
3711         addr += SPAPR_MEMORY_BLOCK_SIZE;
3712     }
3713 
3714     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3715                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3716     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3717                                               nr_lmbs, spapr_drc_index(drc));
3718 }
3719 
3720 /* Callback to be called during DRC release. */
3721 void spapr_core_release(DeviceState *dev)
3722 {
3723     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3724 
3725     /* Call the unplug handler chain. This can never fail. */
3726     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3727     object_unparent(OBJECT(dev));
3728 }
3729 
3730 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3731 {
3732     MachineState *ms = MACHINE(hotplug_dev);
3733     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3734     CPUCore *cc = CPU_CORE(dev);
3735     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3736 
3737     if (smc->pre_2_10_has_unused_icps) {
3738         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3739         int i;
3740 
3741         for (i = 0; i < cc->nr_threads; i++) {
3742             CPUState *cs = CPU(sc->threads[i]);
3743 
3744             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3745         }
3746     }
3747 
3748     assert(core_slot);
3749     core_slot->cpu = NULL;
3750     qdev_unrealize(dev);
3751 }
3752 
3753 static
3754 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3755                                Error **errp)
3756 {
3757     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3758     int index;
3759     SpaprDrc *drc;
3760     CPUCore *cc = CPU_CORE(dev);
3761 
3762     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3763         error_setg(errp, "Unable to find CPU core with core-id: %d",
3764                    cc->core_id);
3765         return;
3766     }
3767     if (index == 0) {
3768         error_setg(errp, "Boot CPU core may not be unplugged");
3769         return;
3770     }
3771 
3772     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3773                           spapr_vcpu_id(spapr, cc->core_id));
3774     g_assert(drc);
3775 
3776     if (!spapr_drc_unplug_requested(drc)) {
3777         spapr_drc_unplug_request(drc);
3778     }
3779 
3780     /*
3781      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3782      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3783      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3784      * attempt (e.g. the kernel will refuse to remove the last online
3785      * CPU), we will never attempt it again because unplug_requested
3786      * will still be 'true' in that case.
3787      */
3788     spapr_hotplug_req_remove_by_index(drc);
3789 }
3790 
3791 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3792                            void *fdt, int *fdt_start_offset, Error **errp)
3793 {
3794     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3795     CPUState *cs = CPU(core->threads[0]);
3796     PowerPCCPU *cpu = POWERPC_CPU(cs);
3797     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3798     int id = spapr_get_vcpu_id(cpu);
3799     g_autofree char *nodename = NULL;
3800     int offset;
3801 
3802     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3803     offset = fdt_add_subnode(fdt, 0, nodename);
3804 
3805     spapr_dt_cpu(cs, fdt, offset, spapr);
3806 
3807     /*
3808      * spapr_dt_cpu() does not fill the 'name' property in the
3809      * CPU node. The function is called during boot process, before
3810      * and after CAS, and overwriting the 'name' property written
3811      * by SLOF is not allowed.
3812      *
3813      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3814      * CPUs more compatible with the coldplugged ones, which have
3815      * the 'name' property. Linux Kernel also relies on this
3816      * property to identify CPU nodes.
3817      */
3818     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3819 
3820     *fdt_start_offset = offset;
3821     return 0;
3822 }
3823 
3824 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3825 {
3826     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3827     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3828     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3829     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3830     CPUCore *cc = CPU_CORE(dev);
3831     CPUState *cs;
3832     SpaprDrc *drc;
3833     CPUArchId *core_slot;
3834     int index;
3835     bool hotplugged = spapr_drc_hotplugged(dev);
3836     int i;
3837 
3838     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3839     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3840 
3841     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3842                           spapr_vcpu_id(spapr, cc->core_id));
3843 
3844     g_assert(drc || !mc->has_hotpluggable_cpus);
3845 
3846     if (drc) {
3847         /*
3848          * spapr_core_pre_plug() already buys us this is a brand new
3849          * core being plugged into a free slot. Nothing should already
3850          * be attached to the corresponding DRC.
3851          */
3852         spapr_drc_attach(drc, dev);
3853 
3854         if (hotplugged) {
3855             /*
3856              * Send hotplug notification interrupt to the guest only
3857              * in case of hotplugged CPUs.
3858              */
3859             spapr_hotplug_req_add_by_index(drc);
3860         } else {
3861             spapr_drc_reset(drc);
3862         }
3863     }
3864 
3865     core_slot->cpu = OBJECT(dev);
3866 
3867     /*
3868      * Set compatibility mode to match the boot CPU, which was either set
3869      * by the machine reset code or by CAS. This really shouldn't fail at
3870      * this point.
3871      */
3872     if (hotplugged) {
3873         for (i = 0; i < cc->nr_threads; i++) {
3874             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3875                            &error_abort);
3876         }
3877     }
3878 
3879     if (smc->pre_2_10_has_unused_icps) {
3880         for (i = 0; i < cc->nr_threads; i++) {
3881             cs = CPU(core->threads[i]);
3882             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3883         }
3884     }
3885 }
3886 
3887 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3888                                 Error **errp)
3889 {
3890     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3891     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3892     CPUCore *cc = CPU_CORE(dev);
3893     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3894     const char *type = object_get_typename(OBJECT(dev));
3895     CPUArchId *core_slot;
3896     int index;
3897     unsigned int smp_threads = machine->smp.threads;
3898 
3899     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3900         error_setg(errp, "CPU hotplug not supported for this machine");
3901         return;
3902     }
3903 
3904     if (strcmp(base_core_type, type)) {
3905         error_setg(errp, "CPU core type should be %s", base_core_type);
3906         return;
3907     }
3908 
3909     if (cc->core_id % smp_threads) {
3910         error_setg(errp, "invalid core id %d", cc->core_id);
3911         return;
3912     }
3913 
3914     /*
3915      * In general we should have homogeneous threads-per-core, but old
3916      * (pre hotplug support) machine types allow the last core to have
3917      * reduced threads as a compatibility hack for when we allowed
3918      * total vcpus not a multiple of threads-per-core.
3919      */
3920     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3921         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3922                    smp_threads);
3923         return;
3924     }
3925 
3926     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3927     if (!core_slot) {
3928         error_setg(errp, "core id %d out of range", cc->core_id);
3929         return;
3930     }
3931 
3932     if (core_slot->cpu) {
3933         error_setg(errp, "core %d already populated", cc->core_id);
3934         return;
3935     }
3936 
3937     numa_cpu_pre_plug(core_slot, dev, errp);
3938 }
3939 
3940 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3941                           void *fdt, int *fdt_start_offset, Error **errp)
3942 {
3943     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3944     int intc_phandle;
3945 
3946     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3947     if (intc_phandle <= 0) {
3948         return -1;
3949     }
3950 
3951     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3952         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3953         return -1;
3954     }
3955 
3956     /* generally SLOF creates these, for hotplug it's up to QEMU */
3957     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3958 
3959     return 0;
3960 }
3961 
3962 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3963                                Error **errp)
3964 {
3965     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3966     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3967     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3968     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3969     SpaprDrc *drc;
3970 
3971     if (dev->hotplugged && !smc->dr_phb_enabled) {
3972         error_setg(errp, "PHB hotplug not supported for this machine");
3973         return false;
3974     }
3975 
3976     if (sphb->index == (uint32_t)-1) {
3977         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3978         return false;
3979     }
3980 
3981     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3982     if (drc && drc->dev) {
3983         error_setg(errp, "PHB %d already attached", sphb->index);
3984         return false;
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     return
3992         smc->phb_placement(spapr, sphb->index,
3993                            &sphb->buid, &sphb->io_win_addr,
3994                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
3995                            windows_supported, sphb->dma_liobn,
3996                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3997                            errp);
3998 }
3999 
4000 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
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 
4008     if (!smc->dr_phb_enabled) {
4009         return;
4010     }
4011 
4012     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4013     /* hotplug hooks should check it's enabled before getting this far */
4014     assert(drc);
4015 
4016     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4017     spapr_drc_attach(drc, dev);
4018 
4019     if (hotplugged) {
4020         spapr_hotplug_req_add_by_index(drc);
4021     } else {
4022         spapr_drc_reset(drc);
4023     }
4024 }
4025 
4026 void spapr_phb_release(DeviceState *dev)
4027 {
4028     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4029 
4030     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4031     object_unparent(OBJECT(dev));
4032 }
4033 
4034 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4035 {
4036     qdev_unrealize(dev);
4037 }
4038 
4039 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4040                                      DeviceState *dev, Error **errp)
4041 {
4042     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4043     SpaprDrc *drc;
4044 
4045     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4046     assert(drc);
4047 
4048     if (!spapr_drc_unplug_requested(drc)) {
4049         spapr_drc_unplug_request(drc);
4050         spapr_hotplug_req_remove_by_index(drc);
4051     } else {
4052         error_setg(errp,
4053                    "PCI Host Bridge unplug already in progress for device %s",
4054                    dev->id);
4055     }
4056 }
4057 
4058 static
4059 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4060                               Error **errp)
4061 {
4062     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4063 
4064     if (spapr->tpm_proxy != NULL) {
4065         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4066         return false;
4067     }
4068 
4069     return true;
4070 }
4071 
4072 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4073 {
4074     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4075     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4076 
4077     /* Already checked in spapr_tpm_proxy_pre_plug() */
4078     g_assert(spapr->tpm_proxy == NULL);
4079 
4080     spapr->tpm_proxy = tpm_proxy;
4081 }
4082 
4083 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4084 {
4085     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4086 
4087     qdev_unrealize(dev);
4088     object_unparent(OBJECT(dev));
4089     spapr->tpm_proxy = NULL;
4090 }
4091 
4092 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4093                                       DeviceState *dev, Error **errp)
4094 {
4095     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4096         spapr_memory_plug(hotplug_dev, dev);
4097     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4098         spapr_core_plug(hotplug_dev, dev);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4100         spapr_phb_plug(hotplug_dev, dev);
4101     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4102         spapr_tpm_proxy_plug(hotplug_dev, dev);
4103     }
4104 }
4105 
4106 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4107                                         DeviceState *dev, Error **errp)
4108 {
4109     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4110         spapr_memory_unplug(hotplug_dev, dev);
4111     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4112         spapr_core_unplug(hotplug_dev, dev);
4113     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4114         spapr_phb_unplug(hotplug_dev, dev);
4115     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4116         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4117     }
4118 }
4119 
4120 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4121 {
4122     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4123         /*
4124          * CAS will process all pending unplug requests.
4125          *
4126          * HACK: a guest could theoretically have cleared all bits in OV5,
4127          * but none of the guests we care for do.
4128          */
4129         spapr_ovec_empty(spapr->ov5_cas);
4130 }
4131 
4132 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4133                                                 DeviceState *dev, Error **errp)
4134 {
4135     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4136     MachineClass *mc = MACHINE_GET_CLASS(sms);
4137     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4138 
4139     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4140         if (spapr_memory_hot_unplug_supported(sms)) {
4141             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4142         } else {
4143             error_setg(errp, "Memory hot unplug not supported for this guest");
4144         }
4145     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4146         if (!mc->has_hotpluggable_cpus) {
4147             error_setg(errp, "CPU hot unplug not supported on this machine");
4148             return;
4149         }
4150         spapr_core_unplug_request(hotplug_dev, dev, errp);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4152         if (!smc->dr_phb_enabled) {
4153             error_setg(errp, "PHB hot unplug not supported on this machine");
4154             return;
4155         }
4156         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4157     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4158         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4159     }
4160 }
4161 
4162 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4163                                           DeviceState *dev, Error **errp)
4164 {
4165     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4166         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4167     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4168         spapr_core_pre_plug(hotplug_dev, dev, errp);
4169     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4170         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4171     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4172         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4173     }
4174 }
4175 
4176 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4177                                                  DeviceState *dev)
4178 {
4179     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4180         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4181         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4182         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4183         return HOTPLUG_HANDLER(machine);
4184     }
4185     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4186         PCIDevice *pcidev = PCI_DEVICE(dev);
4187         PCIBus *root = pci_device_root_bus(pcidev);
4188         SpaprPhbState *phb =
4189             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4190                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4191 
4192         if (phb) {
4193             return HOTPLUG_HANDLER(phb);
4194         }
4195     }
4196     return NULL;
4197 }
4198 
4199 static CpuInstanceProperties
4200 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4201 {
4202     CPUArchId *core_slot;
4203     MachineClass *mc = MACHINE_GET_CLASS(machine);
4204 
4205     /* make sure possible_cpu are intialized */
4206     mc->possible_cpu_arch_ids(machine);
4207     /* get CPU core slot containing thread that matches cpu_index */
4208     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4209     assert(core_slot);
4210     return core_slot->props;
4211 }
4212 
4213 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4214 {
4215     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4216 }
4217 
4218 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4219 {
4220     int i;
4221     unsigned int smp_threads = machine->smp.threads;
4222     unsigned int smp_cpus = machine->smp.cpus;
4223     const char *core_type;
4224     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4225     MachineClass *mc = MACHINE_GET_CLASS(machine);
4226 
4227     if (!mc->has_hotpluggable_cpus) {
4228         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4229     }
4230     if (machine->possible_cpus) {
4231         assert(machine->possible_cpus->len == spapr_max_cores);
4232         return machine->possible_cpus;
4233     }
4234 
4235     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4236     if (!core_type) {
4237         error_report("Unable to find sPAPR CPU Core definition");
4238         exit(1);
4239     }
4240 
4241     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4242                              sizeof(CPUArchId) * spapr_max_cores);
4243     machine->possible_cpus->len = spapr_max_cores;
4244     for (i = 0; i < machine->possible_cpus->len; i++) {
4245         int core_id = i * smp_threads;
4246 
4247         machine->possible_cpus->cpus[i].type = core_type;
4248         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4249         machine->possible_cpus->cpus[i].arch_id = core_id;
4250         machine->possible_cpus->cpus[i].props.has_core_id = true;
4251         machine->possible_cpus->cpus[i].props.core_id = core_id;
4252     }
4253     return machine->possible_cpus;
4254 }
4255 
4256 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4257                                 uint64_t *buid, hwaddr *pio,
4258                                 hwaddr *mmio32, hwaddr *mmio64,
4259                                 unsigned n_dma, uint32_t *liobns,
4260                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4261 {
4262     /*
4263      * New-style PHB window placement.
4264      *
4265      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4266      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4267      * windows.
4268      *
4269      * Some guest kernels can't work with MMIO windows above 1<<46
4270      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4271      *
4272      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4273      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4274      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4275      * 1TiB 64-bit MMIO windows for each PHB.
4276      */
4277     const uint64_t base_buid = 0x800000020000000ULL;
4278     int i;
4279 
4280     /* Sanity check natural alignments */
4281     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4282     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4283     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4284     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4285     /* Sanity check bounds */
4286     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4287                       SPAPR_PCI_MEM32_WIN_SIZE);
4288     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4289                       SPAPR_PCI_MEM64_WIN_SIZE);
4290 
4291     if (index >= SPAPR_MAX_PHBS) {
4292         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4293                    SPAPR_MAX_PHBS - 1);
4294         return false;
4295     }
4296 
4297     *buid = base_buid + index;
4298     for (i = 0; i < n_dma; ++i) {
4299         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4300     }
4301 
4302     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4303     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4304     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4305 
4306     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4307     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4308     return true;
4309 }
4310 
4311 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4312 {
4313     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4314 
4315     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4316 }
4317 
4318 static void spapr_ics_resend(XICSFabric *dev)
4319 {
4320     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4321 
4322     ics_resend(spapr->ics);
4323 }
4324 
4325 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4326 {
4327     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4328 
4329     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4330 }
4331 
4332 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4333                                  Monitor *mon)
4334 {
4335     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4336 
4337     spapr_irq_print_info(spapr, mon);
4338     monitor_printf(mon, "irqchip: %s\n",
4339                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4340 }
4341 
4342 /*
4343  * This is a XIVE only operation
4344  */
4345 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4346                            uint8_t nvt_blk, uint32_t nvt_idx,
4347                            bool cam_ignore, uint8_t priority,
4348                            uint32_t logic_serv, XiveTCTXMatch *match)
4349 {
4350     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4351     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4352     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4353     int count;
4354 
4355     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4356                            priority, logic_serv, match);
4357     if (count < 0) {
4358         return count;
4359     }
4360 
4361     /*
4362      * When we implement the save and restore of the thread interrupt
4363      * contexts in the enter/exit CPU handlers of the machine and the
4364      * escalations in QEMU, we should be able to handle non dispatched
4365      * vCPUs.
4366      *
4367      * Until this is done, the sPAPR machine should find at least one
4368      * matching context always.
4369      */
4370     if (count == 0) {
4371         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4372                       nvt_blk, nvt_idx);
4373     }
4374 
4375     return count;
4376 }
4377 
4378 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4379 {
4380     return cpu->vcpu_id;
4381 }
4382 
4383 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4384 {
4385     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4386     MachineState *ms = MACHINE(spapr);
4387     int vcpu_id;
4388 
4389     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4390 
4391     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4392         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4393         error_append_hint(errp, "Adjust the number of cpus to %d "
4394                           "or try to raise the number of threads per core\n",
4395                           vcpu_id * ms->smp.threads / spapr->vsmt);
4396         return false;
4397     }
4398 
4399     cpu->vcpu_id = vcpu_id;
4400     return true;
4401 }
4402 
4403 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4404 {
4405     CPUState *cs;
4406 
4407     CPU_FOREACH(cs) {
4408         PowerPCCPU *cpu = POWERPC_CPU(cs);
4409 
4410         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4411             return cpu;
4412         }
4413     }
4414 
4415     return NULL;
4416 }
4417 
4418 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4419 {
4420     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4421 
4422     /* These are only called by TCG, KVM maintains dispatch state */
4423 
4424     spapr_cpu->prod = false;
4425     if (spapr_cpu->vpa_addr) {
4426         CPUState *cs = CPU(cpu);
4427         uint32_t dispatch;
4428 
4429         dispatch = ldl_be_phys(cs->as,
4430                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4431         dispatch++;
4432         if ((dispatch & 1) != 0) {
4433             qemu_log_mask(LOG_GUEST_ERROR,
4434                           "VPA: incorrect dispatch counter value for "
4435                           "dispatched partition %u, correcting.\n", dispatch);
4436             dispatch++;
4437         }
4438         stl_be_phys(cs->as,
4439                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4440     }
4441 }
4442 
4443 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4444 {
4445     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4446 
4447     if (spapr_cpu->vpa_addr) {
4448         CPUState *cs = CPU(cpu);
4449         uint32_t dispatch;
4450 
4451         dispatch = ldl_be_phys(cs->as,
4452                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4453         dispatch++;
4454         if ((dispatch & 1) != 1) {
4455             qemu_log_mask(LOG_GUEST_ERROR,
4456                           "VPA: incorrect dispatch counter value for "
4457                           "preempted partition %u, correcting.\n", dispatch);
4458             dispatch++;
4459         }
4460         stl_be_phys(cs->as,
4461                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4462     }
4463 }
4464 
4465 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4466 {
4467     MachineClass *mc = MACHINE_CLASS(oc);
4468     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4469     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4470     NMIClass *nc = NMI_CLASS(oc);
4471     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4472     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4473     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4474     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4475     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4476 
4477     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4478     mc->ignore_boot_device_suffixes = true;
4479 
4480     /*
4481      * We set up the default / latest behaviour here.  The class_init
4482      * functions for the specific versioned machine types can override
4483      * these details for backwards compatibility
4484      */
4485     mc->init = spapr_machine_init;
4486     mc->reset = spapr_machine_reset;
4487     mc->block_default_type = IF_SCSI;
4488 
4489     /*
4490      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4491      * should be limited by the host capability instead of hardcoded.
4492      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4493      * guests are welcome to have as many CPUs as the host are capable
4494      * of emulate.
4495      */
4496     mc->max_cpus = INT32_MAX;
4497 
4498     mc->no_parallel = 1;
4499     mc->default_boot_order = "";
4500     mc->default_ram_size = 512 * MiB;
4501     mc->default_ram_id = "ppc_spapr.ram";
4502     mc->default_display = "std";
4503     mc->kvm_type = spapr_kvm_type;
4504     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4505     mc->pci_allow_0_address = true;
4506     assert(!mc->get_hotplug_handler);
4507     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4508     hc->pre_plug = spapr_machine_device_pre_plug;
4509     hc->plug = spapr_machine_device_plug;
4510     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4511     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4512     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4513     hc->unplug_request = spapr_machine_device_unplug_request;
4514     hc->unplug = spapr_machine_device_unplug;
4515 
4516     smc->dr_lmb_enabled = true;
4517     smc->update_dt_enabled = true;
4518     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4519     mc->has_hotpluggable_cpus = true;
4520     mc->nvdimm_supported = true;
4521     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4522     fwc->get_dev_path = spapr_get_fw_dev_path;
4523     nc->nmi_monitor_handler = spapr_nmi;
4524     smc->phb_placement = spapr_phb_placement;
4525     vhc->hypercall = emulate_spapr_hypercall;
4526     vhc->hpt_mask = spapr_hpt_mask;
4527     vhc->map_hptes = spapr_map_hptes;
4528     vhc->unmap_hptes = spapr_unmap_hptes;
4529     vhc->hpte_set_c = spapr_hpte_set_c;
4530     vhc->hpte_set_r = spapr_hpte_set_r;
4531     vhc->get_pate = spapr_get_pate;
4532     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4533     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4534     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4535     xic->ics_get = spapr_ics_get;
4536     xic->ics_resend = spapr_ics_resend;
4537     xic->icp_get = spapr_icp_get;
4538     ispc->print_info = spapr_pic_print_info;
4539     /* Force NUMA node memory size to be a multiple of
4540      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4541      * in which LMBs are represented and hot-added
4542      */
4543     mc->numa_mem_align_shift = 28;
4544     mc->auto_enable_numa = true;
4545 
4546     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4547     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4548     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4549     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4550     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4551     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4552     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4553     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4554     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4555     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4556     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4557     spapr_caps_add_properties(smc);
4558     smc->irq = &spapr_irq_dual;
4559     smc->dr_phb_enabled = true;
4560     smc->linux_pci_probe = true;
4561     smc->smp_threads_vsmt = true;
4562     smc->nr_xirqs = SPAPR_NR_XIRQS;
4563     xfc->match_nvt = spapr_match_nvt;
4564 }
4565 
4566 static const TypeInfo spapr_machine_info = {
4567     .name          = TYPE_SPAPR_MACHINE,
4568     .parent        = TYPE_MACHINE,
4569     .abstract      = true,
4570     .instance_size = sizeof(SpaprMachineState),
4571     .instance_init = spapr_instance_init,
4572     .instance_finalize = spapr_machine_finalizefn,
4573     .class_size    = sizeof(SpaprMachineClass),
4574     .class_init    = spapr_machine_class_init,
4575     .interfaces = (InterfaceInfo[]) {
4576         { TYPE_FW_PATH_PROVIDER },
4577         { TYPE_NMI },
4578         { TYPE_HOTPLUG_HANDLER },
4579         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4580         { TYPE_XICS_FABRIC },
4581         { TYPE_INTERRUPT_STATS_PROVIDER },
4582         { TYPE_XIVE_FABRIC },
4583         { }
4584     },
4585 };
4586 
4587 static void spapr_machine_latest_class_options(MachineClass *mc)
4588 {
4589     mc->alias = "pseries";
4590     mc->is_default = true;
4591 }
4592 
4593 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4594     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4595                                                     void *data)      \
4596     {                                                                \
4597         MachineClass *mc = MACHINE_CLASS(oc);                        \
4598         spapr_machine_##suffix##_class_options(mc);                  \
4599         if (latest) {                                                \
4600             spapr_machine_latest_class_options(mc);                  \
4601         }                                                            \
4602     }                                                                \
4603     static const TypeInfo spapr_machine_##suffix##_info = {          \
4604         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4605         .parent = TYPE_SPAPR_MACHINE,                                \
4606         .class_init = spapr_machine_##suffix##_class_init,           \
4607     };                                                               \
4608     static void spapr_machine_register_##suffix(void)                \
4609     {                                                                \
4610         type_register(&spapr_machine_##suffix##_info);               \
4611     }                                                                \
4612     type_init(spapr_machine_register_##suffix)
4613 
4614 /*
4615  * pseries-6.1
4616  */
4617 static void spapr_machine_6_1_class_options(MachineClass *mc)
4618 {
4619     /* Defaults for the latest behaviour inherited from the base class */
4620 }
4621 
4622 DEFINE_SPAPR_MACHINE(6_1, "6.1", true);
4623 
4624 /*
4625  * pseries-6.0
4626  */
4627 static void spapr_machine_6_0_class_options(MachineClass *mc)
4628 {
4629     spapr_machine_6_1_class_options(mc);
4630     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4631 }
4632 
4633 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4634 
4635 /*
4636  * pseries-5.2
4637  */
4638 static void spapr_machine_5_2_class_options(MachineClass *mc)
4639 {
4640     spapr_machine_6_0_class_options(mc);
4641     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4642 }
4643 
4644 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4645 
4646 /*
4647  * pseries-5.1
4648  */
4649 static void spapr_machine_5_1_class_options(MachineClass *mc)
4650 {
4651     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4652 
4653     spapr_machine_5_2_class_options(mc);
4654     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4655     smc->pre_5_2_numa_associativity = true;
4656 }
4657 
4658 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4659 
4660 /*
4661  * pseries-5.0
4662  */
4663 static void spapr_machine_5_0_class_options(MachineClass *mc)
4664 {
4665     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4666     static GlobalProperty compat[] = {
4667         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4668     };
4669 
4670     spapr_machine_5_1_class_options(mc);
4671     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4672     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4673     mc->numa_mem_supported = true;
4674     smc->pre_5_1_assoc_refpoints = true;
4675 }
4676 
4677 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4678 
4679 /*
4680  * pseries-4.2
4681  */
4682 static void spapr_machine_4_2_class_options(MachineClass *mc)
4683 {
4684     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4685 
4686     spapr_machine_5_0_class_options(mc);
4687     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4688     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4689     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4690     smc->rma_limit = 16 * GiB;
4691     mc->nvdimm_supported = false;
4692 }
4693 
4694 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4695 
4696 /*
4697  * pseries-4.1
4698  */
4699 static void spapr_machine_4_1_class_options(MachineClass *mc)
4700 {
4701     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4702     static GlobalProperty compat[] = {
4703         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4704         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4705     };
4706 
4707     spapr_machine_4_2_class_options(mc);
4708     smc->linux_pci_probe = false;
4709     smc->smp_threads_vsmt = false;
4710     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4711     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4715 
4716 /*
4717  * pseries-4.0
4718  */
4719 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4720                               uint64_t *buid, hwaddr *pio,
4721                               hwaddr *mmio32, hwaddr *mmio64,
4722                               unsigned n_dma, uint32_t *liobns,
4723                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4724 {
4725     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4726                              liobns, nv2gpa, nv2atsd, errp)) {
4727         return false;
4728     }
4729 
4730     *nv2gpa = 0;
4731     *nv2atsd = 0;
4732     return true;
4733 }
4734 static void spapr_machine_4_0_class_options(MachineClass *mc)
4735 {
4736     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4737 
4738     spapr_machine_4_1_class_options(mc);
4739     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4740     smc->phb_placement = phb_placement_4_0;
4741     smc->irq = &spapr_irq_xics;
4742     smc->pre_4_1_migration = true;
4743 }
4744 
4745 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4746 
4747 /*
4748  * pseries-3.1
4749  */
4750 static void spapr_machine_3_1_class_options(MachineClass *mc)
4751 {
4752     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4753 
4754     spapr_machine_4_0_class_options(mc);
4755     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4756 
4757     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4758     smc->update_dt_enabled = false;
4759     smc->dr_phb_enabled = false;
4760     smc->broken_host_serial_model = true;
4761     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4762     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4763     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4764     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4765 }
4766 
4767 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4768 
4769 /*
4770  * pseries-3.0
4771  */
4772 
4773 static void spapr_machine_3_0_class_options(MachineClass *mc)
4774 {
4775     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4776 
4777     spapr_machine_3_1_class_options(mc);
4778     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4779 
4780     smc->legacy_irq_allocation = true;
4781     smc->nr_xirqs = 0x400;
4782     smc->irq = &spapr_irq_xics_legacy;
4783 }
4784 
4785 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4786 
4787 /*
4788  * pseries-2.12
4789  */
4790 static void spapr_machine_2_12_class_options(MachineClass *mc)
4791 {
4792     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4793     static GlobalProperty compat[] = {
4794         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4795         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4796     };
4797 
4798     spapr_machine_3_0_class_options(mc);
4799     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4800     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4801 
4802     /* We depend on kvm_enabled() to choose a default value for the
4803      * hpt-max-page-size capability. Of course we can't do it here
4804      * because this is too early and the HW accelerator isn't initialzed
4805      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4806      */
4807     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4808 }
4809 
4810 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4811 
4812 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4813 {
4814     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4815 
4816     spapr_machine_2_12_class_options(mc);
4817     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4818     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4819     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4820 }
4821 
4822 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4823 
4824 /*
4825  * pseries-2.11
4826  */
4827 
4828 static void spapr_machine_2_11_class_options(MachineClass *mc)
4829 {
4830     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4831 
4832     spapr_machine_2_12_class_options(mc);
4833     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4834     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4835 }
4836 
4837 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4838 
4839 /*
4840  * pseries-2.10
4841  */
4842 
4843 static void spapr_machine_2_10_class_options(MachineClass *mc)
4844 {
4845     spapr_machine_2_11_class_options(mc);
4846     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4847 }
4848 
4849 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4850 
4851 /*
4852  * pseries-2.9
4853  */
4854 
4855 static void spapr_machine_2_9_class_options(MachineClass *mc)
4856 {
4857     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4858     static GlobalProperty compat[] = {
4859         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4860     };
4861 
4862     spapr_machine_2_10_class_options(mc);
4863     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4864     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4865     smc->pre_2_10_has_unused_icps = true;
4866     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4867 }
4868 
4869 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4870 
4871 /*
4872  * pseries-2.8
4873  */
4874 
4875 static void spapr_machine_2_8_class_options(MachineClass *mc)
4876 {
4877     static GlobalProperty compat[] = {
4878         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4879     };
4880 
4881     spapr_machine_2_9_class_options(mc);
4882     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4883     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4884     mc->numa_mem_align_shift = 23;
4885 }
4886 
4887 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4888 
4889 /*
4890  * pseries-2.7
4891  */
4892 
4893 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4894                               uint64_t *buid, hwaddr *pio,
4895                               hwaddr *mmio32, hwaddr *mmio64,
4896                               unsigned n_dma, uint32_t *liobns,
4897                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4898 {
4899     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4900     const uint64_t base_buid = 0x800000020000000ULL;
4901     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4902     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4903     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4904     const uint32_t max_index = 255;
4905     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4906 
4907     uint64_t ram_top = MACHINE(spapr)->ram_size;
4908     hwaddr phb0_base, phb_base;
4909     int i;
4910 
4911     /* Do we have device memory? */
4912     if (MACHINE(spapr)->maxram_size > ram_top) {
4913         /* Can't just use maxram_size, because there may be an
4914          * alignment gap between normal and device memory regions
4915          */
4916         ram_top = MACHINE(spapr)->device_memory->base +
4917             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4918     }
4919 
4920     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4921 
4922     if (index > max_index) {
4923         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4924                    max_index);
4925         return false;
4926     }
4927 
4928     *buid = base_buid + index;
4929     for (i = 0; i < n_dma; ++i) {
4930         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4931     }
4932 
4933     phb_base = phb0_base + index * phb_spacing;
4934     *pio = phb_base + pio_offset;
4935     *mmio32 = phb_base + mmio_offset;
4936     /*
4937      * We don't set the 64-bit MMIO window, relying on the PHB's
4938      * fallback behaviour of automatically splitting a large "32-bit"
4939      * window into contiguous 32-bit and 64-bit windows
4940      */
4941 
4942     *nv2gpa = 0;
4943     *nv2atsd = 0;
4944     return true;
4945 }
4946 
4947 static void spapr_machine_2_7_class_options(MachineClass *mc)
4948 {
4949     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4950     static GlobalProperty compat[] = {
4951         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4952         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4953         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4954         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4955     };
4956 
4957     spapr_machine_2_8_class_options(mc);
4958     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4959     mc->default_machine_opts = "modern-hotplug-events=off";
4960     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4961     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4962     smc->phb_placement = phb_placement_2_7;
4963 }
4964 
4965 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4966 
4967 /*
4968  * pseries-2.6
4969  */
4970 
4971 static void spapr_machine_2_6_class_options(MachineClass *mc)
4972 {
4973     static GlobalProperty compat[] = {
4974         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4975     };
4976 
4977     spapr_machine_2_7_class_options(mc);
4978     mc->has_hotpluggable_cpus = false;
4979     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4980     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4981 }
4982 
4983 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4984 
4985 /*
4986  * pseries-2.5
4987  */
4988 
4989 static void spapr_machine_2_5_class_options(MachineClass *mc)
4990 {
4991     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4992     static GlobalProperty compat[] = {
4993         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4994     };
4995 
4996     spapr_machine_2_6_class_options(mc);
4997     smc->use_ohci_by_default = true;
4998     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4999     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5000 }
5001 
5002 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5003 
5004 /*
5005  * pseries-2.4
5006  */
5007 
5008 static void spapr_machine_2_4_class_options(MachineClass *mc)
5009 {
5010     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5011 
5012     spapr_machine_2_5_class_options(mc);
5013     smc->dr_lmb_enabled = false;
5014     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5015 }
5016 
5017 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5018 
5019 /*
5020  * pseries-2.3
5021  */
5022 
5023 static void spapr_machine_2_3_class_options(MachineClass *mc)
5024 {
5025     static GlobalProperty compat[] = {
5026         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5027     };
5028     spapr_machine_2_4_class_options(mc);
5029     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5030     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5031 }
5032 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5033 
5034 /*
5035  * pseries-2.2
5036  */
5037 
5038 static void spapr_machine_2_2_class_options(MachineClass *mc)
5039 {
5040     static GlobalProperty compat[] = {
5041         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5042     };
5043 
5044     spapr_machine_2_3_class_options(mc);
5045     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5046     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5047     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5048 }
5049 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5050 
5051 /*
5052  * pseries-2.1
5053  */
5054 
5055 static void spapr_machine_2_1_class_options(MachineClass *mc)
5056 {
5057     spapr_machine_2_2_class_options(mc);
5058     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5059 }
5060 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5061 
5062 static void spapr_machine_register_types(void)
5063 {
5064     type_register_static(&spapr_machine_info);
5065 }
5066 
5067 type_init(spapr_machine_register_types)
5068