xref: /openbmc/qemu/hw/ppc/spapr.c (revision 06831001)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu/datadir.h"
29 #include "qemu/memalign.h"
30 #include "qemu/guest-random.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-events-machine.h"
33 #include "qapi/qapi-events-qdev.h"
34 #include "qapi/visitor.h"
35 #include "sysemu/sysemu.h"
36 #include "sysemu/hostmem.h"
37 #include "sysemu/numa.h"
38 #include "sysemu/qtest.h"
39 #include "sysemu/reset.h"
40 #include "sysemu/runstate.h"
41 #include "qemu/log.h"
42 #include "hw/fw-path-provider.h"
43 #include "elf.h"
44 #include "net/net.h"
45 #include "sysemu/device_tree.h"
46 #include "sysemu/cpus.h"
47 #include "sysemu/hw_accel.h"
48 #include "kvm_ppc.h"
49 #include "migration/misc.h"
50 #include "migration/qemu-file-types.h"
51 #include "migration/global_state.h"
52 #include "migration/register.h"
53 #include "migration/blocker.h"
54 #include "mmu-hash64.h"
55 #include "mmu-book3s-v3.h"
56 #include "cpu-models.h"
57 #include "hw/core/cpu.h"
58 
59 #include "hw/ppc/ppc.h"
60 #include "hw/loader.h"
61 
62 #include "hw/ppc/fdt.h"
63 #include "hw/ppc/spapr.h"
64 #include "hw/ppc/spapr_vio.h"
65 #include "hw/ppc/vof.h"
66 #include "hw/qdev-properties.h"
67 #include "hw/pci-host/spapr.h"
68 #include "hw/pci/msi.h"
69 
70 #include "hw/pci/pci.h"
71 #include "hw/scsi/scsi.h"
72 #include "hw/virtio/virtio-scsi.h"
73 #include "hw/virtio/vhost-scsi-common.h"
74 
75 #include "exec/ram_addr.h"
76 #include "hw/usb.h"
77 #include "qemu/config-file.h"
78 #include "qemu/error-report.h"
79 #include "trace.h"
80 #include "hw/nmi.h"
81 #include "hw/intc/intc.h"
82 
83 #include "hw/ppc/spapr_cpu_core.h"
84 #include "hw/mem/memory-device.h"
85 #include "hw/ppc/spapr_tpm_proxy.h"
86 #include "hw/ppc/spapr_nvdimm.h"
87 #include "hw/ppc/spapr_numa.h"
88 #include "hw/ppc/pef.h"
89 
90 #include "monitor/monitor.h"
91 
92 #include <libfdt.h>
93 
94 /* SLOF memory layout:
95  *
96  * SLOF raw image loaded at 0, copies its romfs right below the flat
97  * device-tree, then position SLOF itself 31M below that
98  *
99  * So we set FW_OVERHEAD to 40MB which should account for all of that
100  * and more
101  *
102  * We load our kernel at 4M, leaving space for SLOF initial image
103  */
104 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
105 #define FW_MAX_SIZE             0x400000
106 #define FW_FILE_NAME            "slof.bin"
107 #define FW_FILE_NAME_VOF        "vof.bin"
108 #define FW_OVERHEAD             0x2800000
109 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
110 
111 #define MIN_RMA_SLOF            (128 * MiB)
112 
113 #define PHANDLE_INTC            0x00001111
114 
115 /* These two functions implement the VCPU id numbering: one to compute them
116  * all and one to identify thread 0 of a VCORE. Any change to the first one
117  * is likely to have an impact on the second one, so let's keep them close.
118  */
119 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
120 {
121     MachineState *ms = MACHINE(spapr);
122     unsigned int smp_threads = ms->smp.threads;
123 
124     assert(spapr->vsmt);
125     return
126         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
127 }
128 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
129                                       PowerPCCPU *cpu)
130 {
131     assert(spapr->vsmt);
132     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
133 }
134 
135 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
136 {
137     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
138      * and newer QEMUs don't even have them. In both cases, we don't want
139      * to send anything on the wire.
140      */
141     return false;
142 }
143 
144 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
145     .name = "icp/server",
146     .version_id = 1,
147     .minimum_version_id = 1,
148     .needed = pre_2_10_vmstate_dummy_icp_needed,
149     .fields = (VMStateField[]) {
150         VMSTATE_UNUSED(4), /* uint32_t xirr */
151         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
152         VMSTATE_UNUSED(1), /* uint8_t mfrr */
153         VMSTATE_END_OF_LIST()
154     },
155 };
156 
157 static void pre_2_10_vmstate_register_dummy_icp(int i)
158 {
159     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
160                      (void *)(uintptr_t) i);
161 }
162 
163 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
164 {
165     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
166                        (void *)(uintptr_t) i);
167 }
168 
169 int spapr_max_server_number(SpaprMachineState *spapr)
170 {
171     MachineState *ms = MACHINE(spapr);
172 
173     assert(spapr->vsmt);
174     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
175 }
176 
177 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
178                                   int smt_threads)
179 {
180     int i, ret = 0;
181     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
182     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
183     int index = spapr_get_vcpu_id(cpu);
184 
185     if (cpu->compat_pvr) {
186         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
187         if (ret < 0) {
188             return ret;
189         }
190     }
191 
192     /* Build interrupt servers and gservers properties */
193     for (i = 0; i < smt_threads; i++) {
194         servers_prop[i] = cpu_to_be32(index + i);
195         /* Hack, direct the group queues back to cpu 0 */
196         gservers_prop[i*2] = cpu_to_be32(index + i);
197         gservers_prop[i*2 + 1] = 0;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
200                       servers_prop, sizeof(*servers_prop) * smt_threads);
201     if (ret < 0) {
202         return ret;
203     }
204     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
205                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
206 
207     return ret;
208 }
209 
210 static void spapr_dt_pa_features(SpaprMachineState *spapr,
211                                  PowerPCCPU *cpu,
212                                  void *fdt, int offset)
213 {
214     uint8_t pa_features_206[] = { 6, 0,
215         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
216     uint8_t pa_features_207[] = { 24, 0,
217         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
218         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
219         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
220         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
221     uint8_t pa_features_300[] = { 66, 0,
222         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
223         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
224         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
225         /* 6: DS207 */
226         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
227         /* 16: Vector */
228         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
229         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
231         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
233         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
234         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
235         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
237         /* 42: PM, 44: PC RA, 46: SC vec'd */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
239         /* 48: SIMD, 50: QP BFP, 52: String */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
241         /* 54: DecFP, 56: DecI, 58: SHA */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
243         /* 60: NM atomic, 62: RNG */
244         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
245     };
246     uint8_t *pa_features = NULL;
247     size_t pa_size;
248 
249     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
250         pa_features = pa_features_206;
251         pa_size = sizeof(pa_features_206);
252     }
253     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
254         pa_features = pa_features_207;
255         pa_size = sizeof(pa_features_207);
256     }
257     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
258         pa_features = pa_features_300;
259         pa_size = sizeof(pa_features_300);
260     }
261     if (!pa_features) {
262         return;
263     }
264 
265     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
266         /*
267          * Note: we keep CI large pages off by default because a 64K capable
268          * guest provisioned with large pages might otherwise try to map a qemu
269          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
270          * even if that qemu runs on a 4k host.
271          * We dd this bit back here if we are confident this is not an issue
272          */
273         pa_features[3] |= 0x20;
274     }
275     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
276         pa_features[24] |= 0x80;    /* Transactional memory support */
277     }
278     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
279         /* Workaround for broken kernels that attempt (guest) radix
280          * mode when they can't handle it, if they see the radix bit set
281          * in pa-features. So hide it from them. */
282         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
283     }
284 
285     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
286 }
287 
288 static hwaddr spapr_node0_size(MachineState *machine)
289 {
290     if (machine->numa_state->num_nodes) {
291         int i;
292         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
293             if (machine->numa_state->nodes[i].node_mem) {
294                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
295                            machine->ram_size);
296             }
297         }
298     }
299     return machine->ram_size;
300 }
301 
302 static void add_str(GString *s, const gchar *s1)
303 {
304     g_string_append_len(s, s1, strlen(s1) + 1);
305 }
306 
307 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
308                                 hwaddr start, hwaddr size)
309 {
310     char mem_name[32];
311     uint64_t mem_reg_property[2];
312     int off;
313 
314     mem_reg_property[0] = cpu_to_be64(start);
315     mem_reg_property[1] = cpu_to_be64(size);
316 
317     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
318     off = fdt_add_subnode(fdt, 0, mem_name);
319     _FDT(off);
320     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
321     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
322                       sizeof(mem_reg_property))));
323     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
324     return off;
325 }
326 
327 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
328 {
329     MemoryDeviceInfoList *info;
330 
331     for (info = list; info; info = info->next) {
332         MemoryDeviceInfo *value = info->value;
333 
334         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
335             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
336 
337             if (addr >= pcdimm_info->addr &&
338                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
339                 return pcdimm_info->node;
340             }
341         }
342     }
343 
344     return -1;
345 }
346 
347 struct sPAPRDrconfCellV2 {
348      uint32_t seq_lmbs;
349      uint64_t base_addr;
350      uint32_t drc_index;
351      uint32_t aa_index;
352      uint32_t flags;
353 } QEMU_PACKED;
354 
355 typedef struct DrconfCellQueue {
356     struct sPAPRDrconfCellV2 cell;
357     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
358 } DrconfCellQueue;
359 
360 static DrconfCellQueue *
361 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
362                       uint32_t drc_index, uint32_t aa_index,
363                       uint32_t flags)
364 {
365     DrconfCellQueue *elem;
366 
367     elem = g_malloc0(sizeof(*elem));
368     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
369     elem->cell.base_addr = cpu_to_be64(base_addr);
370     elem->cell.drc_index = cpu_to_be32(drc_index);
371     elem->cell.aa_index = cpu_to_be32(aa_index);
372     elem->cell.flags = cpu_to_be32(flags);
373 
374     return elem;
375 }
376 
377 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
378                                       int offset, MemoryDeviceInfoList *dimms)
379 {
380     MachineState *machine = MACHINE(spapr);
381     uint8_t *int_buf, *cur_index;
382     int ret;
383     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
384     uint64_t addr, cur_addr, size;
385     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
386     uint64_t mem_end = machine->device_memory->base +
387                        memory_region_size(&machine->device_memory->mr);
388     uint32_t node, buf_len, nr_entries = 0;
389     SpaprDrc *drc;
390     DrconfCellQueue *elem, *next;
391     MemoryDeviceInfoList *info;
392     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
393         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
394 
395     /* Entry to cover RAM and the gap area */
396     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
397                                  SPAPR_LMB_FLAGS_RESERVED |
398                                  SPAPR_LMB_FLAGS_DRC_INVALID);
399     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
400     nr_entries++;
401 
402     cur_addr = machine->device_memory->base;
403     for (info = dimms; info; info = info->next) {
404         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
405 
406         addr = di->addr;
407         size = di->size;
408         node = di->node;
409 
410         /*
411          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
412          * area is marked hotpluggable in the next iteration for the bigger
413          * chunk including the NVDIMM occupied area.
414          */
415         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
416             continue;
417 
418         /* Entry for hot-pluggable area */
419         if (cur_addr < addr) {
420             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
421             g_assert(drc);
422             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
423                                          cur_addr, spapr_drc_index(drc), -1, 0);
424             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
425             nr_entries++;
426         }
427 
428         /* Entry for DIMM */
429         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
430         g_assert(drc);
431         elem = spapr_get_drconf_cell(size / lmb_size, addr,
432                                      spapr_drc_index(drc), node,
433                                      (SPAPR_LMB_FLAGS_ASSIGNED |
434                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
435         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
436         nr_entries++;
437         cur_addr = addr + size;
438     }
439 
440     /* Entry for remaining hotpluggable area */
441     if (cur_addr < mem_end) {
442         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
443         g_assert(drc);
444         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
445                                      cur_addr, spapr_drc_index(drc), -1, 0);
446         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
447         nr_entries++;
448     }
449 
450     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
451     int_buf = cur_index = g_malloc0(buf_len);
452     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
453     cur_index += sizeof(nr_entries);
454 
455     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
456         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
457         cur_index += sizeof(elem->cell);
458         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
459         g_free(elem);
460     }
461 
462     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
463     g_free(int_buf);
464     if (ret < 0) {
465         return -1;
466     }
467     return 0;
468 }
469 
470 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
471                                    int offset, MemoryDeviceInfoList *dimms)
472 {
473     MachineState *machine = MACHINE(spapr);
474     int i, ret;
475     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
476     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
477     uint32_t nr_lmbs = (machine->device_memory->base +
478                        memory_region_size(&machine->device_memory->mr)) /
479                        lmb_size;
480     uint32_t *int_buf, *cur_index, buf_len;
481 
482     /*
483      * Allocate enough buffer size to fit in ibm,dynamic-memory
484      */
485     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
486     cur_index = int_buf = g_malloc0(buf_len);
487     int_buf[0] = cpu_to_be32(nr_lmbs);
488     cur_index++;
489     for (i = 0; i < nr_lmbs; i++) {
490         uint64_t addr = i * lmb_size;
491         uint32_t *dynamic_memory = cur_index;
492 
493         if (i >= device_lmb_start) {
494             SpaprDrc *drc;
495 
496             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
497             g_assert(drc);
498 
499             dynamic_memory[0] = cpu_to_be32(addr >> 32);
500             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
501             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
502             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
503             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
504             if (memory_region_present(get_system_memory(), addr)) {
505                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
506             } else {
507                 dynamic_memory[5] = cpu_to_be32(0);
508             }
509         } else {
510             /*
511              * LMB information for RMA, boot time RAM and gap b/n RAM and
512              * device memory region -- all these are marked as reserved
513              * and as having no valid DRC.
514              */
515             dynamic_memory[0] = cpu_to_be32(addr >> 32);
516             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
517             dynamic_memory[2] = cpu_to_be32(0);
518             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
519             dynamic_memory[4] = cpu_to_be32(-1);
520             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
521                                             SPAPR_LMB_FLAGS_DRC_INVALID);
522         }
523 
524         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
525     }
526     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
527     g_free(int_buf);
528     if (ret < 0) {
529         return -1;
530     }
531     return 0;
532 }
533 
534 /*
535  * Adds ibm,dynamic-reconfiguration-memory node.
536  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
537  * of this device tree node.
538  */
539 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
540                                                    void *fdt)
541 {
542     MachineState *machine = MACHINE(spapr);
543     int ret, offset;
544     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
545     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
546                                 cpu_to_be32(lmb_size & 0xffffffff)};
547     MemoryDeviceInfoList *dimms = NULL;
548 
549     /*
550      * Don't create the node if there is no device memory
551      */
552     if (machine->ram_size == machine->maxram_size) {
553         return 0;
554     }
555 
556     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
557 
558     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
559                     sizeof(prop_lmb_size));
560     if (ret < 0) {
561         return ret;
562     }
563 
564     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
565     if (ret < 0) {
566         return ret;
567     }
568 
569     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
570     if (ret < 0) {
571         return ret;
572     }
573 
574     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
575     dimms = qmp_memory_device_list();
576     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
577         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
578     } else {
579         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
580     }
581     qapi_free_MemoryDeviceInfoList(dimms);
582 
583     if (ret < 0) {
584         return ret;
585     }
586 
587     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
588 
589     return ret;
590 }
591 
592 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
593 {
594     MachineState *machine = MACHINE(spapr);
595     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
596     hwaddr mem_start, node_size;
597     int i, nb_nodes = machine->numa_state->num_nodes;
598     NodeInfo *nodes = machine->numa_state->nodes;
599 
600     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
601         if (!nodes[i].node_mem) {
602             continue;
603         }
604         if (mem_start >= machine->ram_size) {
605             node_size = 0;
606         } else {
607             node_size = nodes[i].node_mem;
608             if (node_size > machine->ram_size - mem_start) {
609                 node_size = machine->ram_size - mem_start;
610             }
611         }
612         if (!mem_start) {
613             /* spapr_machine_init() checks for rma_size <= node0_size
614              * already */
615             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
616             mem_start += spapr->rma_size;
617             node_size -= spapr->rma_size;
618         }
619         for ( ; node_size; ) {
620             hwaddr sizetmp = pow2floor(node_size);
621 
622             /* mem_start != 0 here */
623             if (ctzl(mem_start) < ctzl(sizetmp)) {
624                 sizetmp = 1ULL << ctzl(mem_start);
625             }
626 
627             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
628             node_size -= sizetmp;
629             mem_start += sizetmp;
630         }
631     }
632 
633     /* Generate ibm,dynamic-reconfiguration-memory node if required */
634     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
635         int ret;
636 
637         g_assert(smc->dr_lmb_enabled);
638         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
639         if (ret) {
640             return ret;
641         }
642     }
643 
644     return 0;
645 }
646 
647 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
648                          SpaprMachineState *spapr)
649 {
650     MachineState *ms = MACHINE(spapr);
651     PowerPCCPU *cpu = POWERPC_CPU(cs);
652     CPUPPCState *env = &cpu->env;
653     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
654     int index = spapr_get_vcpu_id(cpu);
655     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
656                        0xffffffff, 0xffffffff};
657     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
658         : SPAPR_TIMEBASE_FREQ;
659     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
660     uint32_t page_sizes_prop[64];
661     size_t page_sizes_prop_size;
662     unsigned int smp_threads = ms->smp.threads;
663     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
664     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
665     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
666     SpaprDrc *drc;
667     int drc_index;
668     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
669     int i;
670 
671     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
672     if (drc) {
673         drc_index = spapr_drc_index(drc);
674         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
675     }
676 
677     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
678     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
679 
680     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
681     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
682                            env->dcache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
684                            env->dcache_line_size)));
685     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
686                            env->icache_line_size)));
687     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
688                            env->icache_line_size)));
689 
690     if (pcc->l1_dcache_size) {
691         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
692                                pcc->l1_dcache_size)));
693     } else {
694         warn_report("Unknown L1 dcache size for cpu");
695     }
696     if (pcc->l1_icache_size) {
697         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
698                                pcc->l1_icache_size)));
699     } else {
700         warn_report("Unknown L1 icache size for cpu");
701     }
702 
703     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
704     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
705     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
706     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
707     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
708     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
709 
710     if (ppc_has_spr(cpu, SPR_PURR)) {
711         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
712     }
713     if (ppc_has_spr(cpu, SPR_PURR)) {
714         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
715     }
716 
717     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
718         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
719                           segs, sizeof(segs))));
720     }
721 
722     /* Advertise VSX (vector extensions) if available
723      *   1               == VMX / Altivec available
724      *   2               == VSX available
725      *
726      * Only CPUs for which we create core types in spapr_cpu_core.c
727      * are possible, and all of those have VMX */
728     if (env->insns_flags & PPC_ALTIVEC) {
729         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
730             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
731         } else {
732             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
733         }
734     }
735 
736     /* Advertise DFP (Decimal Floating Point) if available
737      *   0 / no property == no DFP
738      *   1               == DFP available */
739     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
740         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
741     }
742 
743     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
744                                                       sizeof(page_sizes_prop));
745     if (page_sizes_prop_size) {
746         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
747                           page_sizes_prop, page_sizes_prop_size)));
748     }
749 
750     spapr_dt_pa_features(spapr, cpu, fdt, offset);
751 
752     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
753                            cs->cpu_index / vcpus_per_socket)));
754 
755     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
756                       pft_size_prop, sizeof(pft_size_prop))));
757 
758     if (ms->numa_state->num_nodes > 1) {
759         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
760     }
761 
762     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
763 
764     if (pcc->radix_page_info) {
765         for (i = 0; i < pcc->radix_page_info->count; i++) {
766             radix_AP_encodings[i] =
767                 cpu_to_be32(pcc->radix_page_info->entries[i]);
768         }
769         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
770                           radix_AP_encodings,
771                           pcc->radix_page_info->count *
772                           sizeof(radix_AP_encodings[0]))));
773     }
774 
775     /*
776      * We set this property to let the guest know that it can use the large
777      * decrementer and its width in bits.
778      */
779     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
780         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
781                               pcc->lrg_decr_bits)));
782 }
783 
784 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
785 {
786     CPUState **rev;
787     CPUState *cs;
788     int n_cpus;
789     int cpus_offset;
790     int i;
791 
792     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
793     _FDT(cpus_offset);
794     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
795     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
796 
797     /*
798      * We walk the CPUs in reverse order to ensure that CPU DT nodes
799      * created by fdt_add_subnode() end up in the right order in FDT
800      * for the guest kernel the enumerate the CPUs correctly.
801      *
802      * The CPU list cannot be traversed in reverse order, so we need
803      * to do extra work.
804      */
805     n_cpus = 0;
806     rev = NULL;
807     CPU_FOREACH(cs) {
808         rev = g_renew(CPUState *, rev, n_cpus + 1);
809         rev[n_cpus++] = cs;
810     }
811 
812     for (i = n_cpus - 1; i >= 0; i--) {
813         CPUState *cs = rev[i];
814         PowerPCCPU *cpu = POWERPC_CPU(cs);
815         int index = spapr_get_vcpu_id(cpu);
816         DeviceClass *dc = DEVICE_GET_CLASS(cs);
817         g_autofree char *nodename = NULL;
818         int offset;
819 
820         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
821             continue;
822         }
823 
824         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
825         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
826         _FDT(offset);
827         spapr_dt_cpu(cs, fdt, offset, spapr);
828     }
829 
830     g_free(rev);
831 }
832 
833 static int spapr_dt_rng(void *fdt)
834 {
835     int node;
836     int ret;
837 
838     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
839     if (node <= 0) {
840         return -1;
841     }
842     ret = fdt_setprop_string(fdt, node, "device_type",
843                              "ibm,platform-facilities");
844     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
845     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
846 
847     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
848     if (node <= 0) {
849         return -1;
850     }
851     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
852 
853     return ret ? -1 : 0;
854 }
855 
856 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
857 {
858     MachineState *ms = MACHINE(spapr);
859     int rtas;
860     GString *hypertas = g_string_sized_new(256);
861     GString *qemu_hypertas = g_string_sized_new(256);
862     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
863         memory_region_size(&MACHINE(spapr)->device_memory->mr);
864     uint32_t lrdr_capacity[] = {
865         cpu_to_be32(max_device_addr >> 32),
866         cpu_to_be32(max_device_addr & 0xffffffff),
867         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
868         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
869         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
870     };
871 
872     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
873 
874     /* hypertas */
875     add_str(hypertas, "hcall-pft");
876     add_str(hypertas, "hcall-term");
877     add_str(hypertas, "hcall-dabr");
878     add_str(hypertas, "hcall-interrupt");
879     add_str(hypertas, "hcall-tce");
880     add_str(hypertas, "hcall-vio");
881     add_str(hypertas, "hcall-splpar");
882     add_str(hypertas, "hcall-join");
883     add_str(hypertas, "hcall-bulk");
884     add_str(hypertas, "hcall-set-mode");
885     add_str(hypertas, "hcall-sprg0");
886     add_str(hypertas, "hcall-copy");
887     add_str(hypertas, "hcall-debug");
888     add_str(hypertas, "hcall-vphn");
889     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
890         add_str(hypertas, "hcall-rpt-invalidate");
891     }
892 
893     add_str(qemu_hypertas, "hcall-memop1");
894 
895     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
896         add_str(hypertas, "hcall-multi-tce");
897     }
898 
899     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
900         add_str(hypertas, "hcall-hpt-resize");
901     }
902 
903     add_str(hypertas, "hcall-watchdog");
904 
905     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
906                      hypertas->str, hypertas->len));
907     g_string_free(hypertas, TRUE);
908     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
909                      qemu_hypertas->str, qemu_hypertas->len));
910     g_string_free(qemu_hypertas, TRUE);
911 
912     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
913 
914     /*
915      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
916      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
917      *
918      * The system reset requirements are driven by existing Linux and PowerVM
919      * implementation which (contrary to PAPR) saves r3 in the error log
920      * structure like machine check, so Linux expects to find the saved r3
921      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
922      * does not look at the error value).
923      *
924      * System reset interrupts are not subject to interlock like machine
925      * check, so this memory area could be corrupted if the sreset is
926      * interrupted by a machine check (or vice versa) if it was shared. To
927      * prevent this, system reset uses per-CPU areas for the sreset save
928      * area. A system reset that interrupts a system reset handler could
929      * still overwrite this area, but Linux doesn't try to recover in that
930      * case anyway.
931      *
932      * The extra 8 bytes is required because Linux's FWNMI error log check
933      * is off-by-one.
934      *
935      * RTAS_MIN_SIZE is required for the RTAS blob itself.
936      */
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
938                           RTAS_ERROR_LOG_MAX +
939                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
940                           sizeof(uint64_t)));
941     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
942                           RTAS_ERROR_LOG_MAX));
943     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
944                           RTAS_EVENT_SCAN_RATE));
945 
946     g_assert(msi_nonbroken);
947     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
948 
949     /*
950      * According to PAPR, rtas ibm,os-term does not guarantee a return
951      * back to the guest cpu.
952      *
953      * While an additional ibm,extended-os-term property indicates
954      * that rtas call return will always occur. Set this property.
955      */
956     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
957 
958     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
959                      lrdr_capacity, sizeof(lrdr_capacity)));
960 
961     spapr_dt_rtas_tokens(fdt, rtas);
962 }
963 
964 /*
965  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
966  * and the XIVE features that the guest may request and thus the valid
967  * values for bytes 23..26 of option vector 5:
968  */
969 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
970                                           int chosen)
971 {
972     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
973 
974     char val[2 * 4] = {
975         23, 0x00, /* XICS / XIVE mode */
976         24, 0x00, /* Hash/Radix, filled in below. */
977         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
978         26, 0x40, /* Radix options: GTSE == yes. */
979     };
980 
981     if (spapr->irq->xics && spapr->irq->xive) {
982         val[1] = SPAPR_OV5_XIVE_BOTH;
983     } else if (spapr->irq->xive) {
984         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
985     } else {
986         assert(spapr->irq->xics);
987         val[1] = SPAPR_OV5_XIVE_LEGACY;
988     }
989 
990     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
991                           first_ppc_cpu->compat_pvr)) {
992         /*
993          * If we're in a pre POWER9 compat mode then the guest should
994          * do hash and use the legacy interrupt mode
995          */
996         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
997         val[3] = 0x00; /* Hash */
998         spapr_check_mmu_mode(false);
999     } else if (kvm_enabled()) {
1000         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1001             val[3] = 0x80; /* OV5_MMU_BOTH */
1002         } else if (kvmppc_has_cap_mmu_radix()) {
1003             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1004         } else {
1005             val[3] = 0x00; /* Hash */
1006         }
1007     } else {
1008         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1009         val[3] = 0xC0;
1010     }
1011     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1012                      val, sizeof(val)));
1013 }
1014 
1015 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1016 {
1017     MachineState *machine = MACHINE(spapr);
1018     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1019     uint8_t rng_seed[32];
1020     int chosen;
1021 
1022     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1023 
1024     if (reset) {
1025         const char *boot_device = spapr->boot_device;
1026         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1027         size_t cb = 0;
1028         g_autofree char *bootlist = get_boot_devices_list(&cb);
1029 
1030         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1031             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1032                                     machine->kernel_cmdline));
1033         }
1034 
1035         if (spapr->initrd_size) {
1036             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1037                                   spapr->initrd_base));
1038             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1039                                   spapr->initrd_base + spapr->initrd_size));
1040         }
1041 
1042         if (spapr->kernel_size) {
1043             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1044                                   cpu_to_be64(spapr->kernel_size) };
1045 
1046             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1047                          &kprop, sizeof(kprop)));
1048             if (spapr->kernel_le) {
1049                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1050             }
1051         }
1052         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1053             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1054         }
1055         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1056         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1057         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1058 
1059         if (cb && bootlist) {
1060             int i;
1061 
1062             for (i = 0; i < cb; i++) {
1063                 if (bootlist[i] == '\n') {
1064                     bootlist[i] = ' ';
1065                 }
1066             }
1067             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1068         }
1069 
1070         if (boot_device && strlen(boot_device)) {
1071             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1072         }
1073 
1074         if (spapr->want_stdout_path && stdout_path) {
1075             /*
1076              * "linux,stdout-path" and "stdout" properties are
1077              * deprecated by linux kernel. New platforms should only
1078              * use the "stdout-path" property. Set the new property
1079              * and continue using older property to remain compatible
1080              * with the existing firmware.
1081              */
1082             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1083             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1084         }
1085 
1086         /*
1087          * We can deal with BAR reallocation just fine, advertise it
1088          * to the guest
1089          */
1090         if (smc->linux_pci_probe) {
1091             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1092         }
1093 
1094         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1095     }
1096 
1097     qemu_guest_getrandom_nofail(rng_seed, sizeof(rng_seed));
1098     _FDT(fdt_setprop(fdt, chosen, "rng-seed", rng_seed, sizeof(rng_seed)));
1099 
1100     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1101 }
1102 
1103 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1104 {
1105     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1106      * KVM to work under pHyp with some guest co-operation */
1107     int hypervisor;
1108     uint8_t hypercall[16];
1109 
1110     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1111     /* indicate KVM hypercall interface */
1112     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1113     if (kvmppc_has_cap_fixup_hcalls()) {
1114         /*
1115          * Older KVM versions with older guest kernels were broken
1116          * with the magic page, don't allow the guest to map it.
1117          */
1118         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1119                                   sizeof(hypercall))) {
1120             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1121                              hypercall, sizeof(hypercall)));
1122         }
1123     }
1124 }
1125 
1126 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1127 {
1128     MachineState *machine = MACHINE(spapr);
1129     MachineClass *mc = MACHINE_GET_CLASS(machine);
1130     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1131     uint32_t root_drc_type_mask = 0;
1132     int ret;
1133     void *fdt;
1134     SpaprPhbState *phb;
1135     char *buf;
1136 
1137     fdt = g_malloc0(space);
1138     _FDT((fdt_create_empty_tree(fdt, space)));
1139 
1140     /* Root node */
1141     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1142     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1143     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1144 
1145     /* Guest UUID & Name*/
1146     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1147     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1148     if (qemu_uuid_set) {
1149         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1150     }
1151     g_free(buf);
1152 
1153     if (qemu_get_vm_name()) {
1154         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1155                                 qemu_get_vm_name()));
1156     }
1157 
1158     /* Host Model & Serial Number */
1159     if (spapr->host_model) {
1160         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1161     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1163         g_free(buf);
1164     }
1165 
1166     if (spapr->host_serial) {
1167         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1168     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1169         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1170         g_free(buf);
1171     }
1172 
1173     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1174     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1175 
1176     /* /interrupt controller */
1177     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1178 
1179     ret = spapr_dt_memory(spapr, fdt);
1180     if (ret < 0) {
1181         error_report("couldn't setup memory nodes in fdt");
1182         exit(1);
1183     }
1184 
1185     /* /vdevice */
1186     spapr_dt_vdevice(spapr->vio_bus, fdt);
1187 
1188     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1189         ret = spapr_dt_rng(fdt);
1190         if (ret < 0) {
1191             error_report("could not set up rng device in the fdt");
1192             exit(1);
1193         }
1194     }
1195 
1196     QLIST_FOREACH(phb, &spapr->phbs, list) {
1197         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1198         if (ret < 0) {
1199             error_report("couldn't setup PCI devices in fdt");
1200             exit(1);
1201         }
1202     }
1203 
1204     spapr_dt_cpus(fdt, spapr);
1205 
1206     /* ibm,drc-indexes and friends */
1207     if (smc->dr_lmb_enabled) {
1208         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1209     }
1210     if (smc->dr_phb_enabled) {
1211         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1212     }
1213     if (mc->nvdimm_supported) {
1214         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1215     }
1216     if (root_drc_type_mask) {
1217         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1218     }
1219 
1220     if (mc->has_hotpluggable_cpus) {
1221         int offset = fdt_path_offset(fdt, "/cpus");
1222         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1223         if (ret < 0) {
1224             error_report("Couldn't set up CPU DR device tree properties");
1225             exit(1);
1226         }
1227     }
1228 
1229     /* /event-sources */
1230     spapr_dt_events(spapr, fdt);
1231 
1232     /* /rtas */
1233     spapr_dt_rtas(spapr, fdt);
1234 
1235     /* /chosen */
1236     spapr_dt_chosen(spapr, fdt, reset);
1237 
1238     /* /hypervisor */
1239     if (kvm_enabled()) {
1240         spapr_dt_hypervisor(spapr, fdt);
1241     }
1242 
1243     /* Build memory reserve map */
1244     if (reset) {
1245         if (spapr->kernel_size) {
1246             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1247                                   spapr->kernel_size)));
1248         }
1249         if (spapr->initrd_size) {
1250             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1251                                   spapr->initrd_size)));
1252         }
1253     }
1254 
1255     /* NVDIMM devices */
1256     if (mc->nvdimm_supported) {
1257         spapr_dt_persistent_memory(spapr, fdt);
1258     }
1259 
1260     return fdt;
1261 }
1262 
1263 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1264 {
1265     SpaprMachineState *spapr = opaque;
1266 
1267     return (addr & 0x0fffffff) + spapr->kernel_addr;
1268 }
1269 
1270 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1271                                     PowerPCCPU *cpu)
1272 {
1273     CPUPPCState *env = &cpu->env;
1274 
1275     /* The TCG path should also be holding the BQL at this point */
1276     g_assert(qemu_mutex_iothread_locked());
1277 
1278     g_assert(!vhyp_cpu_in_nested(cpu));
1279 
1280     if (FIELD_EX64(env->msr, MSR, PR)) {
1281         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1282         env->gpr[3] = H_PRIVILEGE;
1283     } else {
1284         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1285     }
1286 }
1287 
1288 struct LPCRSyncState {
1289     target_ulong value;
1290     target_ulong mask;
1291 };
1292 
1293 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1294 {
1295     struct LPCRSyncState *s = arg.host_ptr;
1296     PowerPCCPU *cpu = POWERPC_CPU(cs);
1297     CPUPPCState *env = &cpu->env;
1298     target_ulong lpcr;
1299 
1300     cpu_synchronize_state(cs);
1301     lpcr = env->spr[SPR_LPCR];
1302     lpcr &= ~s->mask;
1303     lpcr |= s->value;
1304     ppc_store_lpcr(cpu, lpcr);
1305 }
1306 
1307 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1308 {
1309     CPUState *cs;
1310     struct LPCRSyncState s = {
1311         .value = value,
1312         .mask = mask
1313     };
1314     CPU_FOREACH(cs) {
1315         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1316     }
1317 }
1318 
1319 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1320                            target_ulong lpid, ppc_v3_pate_t *entry)
1321 {
1322     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1323     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1324 
1325     if (!spapr_cpu->in_nested) {
1326         assert(lpid == 0);
1327 
1328         /* Copy PATE1:GR into PATE0:HR */
1329         entry->dw0 = spapr->patb_entry & PATE0_HR;
1330         entry->dw1 = spapr->patb_entry;
1331 
1332     } else {
1333         uint64_t patb, pats;
1334 
1335         assert(lpid != 0);
1336 
1337         patb = spapr->nested_ptcr & PTCR_PATB;
1338         pats = spapr->nested_ptcr & PTCR_PATS;
1339 
1340         /* Check if partition table is properly aligned */
1341         if (patb & MAKE_64BIT_MASK(0, pats + 12)) {
1342             return false;
1343         }
1344 
1345         /* Calculate number of entries */
1346         pats = 1ull << (pats + 12 - 4);
1347         if (pats <= lpid) {
1348             return false;
1349         }
1350 
1351         /* Grab entry */
1352         patb += 16 * lpid;
1353         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1354         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1355     }
1356 
1357     return true;
1358 }
1359 
1360 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1361 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1362 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1363 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1364 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1365 
1366 /*
1367  * Get the fd to access the kernel htab, re-opening it if necessary
1368  */
1369 static int get_htab_fd(SpaprMachineState *spapr)
1370 {
1371     Error *local_err = NULL;
1372 
1373     if (spapr->htab_fd >= 0) {
1374         return spapr->htab_fd;
1375     }
1376 
1377     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1378     if (spapr->htab_fd < 0) {
1379         error_report_err(local_err);
1380     }
1381 
1382     return spapr->htab_fd;
1383 }
1384 
1385 void close_htab_fd(SpaprMachineState *spapr)
1386 {
1387     if (spapr->htab_fd >= 0) {
1388         close(spapr->htab_fd);
1389     }
1390     spapr->htab_fd = -1;
1391 }
1392 
1393 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1394 {
1395     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1396 
1397     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1398 }
1399 
1400 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1401 {
1402     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1403 
1404     assert(kvm_enabled());
1405 
1406     if (!spapr->htab) {
1407         return 0;
1408     }
1409 
1410     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1411 }
1412 
1413 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1414                                                 hwaddr ptex, int n)
1415 {
1416     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1417     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1418 
1419     if (!spapr->htab) {
1420         /*
1421          * HTAB is controlled by KVM. Fetch into temporary buffer
1422          */
1423         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1424         kvmppc_read_hptes(hptes, ptex, n);
1425         return hptes;
1426     }
1427 
1428     /*
1429      * HTAB is controlled by QEMU. Just point to the internally
1430      * accessible PTEG.
1431      */
1432     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1433 }
1434 
1435 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1436                               const ppc_hash_pte64_t *hptes,
1437                               hwaddr ptex, int n)
1438 {
1439     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1440 
1441     if (!spapr->htab) {
1442         g_free((void *)hptes);
1443     }
1444 
1445     /* Nothing to do for qemu managed HPT */
1446 }
1447 
1448 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1449                       uint64_t pte0, uint64_t pte1)
1450 {
1451     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1452     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1453 
1454     if (!spapr->htab) {
1455         kvmppc_write_hpte(ptex, pte0, pte1);
1456     } else {
1457         if (pte0 & HPTE64_V_VALID) {
1458             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1459             /*
1460              * When setting valid, we write PTE1 first. This ensures
1461              * proper synchronization with the reading code in
1462              * ppc_hash64_pteg_search()
1463              */
1464             smp_wmb();
1465             stq_p(spapr->htab + offset, pte0);
1466         } else {
1467             stq_p(spapr->htab + offset, pte0);
1468             /*
1469              * When clearing it we set PTE0 first. This ensures proper
1470              * synchronization with the reading code in
1471              * ppc_hash64_pteg_search()
1472              */
1473             smp_wmb();
1474             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1475         }
1476     }
1477 }
1478 
1479 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1480                              uint64_t pte1)
1481 {
1482     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1483     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1484 
1485     if (!spapr->htab) {
1486         /* There should always be a hash table when this is called */
1487         error_report("spapr_hpte_set_c called with no hash table !");
1488         return;
1489     }
1490 
1491     /* The HW performs a non-atomic byte update */
1492     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1493 }
1494 
1495 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1496                              uint64_t pte1)
1497 {
1498     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1499     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1500 
1501     if (!spapr->htab) {
1502         /* There should always be a hash table when this is called */
1503         error_report("spapr_hpte_set_r called with no hash table !");
1504         return;
1505     }
1506 
1507     /* The HW performs a non-atomic byte update */
1508     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1509 }
1510 
1511 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1512 {
1513     int shift;
1514 
1515     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1516      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1517      * that's much more than is needed for Linux guests */
1518     shift = ctz64(pow2ceil(ramsize)) - 7;
1519     shift = MAX(shift, 18); /* Minimum architected size */
1520     shift = MIN(shift, 46); /* Maximum architected size */
1521     return shift;
1522 }
1523 
1524 void spapr_free_hpt(SpaprMachineState *spapr)
1525 {
1526     qemu_vfree(spapr->htab);
1527     spapr->htab = NULL;
1528     spapr->htab_shift = 0;
1529     close_htab_fd(spapr);
1530 }
1531 
1532 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1533 {
1534     ERRP_GUARD();
1535     long rc;
1536 
1537     /* Clean up any HPT info from a previous boot */
1538     spapr_free_hpt(spapr);
1539 
1540     rc = kvmppc_reset_htab(shift);
1541 
1542     if (rc == -EOPNOTSUPP) {
1543         error_setg(errp, "HPT not supported in nested guests");
1544         return -EOPNOTSUPP;
1545     }
1546 
1547     if (rc < 0) {
1548         /* kernel-side HPT needed, but couldn't allocate one */
1549         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1550                          shift);
1551         error_append_hint(errp, "Try smaller maxmem?\n");
1552         return -errno;
1553     } else if (rc > 0) {
1554         /* kernel-side HPT allocated */
1555         if (rc != shift) {
1556             error_setg(errp,
1557                        "Requested order %d HPT, but kernel allocated order %ld",
1558                        shift, rc);
1559             error_append_hint(errp, "Try smaller maxmem?\n");
1560             return -ENOSPC;
1561         }
1562 
1563         spapr->htab_shift = shift;
1564         spapr->htab = NULL;
1565     } else {
1566         /* kernel-side HPT not needed, allocate in userspace instead */
1567         size_t size = 1ULL << shift;
1568         int i;
1569 
1570         spapr->htab = qemu_memalign(size, size);
1571         memset(spapr->htab, 0, size);
1572         spapr->htab_shift = shift;
1573 
1574         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1575             DIRTY_HPTE(HPTE(spapr->htab, i));
1576         }
1577     }
1578     /* We're setting up a hash table, so that means we're not radix */
1579     spapr->patb_entry = 0;
1580     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1581     return 0;
1582 }
1583 
1584 void spapr_setup_hpt(SpaprMachineState *spapr)
1585 {
1586     int hpt_shift;
1587 
1588     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1589         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1590     } else {
1591         uint64_t current_ram_size;
1592 
1593         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1594         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1595     }
1596     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1597 
1598     if (kvm_enabled()) {
1599         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1600 
1601         /* Check our RMA fits in the possible VRMA */
1602         if (vrma_limit < spapr->rma_size) {
1603             error_report("Unable to create %" HWADDR_PRIu
1604                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1605                          spapr->rma_size / MiB, vrma_limit / MiB);
1606             exit(EXIT_FAILURE);
1607         }
1608     }
1609 }
1610 
1611 void spapr_check_mmu_mode(bool guest_radix)
1612 {
1613     if (guest_radix) {
1614         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1615             error_report("Guest requested unavailable MMU mode (radix).");
1616             exit(EXIT_FAILURE);
1617         }
1618     } else {
1619         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1620             && !kvmppc_has_cap_mmu_hash_v3()) {
1621             error_report("Guest requested unavailable MMU mode (hash).");
1622             exit(EXIT_FAILURE);
1623         }
1624     }
1625 }
1626 
1627 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1628 {
1629     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1630     PowerPCCPU *first_ppc_cpu;
1631     hwaddr fdt_addr;
1632     void *fdt;
1633     int rc;
1634 
1635     pef_kvm_reset(machine->cgs, &error_fatal);
1636     spapr_caps_apply(spapr);
1637 
1638     first_ppc_cpu = POWERPC_CPU(first_cpu);
1639     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1640         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1641                               spapr->max_compat_pvr)) {
1642         /*
1643          * If using KVM with radix mode available, VCPUs can be started
1644          * without a HPT because KVM will start them in radix mode.
1645          * Set the GR bit in PATE so that we know there is no HPT.
1646          */
1647         spapr->patb_entry = PATE1_GR;
1648         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1649     } else {
1650         spapr_setup_hpt(spapr);
1651     }
1652 
1653     qemu_devices_reset(reason);
1654 
1655     spapr_ovec_cleanup(spapr->ov5_cas);
1656     spapr->ov5_cas = spapr_ovec_new();
1657 
1658     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1659 
1660     /*
1661      * This is fixing some of the default configuration of the XIVE
1662      * devices. To be called after the reset of the machine devices.
1663      */
1664     spapr_irq_reset(spapr, &error_fatal);
1665 
1666     /*
1667      * There is no CAS under qtest. Simulate one to please the code that
1668      * depends on spapr->ov5_cas. This is especially needed to test device
1669      * unplug, so we do that before resetting the DRCs.
1670      */
1671     if (qtest_enabled()) {
1672         spapr_ovec_cleanup(spapr->ov5_cas);
1673         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1674     }
1675 
1676     spapr_nvdimm_finish_flushes();
1677 
1678     /* DRC reset may cause a device to be unplugged. This will cause troubles
1679      * if this device is used by another device (eg, a running vhost backend
1680      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1681      * situations, we reset DRCs after all devices have been reset.
1682      */
1683     spapr_drc_reset_all(spapr);
1684 
1685     spapr_clear_pending_events(spapr);
1686 
1687     /*
1688      * We place the device tree just below either the top of the RMA,
1689      * or just below 2GB, whichever is lower, so that it can be
1690      * processed with 32-bit real mode code if necessary
1691      */
1692     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1693 
1694     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1695     if (spapr->vof) {
1696         spapr_vof_reset(spapr, fdt, &error_fatal);
1697         /*
1698          * Do not pack the FDT as the client may change properties.
1699          * VOF client does not expect the FDT so we do not load it to the VM.
1700          */
1701     } else {
1702         rc = fdt_pack(fdt);
1703         /* Should only fail if we've built a corrupted tree */
1704         assert(rc == 0);
1705 
1706         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1707                                   0, fdt_addr, 0);
1708         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1709     }
1710     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1711 
1712     g_free(spapr->fdt_blob);
1713     spapr->fdt_size = fdt_totalsize(fdt);
1714     spapr->fdt_initial_size = spapr->fdt_size;
1715     spapr->fdt_blob = fdt;
1716 
1717     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1718     machine->fdt = fdt;
1719 
1720     /* Set up the entry state */
1721     first_ppc_cpu->env.gpr[5] = 0;
1722 
1723     spapr->fwnmi_system_reset_addr = -1;
1724     spapr->fwnmi_machine_check_addr = -1;
1725     spapr->fwnmi_machine_check_interlock = -1;
1726 
1727     /* Signal all vCPUs waiting on this condition */
1728     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1729 
1730     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1731 }
1732 
1733 static void spapr_create_nvram(SpaprMachineState *spapr)
1734 {
1735     DeviceState *dev = qdev_new("spapr-nvram");
1736     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1737 
1738     if (dinfo) {
1739         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1740                                 &error_fatal);
1741     }
1742 
1743     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1744 
1745     spapr->nvram = (struct SpaprNvram *)dev;
1746 }
1747 
1748 static void spapr_rtc_create(SpaprMachineState *spapr)
1749 {
1750     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1751                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1752                                        &error_fatal, NULL);
1753     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1754     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1755                               "date");
1756 }
1757 
1758 /* Returns whether we want to use VGA or not */
1759 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1760 {
1761     vga_interface_created = true;
1762     switch (vga_interface_type) {
1763     case VGA_NONE:
1764         return false;
1765     case VGA_DEVICE:
1766         return true;
1767     case VGA_STD:
1768     case VGA_VIRTIO:
1769     case VGA_CIRRUS:
1770         return pci_vga_init(pci_bus) != NULL;
1771     default:
1772         error_setg(errp,
1773                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1774         return false;
1775     }
1776 }
1777 
1778 static int spapr_pre_load(void *opaque)
1779 {
1780     int rc;
1781 
1782     rc = spapr_caps_pre_load(opaque);
1783     if (rc) {
1784         return rc;
1785     }
1786 
1787     return 0;
1788 }
1789 
1790 static int spapr_post_load(void *opaque, int version_id)
1791 {
1792     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1793     int err = 0;
1794 
1795     err = spapr_caps_post_migration(spapr);
1796     if (err) {
1797         return err;
1798     }
1799 
1800     /*
1801      * In earlier versions, there was no separate qdev for the PAPR
1802      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1803      * So when migrating from those versions, poke the incoming offset
1804      * value into the RTC device
1805      */
1806     if (version_id < 3) {
1807         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1808         if (err) {
1809             return err;
1810         }
1811     }
1812 
1813     if (kvm_enabled() && spapr->patb_entry) {
1814         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1815         bool radix = !!(spapr->patb_entry & PATE1_GR);
1816         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1817 
1818         /*
1819          * Update LPCR:HR and UPRT as they may not be set properly in
1820          * the stream
1821          */
1822         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1823                             LPCR_HR | LPCR_UPRT);
1824 
1825         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1826         if (err) {
1827             error_report("Process table config unsupported by the host");
1828             return -EINVAL;
1829         }
1830     }
1831 
1832     err = spapr_irq_post_load(spapr, version_id);
1833     if (err) {
1834         return err;
1835     }
1836 
1837     return err;
1838 }
1839 
1840 static int spapr_pre_save(void *opaque)
1841 {
1842     int rc;
1843 
1844     rc = spapr_caps_pre_save(opaque);
1845     if (rc) {
1846         return rc;
1847     }
1848 
1849     return 0;
1850 }
1851 
1852 static bool version_before_3(void *opaque, int version_id)
1853 {
1854     return version_id < 3;
1855 }
1856 
1857 static bool spapr_pending_events_needed(void *opaque)
1858 {
1859     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1860     return !QTAILQ_EMPTY(&spapr->pending_events);
1861 }
1862 
1863 static const VMStateDescription vmstate_spapr_event_entry = {
1864     .name = "spapr_event_log_entry",
1865     .version_id = 1,
1866     .minimum_version_id = 1,
1867     .fields = (VMStateField[]) {
1868         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1869         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1870         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1871                                      NULL, extended_length),
1872         VMSTATE_END_OF_LIST()
1873     },
1874 };
1875 
1876 static const VMStateDescription vmstate_spapr_pending_events = {
1877     .name = "spapr_pending_events",
1878     .version_id = 1,
1879     .minimum_version_id = 1,
1880     .needed = spapr_pending_events_needed,
1881     .fields = (VMStateField[]) {
1882         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1883                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1884         VMSTATE_END_OF_LIST()
1885     },
1886 };
1887 
1888 static bool spapr_ov5_cas_needed(void *opaque)
1889 {
1890     SpaprMachineState *spapr = opaque;
1891     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1892     bool cas_needed;
1893 
1894     /* Prior to the introduction of SpaprOptionVector, we had two option
1895      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1896      * Both of these options encode machine topology into the device-tree
1897      * in such a way that the now-booted OS should still be able to interact
1898      * appropriately with QEMU regardless of what options were actually
1899      * negotiatied on the source side.
1900      *
1901      * As such, we can avoid migrating the CAS-negotiated options if these
1902      * are the only options available on the current machine/platform.
1903      * Since these are the only options available for pseries-2.7 and
1904      * earlier, this allows us to maintain old->new/new->old migration
1905      * compatibility.
1906      *
1907      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1908      * via default pseries-2.8 machines and explicit command-line parameters.
1909      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1910      * of the actual CAS-negotiated values to continue working properly. For
1911      * example, availability of memory unplug depends on knowing whether
1912      * OV5_HP_EVT was negotiated via CAS.
1913      *
1914      * Thus, for any cases where the set of available CAS-negotiatable
1915      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1916      * include the CAS-negotiated options in the migration stream, unless
1917      * if they affect boot time behaviour only.
1918      */
1919     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1920     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1921     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1922 
1923     /* We need extra information if we have any bits outside the mask
1924      * defined above */
1925     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1926 
1927     spapr_ovec_cleanup(ov5_mask);
1928 
1929     return cas_needed;
1930 }
1931 
1932 static const VMStateDescription vmstate_spapr_ov5_cas = {
1933     .name = "spapr_option_vector_ov5_cas",
1934     .version_id = 1,
1935     .minimum_version_id = 1,
1936     .needed = spapr_ov5_cas_needed,
1937     .fields = (VMStateField[]) {
1938         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1939                                  vmstate_spapr_ovec, SpaprOptionVector),
1940         VMSTATE_END_OF_LIST()
1941     },
1942 };
1943 
1944 static bool spapr_patb_entry_needed(void *opaque)
1945 {
1946     SpaprMachineState *spapr = opaque;
1947 
1948     return !!spapr->patb_entry;
1949 }
1950 
1951 static const VMStateDescription vmstate_spapr_patb_entry = {
1952     .name = "spapr_patb_entry",
1953     .version_id = 1,
1954     .minimum_version_id = 1,
1955     .needed = spapr_patb_entry_needed,
1956     .fields = (VMStateField[]) {
1957         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1958         VMSTATE_END_OF_LIST()
1959     },
1960 };
1961 
1962 static bool spapr_irq_map_needed(void *opaque)
1963 {
1964     SpaprMachineState *spapr = opaque;
1965 
1966     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1967 }
1968 
1969 static const VMStateDescription vmstate_spapr_irq_map = {
1970     .name = "spapr_irq_map",
1971     .version_id = 1,
1972     .minimum_version_id = 1,
1973     .needed = spapr_irq_map_needed,
1974     .fields = (VMStateField[]) {
1975         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1976         VMSTATE_END_OF_LIST()
1977     },
1978 };
1979 
1980 static bool spapr_dtb_needed(void *opaque)
1981 {
1982     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1983 
1984     return smc->update_dt_enabled;
1985 }
1986 
1987 static int spapr_dtb_pre_load(void *opaque)
1988 {
1989     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1990 
1991     g_free(spapr->fdt_blob);
1992     spapr->fdt_blob = NULL;
1993     spapr->fdt_size = 0;
1994 
1995     return 0;
1996 }
1997 
1998 static const VMStateDescription vmstate_spapr_dtb = {
1999     .name = "spapr_dtb",
2000     .version_id = 1,
2001     .minimum_version_id = 1,
2002     .needed = spapr_dtb_needed,
2003     .pre_load = spapr_dtb_pre_load,
2004     .fields = (VMStateField[]) {
2005         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2006         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2007         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2008                                      fdt_size),
2009         VMSTATE_END_OF_LIST()
2010     },
2011 };
2012 
2013 static bool spapr_fwnmi_needed(void *opaque)
2014 {
2015     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2016 
2017     return spapr->fwnmi_machine_check_addr != -1;
2018 }
2019 
2020 static int spapr_fwnmi_pre_save(void *opaque)
2021 {
2022     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2023 
2024     /*
2025      * Check if machine check handling is in progress and print a
2026      * warning message.
2027      */
2028     if (spapr->fwnmi_machine_check_interlock != -1) {
2029         warn_report("A machine check is being handled during migration. The"
2030                 "handler may run and log hardware error on the destination");
2031     }
2032 
2033     return 0;
2034 }
2035 
2036 static const VMStateDescription vmstate_spapr_fwnmi = {
2037     .name = "spapr_fwnmi",
2038     .version_id = 1,
2039     .minimum_version_id = 1,
2040     .needed = spapr_fwnmi_needed,
2041     .pre_save = spapr_fwnmi_pre_save,
2042     .fields = (VMStateField[]) {
2043         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2044         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2045         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2046         VMSTATE_END_OF_LIST()
2047     },
2048 };
2049 
2050 static const VMStateDescription vmstate_spapr = {
2051     .name = "spapr",
2052     .version_id = 3,
2053     .minimum_version_id = 1,
2054     .pre_load = spapr_pre_load,
2055     .post_load = spapr_post_load,
2056     .pre_save = spapr_pre_save,
2057     .fields = (VMStateField[]) {
2058         /* used to be @next_irq */
2059         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2060 
2061         /* RTC offset */
2062         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2063 
2064         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2065         VMSTATE_END_OF_LIST()
2066     },
2067     .subsections = (const VMStateDescription*[]) {
2068         &vmstate_spapr_ov5_cas,
2069         &vmstate_spapr_patb_entry,
2070         &vmstate_spapr_pending_events,
2071         &vmstate_spapr_cap_htm,
2072         &vmstate_spapr_cap_vsx,
2073         &vmstate_spapr_cap_dfp,
2074         &vmstate_spapr_cap_cfpc,
2075         &vmstate_spapr_cap_sbbc,
2076         &vmstate_spapr_cap_ibs,
2077         &vmstate_spapr_cap_hpt_maxpagesize,
2078         &vmstate_spapr_irq_map,
2079         &vmstate_spapr_cap_nested_kvm_hv,
2080         &vmstate_spapr_dtb,
2081         &vmstate_spapr_cap_large_decr,
2082         &vmstate_spapr_cap_ccf_assist,
2083         &vmstate_spapr_cap_fwnmi,
2084         &vmstate_spapr_fwnmi,
2085         &vmstate_spapr_cap_rpt_invalidate,
2086         NULL
2087     }
2088 };
2089 
2090 static int htab_save_setup(QEMUFile *f, void *opaque)
2091 {
2092     SpaprMachineState *spapr = opaque;
2093 
2094     /* "Iteration" header */
2095     if (!spapr->htab_shift) {
2096         qemu_put_be32(f, -1);
2097     } else {
2098         qemu_put_be32(f, spapr->htab_shift);
2099     }
2100 
2101     if (spapr->htab) {
2102         spapr->htab_save_index = 0;
2103         spapr->htab_first_pass = true;
2104     } else {
2105         if (spapr->htab_shift) {
2106             assert(kvm_enabled());
2107         }
2108     }
2109 
2110 
2111     return 0;
2112 }
2113 
2114 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2115                             int chunkstart, int n_valid, int n_invalid)
2116 {
2117     qemu_put_be32(f, chunkstart);
2118     qemu_put_be16(f, n_valid);
2119     qemu_put_be16(f, n_invalid);
2120     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2121                     HASH_PTE_SIZE_64 * n_valid);
2122 }
2123 
2124 static void htab_save_end_marker(QEMUFile *f)
2125 {
2126     qemu_put_be32(f, 0);
2127     qemu_put_be16(f, 0);
2128     qemu_put_be16(f, 0);
2129 }
2130 
2131 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2132                                  int64_t max_ns)
2133 {
2134     bool has_timeout = max_ns != -1;
2135     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2136     int index = spapr->htab_save_index;
2137     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2138 
2139     assert(spapr->htab_first_pass);
2140 
2141     do {
2142         int chunkstart;
2143 
2144         /* Consume invalid HPTEs */
2145         while ((index < htabslots)
2146                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2147             CLEAN_HPTE(HPTE(spapr->htab, index));
2148             index++;
2149         }
2150 
2151         /* Consume valid HPTEs */
2152         chunkstart = index;
2153         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2154                && HPTE_VALID(HPTE(spapr->htab, index))) {
2155             CLEAN_HPTE(HPTE(spapr->htab, index));
2156             index++;
2157         }
2158 
2159         if (index > chunkstart) {
2160             int n_valid = index - chunkstart;
2161 
2162             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2163 
2164             if (has_timeout &&
2165                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2166                 break;
2167             }
2168         }
2169     } while ((index < htabslots) && !migration_rate_exceeded(f));
2170 
2171     if (index >= htabslots) {
2172         assert(index == htabslots);
2173         index = 0;
2174         spapr->htab_first_pass = false;
2175     }
2176     spapr->htab_save_index = index;
2177 }
2178 
2179 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2180                                 int64_t max_ns)
2181 {
2182     bool final = max_ns < 0;
2183     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2184     int examined = 0, sent = 0;
2185     int index = spapr->htab_save_index;
2186     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2187 
2188     assert(!spapr->htab_first_pass);
2189 
2190     do {
2191         int chunkstart, invalidstart;
2192 
2193         /* Consume non-dirty HPTEs */
2194         while ((index < htabslots)
2195                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2196             index++;
2197             examined++;
2198         }
2199 
2200         chunkstart = index;
2201         /* Consume valid dirty HPTEs */
2202         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2203                && HPTE_DIRTY(HPTE(spapr->htab, index))
2204                && HPTE_VALID(HPTE(spapr->htab, index))) {
2205             CLEAN_HPTE(HPTE(spapr->htab, index));
2206             index++;
2207             examined++;
2208         }
2209 
2210         invalidstart = index;
2211         /* Consume invalid dirty HPTEs */
2212         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2213                && HPTE_DIRTY(HPTE(spapr->htab, index))
2214                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2215             CLEAN_HPTE(HPTE(spapr->htab, index));
2216             index++;
2217             examined++;
2218         }
2219 
2220         if (index > chunkstart) {
2221             int n_valid = invalidstart - chunkstart;
2222             int n_invalid = index - invalidstart;
2223 
2224             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2225             sent += index - chunkstart;
2226 
2227             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2228                 break;
2229             }
2230         }
2231 
2232         if (examined >= htabslots) {
2233             break;
2234         }
2235 
2236         if (index >= htabslots) {
2237             assert(index == htabslots);
2238             index = 0;
2239         }
2240     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2241 
2242     if (index >= htabslots) {
2243         assert(index == htabslots);
2244         index = 0;
2245     }
2246 
2247     spapr->htab_save_index = index;
2248 
2249     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2250 }
2251 
2252 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2253 #define MAX_KVM_BUF_SIZE    2048
2254 
2255 static int htab_save_iterate(QEMUFile *f, void *opaque)
2256 {
2257     SpaprMachineState *spapr = opaque;
2258     int fd;
2259     int rc = 0;
2260 
2261     /* Iteration header */
2262     if (!spapr->htab_shift) {
2263         qemu_put_be32(f, -1);
2264         return 1;
2265     } else {
2266         qemu_put_be32(f, 0);
2267     }
2268 
2269     if (!spapr->htab) {
2270         assert(kvm_enabled());
2271 
2272         fd = get_htab_fd(spapr);
2273         if (fd < 0) {
2274             return fd;
2275         }
2276 
2277         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2278         if (rc < 0) {
2279             return rc;
2280         }
2281     } else  if (spapr->htab_first_pass) {
2282         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2283     } else {
2284         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2285     }
2286 
2287     htab_save_end_marker(f);
2288 
2289     return rc;
2290 }
2291 
2292 static int htab_save_complete(QEMUFile *f, void *opaque)
2293 {
2294     SpaprMachineState *spapr = opaque;
2295     int fd;
2296 
2297     /* Iteration header */
2298     if (!spapr->htab_shift) {
2299         qemu_put_be32(f, -1);
2300         return 0;
2301     } else {
2302         qemu_put_be32(f, 0);
2303     }
2304 
2305     if (!spapr->htab) {
2306         int rc;
2307 
2308         assert(kvm_enabled());
2309 
2310         fd = get_htab_fd(spapr);
2311         if (fd < 0) {
2312             return fd;
2313         }
2314 
2315         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2316         if (rc < 0) {
2317             return rc;
2318         }
2319     } else {
2320         if (spapr->htab_first_pass) {
2321             htab_save_first_pass(f, spapr, -1);
2322         }
2323         htab_save_later_pass(f, spapr, -1);
2324     }
2325 
2326     /* End marker */
2327     htab_save_end_marker(f);
2328 
2329     return 0;
2330 }
2331 
2332 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2333 {
2334     SpaprMachineState *spapr = opaque;
2335     uint32_t section_hdr;
2336     int fd = -1;
2337     Error *local_err = NULL;
2338 
2339     if (version_id < 1 || version_id > 1) {
2340         error_report("htab_load() bad version");
2341         return -EINVAL;
2342     }
2343 
2344     section_hdr = qemu_get_be32(f);
2345 
2346     if (section_hdr == -1) {
2347         spapr_free_hpt(spapr);
2348         return 0;
2349     }
2350 
2351     if (section_hdr) {
2352         int ret;
2353 
2354         /* First section gives the htab size */
2355         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2356         if (ret < 0) {
2357             error_report_err(local_err);
2358             return ret;
2359         }
2360         return 0;
2361     }
2362 
2363     if (!spapr->htab) {
2364         assert(kvm_enabled());
2365 
2366         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2367         if (fd < 0) {
2368             error_report_err(local_err);
2369             return fd;
2370         }
2371     }
2372 
2373     while (true) {
2374         uint32_t index;
2375         uint16_t n_valid, n_invalid;
2376 
2377         index = qemu_get_be32(f);
2378         n_valid = qemu_get_be16(f);
2379         n_invalid = qemu_get_be16(f);
2380 
2381         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2382             /* End of Stream */
2383             break;
2384         }
2385 
2386         if ((index + n_valid + n_invalid) >
2387             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2388             /* Bad index in stream */
2389             error_report(
2390                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2391                 index, n_valid, n_invalid, spapr->htab_shift);
2392             return -EINVAL;
2393         }
2394 
2395         if (spapr->htab) {
2396             if (n_valid) {
2397                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2398                                 HASH_PTE_SIZE_64 * n_valid);
2399             }
2400             if (n_invalid) {
2401                 memset(HPTE(spapr->htab, index + n_valid), 0,
2402                        HASH_PTE_SIZE_64 * n_invalid);
2403             }
2404         } else {
2405             int rc;
2406 
2407             assert(fd >= 0);
2408 
2409             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2410                                         &local_err);
2411             if (rc < 0) {
2412                 error_report_err(local_err);
2413                 return rc;
2414             }
2415         }
2416     }
2417 
2418     if (!spapr->htab) {
2419         assert(fd >= 0);
2420         close(fd);
2421     }
2422 
2423     return 0;
2424 }
2425 
2426 static void htab_save_cleanup(void *opaque)
2427 {
2428     SpaprMachineState *spapr = opaque;
2429 
2430     close_htab_fd(spapr);
2431 }
2432 
2433 static SaveVMHandlers savevm_htab_handlers = {
2434     .save_setup = htab_save_setup,
2435     .save_live_iterate = htab_save_iterate,
2436     .save_live_complete_precopy = htab_save_complete,
2437     .save_cleanup = htab_save_cleanup,
2438     .load_state = htab_load,
2439 };
2440 
2441 static void spapr_boot_set(void *opaque, const char *boot_device,
2442                            Error **errp)
2443 {
2444     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2445 
2446     g_free(spapr->boot_device);
2447     spapr->boot_device = g_strdup(boot_device);
2448 }
2449 
2450 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2451 {
2452     MachineState *machine = MACHINE(spapr);
2453     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2454     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2455     int i;
2456 
2457     for (i = 0; i < nr_lmbs; i++) {
2458         uint64_t addr;
2459 
2460         addr = i * lmb_size + machine->device_memory->base;
2461         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2462                                addr / lmb_size);
2463     }
2464 }
2465 
2466 /*
2467  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2468  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2469  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2470  */
2471 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2472 {
2473     int i;
2474 
2475     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2476         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2477                    " is not aligned to %" PRIu64 " MiB",
2478                    machine->ram_size,
2479                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2480         return;
2481     }
2482 
2483     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2484         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2485                    " is not aligned to %" PRIu64 " MiB",
2486                    machine->ram_size,
2487                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2488         return;
2489     }
2490 
2491     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2492         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2493             error_setg(errp,
2494                        "Node %d memory size 0x%" PRIx64
2495                        " is not aligned to %" PRIu64 " MiB",
2496                        i, machine->numa_state->nodes[i].node_mem,
2497                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2498             return;
2499         }
2500     }
2501 }
2502 
2503 /* find cpu slot in machine->possible_cpus by core_id */
2504 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2505 {
2506     int index = id / ms->smp.threads;
2507 
2508     if (index >= ms->possible_cpus->len) {
2509         return NULL;
2510     }
2511     if (idx) {
2512         *idx = index;
2513     }
2514     return &ms->possible_cpus->cpus[index];
2515 }
2516 
2517 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2518 {
2519     MachineState *ms = MACHINE(spapr);
2520     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2521     Error *local_err = NULL;
2522     bool vsmt_user = !!spapr->vsmt;
2523     int kvm_smt = kvmppc_smt_threads();
2524     int ret;
2525     unsigned int smp_threads = ms->smp.threads;
2526 
2527     if (!kvm_enabled() && (smp_threads > 1)) {
2528         error_setg(errp, "TCG cannot support more than 1 thread/core "
2529                    "on a pseries machine");
2530         return;
2531     }
2532     if (!is_power_of_2(smp_threads)) {
2533         error_setg(errp, "Cannot support %d threads/core on a pseries "
2534                    "machine because it must be a power of 2", smp_threads);
2535         return;
2536     }
2537 
2538     /* Detemine the VSMT mode to use: */
2539     if (vsmt_user) {
2540         if (spapr->vsmt < smp_threads) {
2541             error_setg(errp, "Cannot support VSMT mode %d"
2542                        " because it must be >= threads/core (%d)",
2543                        spapr->vsmt, smp_threads);
2544             return;
2545         }
2546         /* In this case, spapr->vsmt has been set by the command line */
2547     } else if (!smc->smp_threads_vsmt) {
2548         /*
2549          * Default VSMT value is tricky, because we need it to be as
2550          * consistent as possible (for migration), but this requires
2551          * changing it for at least some existing cases.  We pick 8 as
2552          * the value that we'd get with KVM on POWER8, the
2553          * overwhelmingly common case in production systems.
2554          */
2555         spapr->vsmt = MAX(8, smp_threads);
2556     } else {
2557         spapr->vsmt = smp_threads;
2558     }
2559 
2560     /* KVM: If necessary, set the SMT mode: */
2561     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2562         ret = kvmppc_set_smt_threads(spapr->vsmt);
2563         if (ret) {
2564             /* Looks like KVM isn't able to change VSMT mode */
2565             error_setg(&local_err,
2566                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2567                        spapr->vsmt, ret);
2568             /* We can live with that if the default one is big enough
2569              * for the number of threads, and a submultiple of the one
2570              * we want.  In this case we'll waste some vcpu ids, but
2571              * behaviour will be correct */
2572             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2573                 warn_report_err(local_err);
2574             } else {
2575                 if (!vsmt_user) {
2576                     error_append_hint(&local_err,
2577                                       "On PPC, a VM with %d threads/core"
2578                                       " on a host with %d threads/core"
2579                                       " requires the use of VSMT mode %d.\n",
2580                                       smp_threads, kvm_smt, spapr->vsmt);
2581                 }
2582                 kvmppc_error_append_smt_possible_hint(&local_err);
2583                 error_propagate(errp, local_err);
2584             }
2585         }
2586     }
2587     /* else TCG: nothing to do currently */
2588 }
2589 
2590 static void spapr_init_cpus(SpaprMachineState *spapr)
2591 {
2592     MachineState *machine = MACHINE(spapr);
2593     MachineClass *mc = MACHINE_GET_CLASS(machine);
2594     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2595     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2596     const CPUArchIdList *possible_cpus;
2597     unsigned int smp_cpus = machine->smp.cpus;
2598     unsigned int smp_threads = machine->smp.threads;
2599     unsigned int max_cpus = machine->smp.max_cpus;
2600     int boot_cores_nr = smp_cpus / smp_threads;
2601     int i;
2602 
2603     possible_cpus = mc->possible_cpu_arch_ids(machine);
2604     if (mc->has_hotpluggable_cpus) {
2605         if (smp_cpus % smp_threads) {
2606             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2607                          smp_cpus, smp_threads);
2608             exit(1);
2609         }
2610         if (max_cpus % smp_threads) {
2611             error_report("max_cpus (%u) must be multiple of threads (%u)",
2612                          max_cpus, smp_threads);
2613             exit(1);
2614         }
2615     } else {
2616         if (max_cpus != smp_cpus) {
2617             error_report("This machine version does not support CPU hotplug");
2618             exit(1);
2619         }
2620         boot_cores_nr = possible_cpus->len;
2621     }
2622 
2623     if (smc->pre_2_10_has_unused_icps) {
2624         int i;
2625 
2626         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2627             /* Dummy entries get deregistered when real ICPState objects
2628              * are registered during CPU core hotplug.
2629              */
2630             pre_2_10_vmstate_register_dummy_icp(i);
2631         }
2632     }
2633 
2634     for (i = 0; i < possible_cpus->len; i++) {
2635         int core_id = i * smp_threads;
2636 
2637         if (mc->has_hotpluggable_cpus) {
2638             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2639                                    spapr_vcpu_id(spapr, core_id));
2640         }
2641 
2642         if (i < boot_cores_nr) {
2643             Object *core  = object_new(type);
2644             int nr_threads = smp_threads;
2645 
2646             /* Handle the partially filled core for older machine types */
2647             if ((i + 1) * smp_threads >= smp_cpus) {
2648                 nr_threads = smp_cpus - i * smp_threads;
2649             }
2650 
2651             object_property_set_int(core, "nr-threads", nr_threads,
2652                                     &error_fatal);
2653             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2654                                     &error_fatal);
2655             qdev_realize(DEVICE(core), NULL, &error_fatal);
2656 
2657             object_unref(core);
2658         }
2659     }
2660 }
2661 
2662 static PCIHostState *spapr_create_default_phb(void)
2663 {
2664     DeviceState *dev;
2665 
2666     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2667     qdev_prop_set_uint32(dev, "index", 0);
2668     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2669 
2670     return PCI_HOST_BRIDGE(dev);
2671 }
2672 
2673 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2674 {
2675     MachineState *machine = MACHINE(spapr);
2676     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2677     hwaddr rma_size = machine->ram_size;
2678     hwaddr node0_size = spapr_node0_size(machine);
2679 
2680     /* RMA has to fit in the first NUMA node */
2681     rma_size = MIN(rma_size, node0_size);
2682 
2683     /*
2684      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2685      * never exceed that
2686      */
2687     rma_size = MIN(rma_size, 1 * TiB);
2688 
2689     /*
2690      * Clamp the RMA size based on machine type.  This is for
2691      * migration compatibility with older qemu versions, which limited
2692      * the RMA size for complicated and mostly bad reasons.
2693      */
2694     if (smc->rma_limit) {
2695         rma_size = MIN(rma_size, smc->rma_limit);
2696     }
2697 
2698     if (rma_size < MIN_RMA_SLOF) {
2699         error_setg(errp,
2700                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2701                    "ldMiB guest RMA (Real Mode Area memory)",
2702                    MIN_RMA_SLOF / MiB);
2703         return 0;
2704     }
2705 
2706     return rma_size;
2707 }
2708 
2709 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2710 {
2711     MachineState *machine = MACHINE(spapr);
2712     int i;
2713 
2714     for (i = 0; i < machine->ram_slots; i++) {
2715         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2716     }
2717 }
2718 
2719 /* pSeries LPAR / sPAPR hardware init */
2720 static void spapr_machine_init(MachineState *machine)
2721 {
2722     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2723     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2724     MachineClass *mc = MACHINE_GET_CLASS(machine);
2725     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2726     const char *bios_name = machine->firmware ?: bios_default;
2727     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2728     const char *kernel_filename = machine->kernel_filename;
2729     const char *initrd_filename = machine->initrd_filename;
2730     PCIHostState *phb;
2731     bool has_vga;
2732     int i;
2733     MemoryRegion *sysmem = get_system_memory();
2734     long load_limit, fw_size;
2735     Error *resize_hpt_err = NULL;
2736 
2737     if (!filename) {
2738         error_report("Could not find LPAR firmware '%s'", bios_name);
2739         exit(1);
2740     }
2741     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2742     if (fw_size <= 0) {
2743         error_report("Could not load LPAR firmware '%s'", filename);
2744         exit(1);
2745     }
2746 
2747     /*
2748      * if Secure VM (PEF) support is configured, then initialize it
2749      */
2750     pef_kvm_init(machine->cgs, &error_fatal);
2751 
2752     msi_nonbroken = true;
2753 
2754     QLIST_INIT(&spapr->phbs);
2755     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2756 
2757     /* Determine capabilities to run with */
2758     spapr_caps_init(spapr);
2759 
2760     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2761     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2762         /*
2763          * If the user explicitly requested a mode we should either
2764          * supply it, or fail completely (which we do below).  But if
2765          * it's not set explicitly, we reset our mode to something
2766          * that works
2767          */
2768         if (resize_hpt_err) {
2769             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2770             error_free(resize_hpt_err);
2771             resize_hpt_err = NULL;
2772         } else {
2773             spapr->resize_hpt = smc->resize_hpt_default;
2774         }
2775     }
2776 
2777     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2778 
2779     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2780         /*
2781          * User requested HPT resize, but this host can't supply it.  Bail out
2782          */
2783         error_report_err(resize_hpt_err);
2784         exit(1);
2785     }
2786     error_free(resize_hpt_err);
2787 
2788     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2789 
2790     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2791     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2792 
2793     /*
2794      * VSMT must be set in order to be able to compute VCPU ids, ie to
2795      * call spapr_max_server_number() or spapr_vcpu_id().
2796      */
2797     spapr_set_vsmt_mode(spapr, &error_fatal);
2798 
2799     /* Set up Interrupt Controller before we create the VCPUs */
2800     spapr_irq_init(spapr, &error_fatal);
2801 
2802     /* Set up containers for ibm,client-architecture-support negotiated options
2803      */
2804     spapr->ov5 = spapr_ovec_new();
2805     spapr->ov5_cas = spapr_ovec_new();
2806 
2807     if (smc->dr_lmb_enabled) {
2808         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2809         spapr_validate_node_memory(machine, &error_fatal);
2810     }
2811 
2812     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2813 
2814     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2815     if (!smc->pre_6_2_numa_affinity) {
2816         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2817     }
2818 
2819     /* advertise support for dedicated HP event source to guests */
2820     if (spapr->use_hotplug_event_source) {
2821         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2822     }
2823 
2824     /* advertise support for HPT resizing */
2825     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2826         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2827     }
2828 
2829     /* advertise support for ibm,dyamic-memory-v2 */
2830     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2831 
2832     /* advertise XIVE on POWER9 machines */
2833     if (spapr->irq->xive) {
2834         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2835     }
2836 
2837     /* init CPUs */
2838     spapr_init_cpus(spapr);
2839 
2840     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2841 
2842     /* Init numa_assoc_array */
2843     spapr_numa_associativity_init(spapr, machine);
2844 
2845     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2846         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2847                               spapr->max_compat_pvr)) {
2848         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2849         /* KVM and TCG always allow GTSE with radix... */
2850         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2851     }
2852     /* ... but not with hash (currently). */
2853 
2854     if (kvm_enabled()) {
2855         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2856         kvmppc_enable_logical_ci_hcalls();
2857         kvmppc_enable_set_mode_hcall();
2858 
2859         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2860         kvmppc_enable_clear_ref_mod_hcalls();
2861 
2862         /* Enable H_PAGE_INIT */
2863         kvmppc_enable_h_page_init();
2864     }
2865 
2866     /* map RAM */
2867     memory_region_add_subregion(sysmem, 0, machine->ram);
2868 
2869     /* always allocate the device memory information */
2870     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2871 
2872     /* initialize hotplug memory address space */
2873     if (machine->ram_size < machine->maxram_size) {
2874         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2875         /*
2876          * Limit the number of hotpluggable memory slots to half the number
2877          * slots that KVM supports, leaving the other half for PCI and other
2878          * devices. However ensure that number of slots doesn't drop below 32.
2879          */
2880         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2881                            SPAPR_MAX_RAM_SLOTS;
2882 
2883         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2884             max_memslots = SPAPR_MAX_RAM_SLOTS;
2885         }
2886         if (machine->ram_slots > max_memslots) {
2887             error_report("Specified number of memory slots %"
2888                          PRIu64" exceeds max supported %d",
2889                          machine->ram_slots, max_memslots);
2890             exit(1);
2891         }
2892 
2893         machine->device_memory->base = ROUND_UP(machine->ram_size,
2894                                                 SPAPR_DEVICE_MEM_ALIGN);
2895         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2896                            "device-memory", device_mem_size);
2897         memory_region_add_subregion(sysmem, machine->device_memory->base,
2898                                     &machine->device_memory->mr);
2899     }
2900 
2901     if (smc->dr_lmb_enabled) {
2902         spapr_create_lmb_dr_connectors(spapr);
2903     }
2904 
2905     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2906         /* Create the error string for live migration blocker */
2907         error_setg(&spapr->fwnmi_migration_blocker,
2908             "A machine check is being handled during migration. The handler"
2909             "may run and log hardware error on the destination");
2910     }
2911 
2912     if (mc->nvdimm_supported) {
2913         spapr_create_nvdimm_dr_connectors(spapr);
2914     }
2915 
2916     /* Set up RTAS event infrastructure */
2917     spapr_events_init(spapr);
2918 
2919     /* Set up the RTC RTAS interfaces */
2920     spapr_rtc_create(spapr);
2921 
2922     /* Set up VIO bus */
2923     spapr->vio_bus = spapr_vio_bus_init();
2924 
2925     for (i = 0; serial_hd(i); i++) {
2926         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2927     }
2928 
2929     /* We always have at least the nvram device on VIO */
2930     spapr_create_nvram(spapr);
2931 
2932     /*
2933      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2934      * connectors (described in root DT node's "ibm,drc-types" property)
2935      * are pre-initialized here. additional child connectors (such as
2936      * connectors for a PHBs PCI slots) are added as needed during their
2937      * parent's realization.
2938      */
2939     if (smc->dr_phb_enabled) {
2940         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2941             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2942         }
2943     }
2944 
2945     /* Set up PCI */
2946     spapr_pci_rtas_init();
2947 
2948     phb = spapr_create_default_phb();
2949 
2950     for (i = 0; i < nb_nics; i++) {
2951         NICInfo *nd = &nd_table[i];
2952 
2953         if (!nd->model) {
2954             nd->model = g_strdup("spapr-vlan");
2955         }
2956 
2957         if (g_str_equal(nd->model, "spapr-vlan") ||
2958             g_str_equal(nd->model, "ibmveth")) {
2959             spapr_vlan_create(spapr->vio_bus, nd);
2960         } else {
2961             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2962         }
2963     }
2964 
2965     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2966         spapr_vscsi_create(spapr->vio_bus);
2967     }
2968 
2969     /* Graphics */
2970     has_vga = spapr_vga_init(phb->bus, &error_fatal);
2971     if (has_vga) {
2972         spapr->want_stdout_path = !machine->enable_graphics;
2973         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2974     } else {
2975         spapr->want_stdout_path = true;
2976     }
2977 
2978     if (machine->usb) {
2979         if (smc->use_ohci_by_default) {
2980             pci_create_simple(phb->bus, -1, "pci-ohci");
2981         } else {
2982             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2983         }
2984 
2985         if (has_vga) {
2986             USBBus *usb_bus = usb_bus_find(-1);
2987 
2988             usb_create_simple(usb_bus, "usb-kbd");
2989             usb_create_simple(usb_bus, "usb-mouse");
2990         }
2991     }
2992 
2993     if (kernel_filename) {
2994         uint64_t loaded_addr = 0;
2995 
2996         spapr->kernel_size = load_elf(kernel_filename, NULL,
2997                                       translate_kernel_address, spapr,
2998                                       NULL, &loaded_addr, NULL, NULL, 1,
2999                                       PPC_ELF_MACHINE, 0, 0);
3000         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3001             spapr->kernel_size = load_elf(kernel_filename, NULL,
3002                                           translate_kernel_address, spapr,
3003                                           NULL, &loaded_addr, NULL, NULL, 0,
3004                                           PPC_ELF_MACHINE, 0, 0);
3005             spapr->kernel_le = spapr->kernel_size > 0;
3006         }
3007         if (spapr->kernel_size < 0) {
3008             error_report("error loading %s: %s", kernel_filename,
3009                          load_elf_strerror(spapr->kernel_size));
3010             exit(1);
3011         }
3012 
3013         if (spapr->kernel_addr != loaded_addr) {
3014             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3015                         " to 0x%"PRIx64,
3016                         spapr->kernel_addr, loaded_addr);
3017             spapr->kernel_addr = loaded_addr;
3018         }
3019 
3020         /* load initrd */
3021         if (initrd_filename) {
3022             /* Try to locate the initrd in the gap between the kernel
3023              * and the firmware. Add a bit of space just in case
3024              */
3025             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3026                                   + 0x1ffff) & ~0xffff;
3027             spapr->initrd_size = load_image_targphys(initrd_filename,
3028                                                      spapr->initrd_base,
3029                                                      load_limit
3030                                                      - spapr->initrd_base);
3031             if (spapr->initrd_size < 0) {
3032                 error_report("could not load initial ram disk '%s'",
3033                              initrd_filename);
3034                 exit(1);
3035             }
3036         }
3037     }
3038 
3039     /* FIXME: Should register things through the MachineState's qdev
3040      * interface, this is a legacy from the sPAPREnvironment structure
3041      * which predated MachineState but had a similar function */
3042     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3043     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3044                          &savevm_htab_handlers, spapr);
3045 
3046     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3047 
3048     qemu_register_boot_set(spapr_boot_set, spapr);
3049 
3050     /*
3051      * Nothing needs to be done to resume a suspended guest because
3052      * suspending does not change the machine state, so no need for
3053      * a ->wakeup method.
3054      */
3055     qemu_register_wakeup_support();
3056 
3057     if (kvm_enabled()) {
3058         /* to stop and start vmclock */
3059         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3060                                          &spapr->tb);
3061 
3062         kvmppc_spapr_enable_inkernel_multitce();
3063     }
3064 
3065     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3066     if (spapr->vof) {
3067         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3068         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3069     }
3070 
3071     spapr_watchdog_init(spapr);
3072 }
3073 
3074 #define DEFAULT_KVM_TYPE "auto"
3075 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3076 {
3077     /*
3078      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3079      * accomodate the 'HV' and 'PV' formats that exists in the
3080      * wild. The 'auto' mode is being introduced already as
3081      * lower-case, thus we don't need to bother checking for
3082      * "AUTO".
3083      */
3084     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3085         return 0;
3086     }
3087 
3088     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3089         return 1;
3090     }
3091 
3092     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3093         return 2;
3094     }
3095 
3096     error_report("Unknown kvm-type specified '%s'", vm_type);
3097     exit(1);
3098 }
3099 
3100 /*
3101  * Implementation of an interface to adjust firmware path
3102  * for the bootindex property handling.
3103  */
3104 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3105                                    DeviceState *dev)
3106 {
3107 #define CAST(type, obj, name) \
3108     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3109     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3110     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3111     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3112     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3113 
3114     if (d && bus) {
3115         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3116         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3117         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3118 
3119         if (spapr) {
3120             /*
3121              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3122              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3123              * 0x8000 | (target << 8) | (bus << 5) | lun
3124              * (see the "Logical unit addressing format" table in SAM5)
3125              */
3126             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3127             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3128                                    (uint64_t)id << 48);
3129         } else if (virtio) {
3130             /*
3131              * We use SRP luns of the form 01000000 | (target << 8) | lun
3132              * in the top 32 bits of the 64-bit LUN
3133              * Note: the quote above is from SLOF and it is wrong,
3134              * the actual binding is:
3135              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3136              */
3137             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3138             if (d->lun >= 256) {
3139                 /* Use the LUN "flat space addressing method" */
3140                 id |= 0x4000;
3141             }
3142             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3143                                    (uint64_t)id << 32);
3144         } else if (usb) {
3145             /*
3146              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3147              * in the top 32 bits of the 64-bit LUN
3148              */
3149             unsigned usb_port = atoi(usb->port->path);
3150             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3151             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3152                                    (uint64_t)id << 32);
3153         }
3154     }
3155 
3156     /*
3157      * SLOF probes the USB devices, and if it recognizes that the device is a
3158      * storage device, it changes its name to "storage" instead of "usb-host",
3159      * and additionally adds a child node for the SCSI LUN, so the correct
3160      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3161      */
3162     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3163         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3164         if (usb_device_is_scsi_storage(usbdev)) {
3165             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3166         }
3167     }
3168 
3169     if (phb) {
3170         /* Replace "pci" with "pci@800000020000000" */
3171         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3172     }
3173 
3174     if (vsc) {
3175         /* Same logic as virtio above */
3176         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3177         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3178     }
3179 
3180     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3181         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3182         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3183         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3184     }
3185 
3186     if (pcidev) {
3187         return spapr_pci_fw_dev_name(pcidev);
3188     }
3189 
3190     return NULL;
3191 }
3192 
3193 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3194 {
3195     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3196 
3197     return g_strdup(spapr->kvm_type);
3198 }
3199 
3200 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3201 {
3202     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3203 
3204     g_free(spapr->kvm_type);
3205     spapr->kvm_type = g_strdup(value);
3206 }
3207 
3208 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3209 {
3210     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3211 
3212     return spapr->use_hotplug_event_source;
3213 }
3214 
3215 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3216                                             Error **errp)
3217 {
3218     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3219 
3220     spapr->use_hotplug_event_source = value;
3221 }
3222 
3223 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3224 {
3225     return true;
3226 }
3227 
3228 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3229 {
3230     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3231 
3232     switch (spapr->resize_hpt) {
3233     case SPAPR_RESIZE_HPT_DEFAULT:
3234         return g_strdup("default");
3235     case SPAPR_RESIZE_HPT_DISABLED:
3236         return g_strdup("disabled");
3237     case SPAPR_RESIZE_HPT_ENABLED:
3238         return g_strdup("enabled");
3239     case SPAPR_RESIZE_HPT_REQUIRED:
3240         return g_strdup("required");
3241     }
3242     g_assert_not_reached();
3243 }
3244 
3245 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3246 {
3247     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3248 
3249     if (strcmp(value, "default") == 0) {
3250         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3251     } else if (strcmp(value, "disabled") == 0) {
3252         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3253     } else if (strcmp(value, "enabled") == 0) {
3254         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3255     } else if (strcmp(value, "required") == 0) {
3256         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3257     } else {
3258         error_setg(errp, "Bad value for \"resize-hpt\" property");
3259     }
3260 }
3261 
3262 static bool spapr_get_vof(Object *obj, Error **errp)
3263 {
3264     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3265 
3266     return spapr->vof != NULL;
3267 }
3268 
3269 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3270 {
3271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272 
3273     if (spapr->vof) {
3274         vof_cleanup(spapr->vof);
3275         g_free(spapr->vof);
3276         spapr->vof = NULL;
3277     }
3278     if (!value) {
3279         return;
3280     }
3281     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3282 }
3283 
3284 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3285 {
3286     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3287 
3288     if (spapr->irq == &spapr_irq_xics_legacy) {
3289         return g_strdup("legacy");
3290     } else if (spapr->irq == &spapr_irq_xics) {
3291         return g_strdup("xics");
3292     } else if (spapr->irq == &spapr_irq_xive) {
3293         return g_strdup("xive");
3294     } else if (spapr->irq == &spapr_irq_dual) {
3295         return g_strdup("dual");
3296     }
3297     g_assert_not_reached();
3298 }
3299 
3300 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3301 {
3302     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3303 
3304     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3305         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3306         return;
3307     }
3308 
3309     /* The legacy IRQ backend can not be set */
3310     if (strcmp(value, "xics") == 0) {
3311         spapr->irq = &spapr_irq_xics;
3312     } else if (strcmp(value, "xive") == 0) {
3313         spapr->irq = &spapr_irq_xive;
3314     } else if (strcmp(value, "dual") == 0) {
3315         spapr->irq = &spapr_irq_dual;
3316     } else {
3317         error_setg(errp, "Bad value for \"ic-mode\" property");
3318     }
3319 }
3320 
3321 static char *spapr_get_host_model(Object *obj, Error **errp)
3322 {
3323     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3324 
3325     return g_strdup(spapr->host_model);
3326 }
3327 
3328 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3329 {
3330     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3331 
3332     g_free(spapr->host_model);
3333     spapr->host_model = g_strdup(value);
3334 }
3335 
3336 static char *spapr_get_host_serial(Object *obj, Error **errp)
3337 {
3338     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3339 
3340     return g_strdup(spapr->host_serial);
3341 }
3342 
3343 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3344 {
3345     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3346 
3347     g_free(spapr->host_serial);
3348     spapr->host_serial = g_strdup(value);
3349 }
3350 
3351 static void spapr_instance_init(Object *obj)
3352 {
3353     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3354     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3355     MachineState *ms = MACHINE(spapr);
3356     MachineClass *mc = MACHINE_GET_CLASS(ms);
3357 
3358     /*
3359      * NVDIMM support went live in 5.1 without considering that, in
3360      * other archs, the user needs to enable NVDIMM support with the
3361      * 'nvdimm' machine option and the default behavior is NVDIMM
3362      * support disabled. It is too late to roll back to the standard
3363      * behavior without breaking 5.1 guests.
3364      */
3365     if (mc->nvdimm_supported) {
3366         ms->nvdimms_state->is_enabled = true;
3367     }
3368 
3369     spapr->htab_fd = -1;
3370     spapr->use_hotplug_event_source = true;
3371     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3372     object_property_add_str(obj, "kvm-type",
3373                             spapr_get_kvm_type, spapr_set_kvm_type);
3374     object_property_set_description(obj, "kvm-type",
3375                                     "Specifies the KVM virtualization mode (auto,"
3376                                     " hv, pr). Defaults to 'auto'. This mode will use"
3377                                     " any available KVM module loaded in the host,"
3378                                     " where kvm_hv takes precedence if both kvm_hv and"
3379                                     " kvm_pr are loaded.");
3380     object_property_add_bool(obj, "modern-hotplug-events",
3381                             spapr_get_modern_hotplug_events,
3382                             spapr_set_modern_hotplug_events);
3383     object_property_set_description(obj, "modern-hotplug-events",
3384                                     "Use dedicated hotplug event mechanism in"
3385                                     " place of standard EPOW events when possible"
3386                                     " (required for memory hot-unplug support)");
3387     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3388                             "Maximum permitted CPU compatibility mode");
3389 
3390     object_property_add_str(obj, "resize-hpt",
3391                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3392     object_property_set_description(obj, "resize-hpt",
3393                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3394     object_property_add_uint32_ptr(obj, "vsmt",
3395                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3396     object_property_set_description(obj, "vsmt",
3397                                     "Virtual SMT: KVM behaves as if this were"
3398                                     " the host's SMT mode");
3399 
3400     object_property_add_bool(obj, "vfio-no-msix-emulation",
3401                              spapr_get_msix_emulation, NULL);
3402 
3403     object_property_add_uint64_ptr(obj, "kernel-addr",
3404                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3405     object_property_set_description(obj, "kernel-addr",
3406                                     stringify(KERNEL_LOAD_ADDR)
3407                                     " for -kernel is the default");
3408     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3409 
3410     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3411     object_property_set_description(obj, "x-vof",
3412                                     "Enable Virtual Open Firmware (experimental)");
3413 
3414     /* The machine class defines the default interrupt controller mode */
3415     spapr->irq = smc->irq;
3416     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3417                             spapr_set_ic_mode);
3418     object_property_set_description(obj, "ic-mode",
3419                  "Specifies the interrupt controller mode (xics, xive, dual)");
3420 
3421     object_property_add_str(obj, "host-model",
3422         spapr_get_host_model, spapr_set_host_model);
3423     object_property_set_description(obj, "host-model",
3424         "Host model to advertise in guest device tree");
3425     object_property_add_str(obj, "host-serial",
3426         spapr_get_host_serial, spapr_set_host_serial);
3427     object_property_set_description(obj, "host-serial",
3428         "Host serial number to advertise in guest device tree");
3429 }
3430 
3431 static void spapr_machine_finalizefn(Object *obj)
3432 {
3433     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3434 
3435     g_free(spapr->kvm_type);
3436 }
3437 
3438 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3439 {
3440     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3441     PowerPCCPU *cpu = POWERPC_CPU(cs);
3442     CPUPPCState *env = &cpu->env;
3443 
3444     cpu_synchronize_state(cs);
3445     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3446     if (spapr->fwnmi_system_reset_addr != -1) {
3447         uint64_t rtas_addr, addr;
3448 
3449         /* get rtas addr from fdt */
3450         rtas_addr = spapr_get_rtas_addr();
3451         if (!rtas_addr) {
3452             qemu_system_guest_panicked(NULL);
3453             return;
3454         }
3455 
3456         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3457         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3458         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3459         env->gpr[3] = addr;
3460     }
3461     ppc_cpu_do_system_reset(cs);
3462     if (spapr->fwnmi_system_reset_addr != -1) {
3463         env->nip = spapr->fwnmi_system_reset_addr;
3464     }
3465 }
3466 
3467 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3468 {
3469     CPUState *cs;
3470 
3471     CPU_FOREACH(cs) {
3472         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3473     }
3474 }
3475 
3476 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3477                           void *fdt, int *fdt_start_offset, Error **errp)
3478 {
3479     uint64_t addr;
3480     uint32_t node;
3481 
3482     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3483     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3484                                     &error_abort);
3485     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3486                                              SPAPR_MEMORY_BLOCK_SIZE);
3487     return 0;
3488 }
3489 
3490 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3491                            bool dedicated_hp_event_source)
3492 {
3493     SpaprDrc *drc;
3494     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3495     int i;
3496     uint64_t addr = addr_start;
3497     bool hotplugged = spapr_drc_hotplugged(dev);
3498 
3499     for (i = 0; i < nr_lmbs; i++) {
3500         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3501                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3502         g_assert(drc);
3503 
3504         /*
3505          * memory_device_get_free_addr() provided a range of free addresses
3506          * that doesn't overlap with any existing mapping at pre-plug. The
3507          * corresponding LMB DRCs are thus assumed to be all attachable.
3508          */
3509         spapr_drc_attach(drc, dev);
3510         if (!hotplugged) {
3511             spapr_drc_reset(drc);
3512         }
3513         addr += SPAPR_MEMORY_BLOCK_SIZE;
3514     }
3515     /* send hotplug notification to the
3516      * guest only in case of hotplugged memory
3517      */
3518     if (hotplugged) {
3519         if (dedicated_hp_event_source) {
3520             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3521                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3522             g_assert(drc);
3523             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3524                                                    nr_lmbs,
3525                                                    spapr_drc_index(drc));
3526         } else {
3527             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3528                                            nr_lmbs);
3529         }
3530     }
3531 }
3532 
3533 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3534 {
3535     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3536     PCDIMMDevice *dimm = PC_DIMM(dev);
3537     uint64_t size, addr;
3538     int64_t slot;
3539     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3540 
3541     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3542 
3543     pc_dimm_plug(dimm, MACHINE(ms));
3544 
3545     if (!is_nvdimm) {
3546         addr = object_property_get_uint(OBJECT(dimm),
3547                                         PC_DIMM_ADDR_PROP, &error_abort);
3548         spapr_add_lmbs(dev, addr, size,
3549                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3550     } else {
3551         slot = object_property_get_int(OBJECT(dimm),
3552                                        PC_DIMM_SLOT_PROP, &error_abort);
3553         /* We should have valid slot number at this point */
3554         g_assert(slot >= 0);
3555         spapr_add_nvdimm(dev, slot);
3556     }
3557 }
3558 
3559 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3560                                   Error **errp)
3561 {
3562     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3563     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3564     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3565     PCDIMMDevice *dimm = PC_DIMM(dev);
3566     Error *local_err = NULL;
3567     uint64_t size;
3568     Object *memdev;
3569     hwaddr pagesize;
3570 
3571     if (!smc->dr_lmb_enabled) {
3572         error_setg(errp, "Memory hotplug not supported for this machine");
3573         return;
3574     }
3575 
3576     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3577     if (local_err) {
3578         error_propagate(errp, local_err);
3579         return;
3580     }
3581 
3582     if (is_nvdimm) {
3583         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3584             return;
3585         }
3586     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3587         error_setg(errp, "Hotplugged memory size must be a multiple of "
3588                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3589         return;
3590     }
3591 
3592     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3593                                       &error_abort);
3594     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3595     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3596         return;
3597     }
3598 
3599     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3600 }
3601 
3602 struct SpaprDimmState {
3603     PCDIMMDevice *dimm;
3604     uint32_t nr_lmbs;
3605     QTAILQ_ENTRY(SpaprDimmState) next;
3606 };
3607 
3608 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3609                                                        PCDIMMDevice *dimm)
3610 {
3611     SpaprDimmState *dimm_state = NULL;
3612 
3613     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3614         if (dimm_state->dimm == dimm) {
3615             break;
3616         }
3617     }
3618     return dimm_state;
3619 }
3620 
3621 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3622                                                       uint32_t nr_lmbs,
3623                                                       PCDIMMDevice *dimm)
3624 {
3625     SpaprDimmState *ds = NULL;
3626 
3627     /*
3628      * If this request is for a DIMM whose removal had failed earlier
3629      * (due to guest's refusal to remove the LMBs), we would have this
3630      * dimm already in the pending_dimm_unplugs list. In that
3631      * case don't add again.
3632      */
3633     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3634     if (!ds) {
3635         ds = g_new0(SpaprDimmState, 1);
3636         ds->nr_lmbs = nr_lmbs;
3637         ds->dimm = dimm;
3638         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3639     }
3640     return ds;
3641 }
3642 
3643 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3644                                               SpaprDimmState *dimm_state)
3645 {
3646     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3647     g_free(dimm_state);
3648 }
3649 
3650 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3651                                                         PCDIMMDevice *dimm)
3652 {
3653     SpaprDrc *drc;
3654     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3655                                                   &error_abort);
3656     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3657     uint32_t avail_lmbs = 0;
3658     uint64_t addr_start, addr;
3659     int i;
3660 
3661     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3662                                           &error_abort);
3663 
3664     addr = addr_start;
3665     for (i = 0; i < nr_lmbs; i++) {
3666         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3667                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3668         g_assert(drc);
3669         if (drc->dev) {
3670             avail_lmbs++;
3671         }
3672         addr += SPAPR_MEMORY_BLOCK_SIZE;
3673     }
3674 
3675     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3676 }
3677 
3678 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3679 {
3680     SpaprDimmState *ds;
3681     PCDIMMDevice *dimm;
3682     SpaprDrc *drc;
3683     uint32_t nr_lmbs;
3684     uint64_t size, addr_start, addr;
3685     g_autofree char *qapi_error = NULL;
3686     int i;
3687 
3688     if (!dev) {
3689         return;
3690     }
3691 
3692     dimm = PC_DIMM(dev);
3693     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3694 
3695     /*
3696      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3697      * unplug state, but one of its DRC is marked as unplug_requested.
3698      * This is bad and weird enough to g_assert() out.
3699      */
3700     g_assert(ds);
3701 
3702     spapr_pending_dimm_unplugs_remove(spapr, ds);
3703 
3704     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3705     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3706 
3707     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3708                                           &error_abort);
3709 
3710     addr = addr_start;
3711     for (i = 0; i < nr_lmbs; i++) {
3712         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3713                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3714         g_assert(drc);
3715 
3716         drc->unplug_requested = false;
3717         addr += SPAPR_MEMORY_BLOCK_SIZE;
3718     }
3719 
3720     /*
3721      * Tell QAPI that something happened and the memory
3722      * hotunplug wasn't successful. Keep sending
3723      * MEM_UNPLUG_ERROR even while sending
3724      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3725      * MEM_UNPLUG_ERROR is due.
3726      */
3727     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3728                                  "for device %s", dev->id);
3729 
3730     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3731 
3732     qapi_event_send_device_unplug_guest_error(dev->id,
3733                                               dev->canonical_path);
3734 }
3735 
3736 /* Callback to be called during DRC release. */
3737 void spapr_lmb_release(DeviceState *dev)
3738 {
3739     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3740     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3741     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3742 
3743     /* This information will get lost if a migration occurs
3744      * during the unplug process. In this case recover it. */
3745     if (ds == NULL) {
3746         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3747         g_assert(ds);
3748         /* The DRC being examined by the caller at least must be counted */
3749         g_assert(ds->nr_lmbs);
3750     }
3751 
3752     if (--ds->nr_lmbs) {
3753         return;
3754     }
3755 
3756     /*
3757      * Now that all the LMBs have been removed by the guest, call the
3758      * unplug handler chain. This can never fail.
3759      */
3760     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3761     object_unparent(OBJECT(dev));
3762 }
3763 
3764 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3765 {
3766     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3767     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3768 
3769     /* We really shouldn't get this far without anything to unplug */
3770     g_assert(ds);
3771 
3772     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3773     qdev_unrealize(dev);
3774     spapr_pending_dimm_unplugs_remove(spapr, ds);
3775 }
3776 
3777 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3778                                         DeviceState *dev, Error **errp)
3779 {
3780     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3781     PCDIMMDevice *dimm = PC_DIMM(dev);
3782     uint32_t nr_lmbs;
3783     uint64_t size, addr_start, addr;
3784     int i;
3785     SpaprDrc *drc;
3786 
3787     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3788         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3789         return;
3790     }
3791 
3792     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3793     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3794 
3795     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3796                                           &error_abort);
3797 
3798     /*
3799      * An existing pending dimm state for this DIMM means that there is an
3800      * unplug operation in progress, waiting for the spapr_lmb_release
3801      * callback to complete the job (BQL can't cover that far). In this case,
3802      * bail out to avoid detaching DRCs that were already released.
3803      */
3804     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3805         error_setg(errp, "Memory unplug already in progress for device %s",
3806                    dev->id);
3807         return;
3808     }
3809 
3810     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3811 
3812     addr = addr_start;
3813     for (i = 0; i < nr_lmbs; i++) {
3814         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3815                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3816         g_assert(drc);
3817 
3818         spapr_drc_unplug_request(drc);
3819         addr += SPAPR_MEMORY_BLOCK_SIZE;
3820     }
3821 
3822     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3823                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3824     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3825                                               nr_lmbs, spapr_drc_index(drc));
3826 }
3827 
3828 /* Callback to be called during DRC release. */
3829 void spapr_core_release(DeviceState *dev)
3830 {
3831     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3832 
3833     /* Call the unplug handler chain. This can never fail. */
3834     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3835     object_unparent(OBJECT(dev));
3836 }
3837 
3838 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3839 {
3840     MachineState *ms = MACHINE(hotplug_dev);
3841     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3842     CPUCore *cc = CPU_CORE(dev);
3843     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3844 
3845     if (smc->pre_2_10_has_unused_icps) {
3846         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3847         int i;
3848 
3849         for (i = 0; i < cc->nr_threads; i++) {
3850             CPUState *cs = CPU(sc->threads[i]);
3851 
3852             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3853         }
3854     }
3855 
3856     assert(core_slot);
3857     core_slot->cpu = NULL;
3858     qdev_unrealize(dev);
3859 }
3860 
3861 static
3862 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3863                                Error **errp)
3864 {
3865     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3866     int index;
3867     SpaprDrc *drc;
3868     CPUCore *cc = CPU_CORE(dev);
3869 
3870     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3871         error_setg(errp, "Unable to find CPU core with core-id: %d",
3872                    cc->core_id);
3873         return;
3874     }
3875     if (index == 0) {
3876         error_setg(errp, "Boot CPU core may not be unplugged");
3877         return;
3878     }
3879 
3880     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3881                           spapr_vcpu_id(spapr, cc->core_id));
3882     g_assert(drc);
3883 
3884     if (!spapr_drc_unplug_requested(drc)) {
3885         spapr_drc_unplug_request(drc);
3886     }
3887 
3888     /*
3889      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3890      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3891      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3892      * attempt (e.g. the kernel will refuse to remove the last online
3893      * CPU), we will never attempt it again because unplug_requested
3894      * will still be 'true' in that case.
3895      */
3896     spapr_hotplug_req_remove_by_index(drc);
3897 }
3898 
3899 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3900                            void *fdt, int *fdt_start_offset, Error **errp)
3901 {
3902     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3903     CPUState *cs = CPU(core->threads[0]);
3904     PowerPCCPU *cpu = POWERPC_CPU(cs);
3905     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3906     int id = spapr_get_vcpu_id(cpu);
3907     g_autofree char *nodename = NULL;
3908     int offset;
3909 
3910     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3911     offset = fdt_add_subnode(fdt, 0, nodename);
3912 
3913     spapr_dt_cpu(cs, fdt, offset, spapr);
3914 
3915     /*
3916      * spapr_dt_cpu() does not fill the 'name' property in the
3917      * CPU node. The function is called during boot process, before
3918      * and after CAS, and overwriting the 'name' property written
3919      * by SLOF is not allowed.
3920      *
3921      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3922      * CPUs more compatible with the coldplugged ones, which have
3923      * the 'name' property. Linux Kernel also relies on this
3924      * property to identify CPU nodes.
3925      */
3926     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3927 
3928     *fdt_start_offset = offset;
3929     return 0;
3930 }
3931 
3932 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3933 {
3934     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3935     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3936     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3937     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3938     CPUCore *cc = CPU_CORE(dev);
3939     CPUState *cs;
3940     SpaprDrc *drc;
3941     CPUArchId *core_slot;
3942     int index;
3943     bool hotplugged = spapr_drc_hotplugged(dev);
3944     int i;
3945 
3946     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3947     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3948 
3949     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3950                           spapr_vcpu_id(spapr, cc->core_id));
3951 
3952     g_assert(drc || !mc->has_hotpluggable_cpus);
3953 
3954     if (drc) {
3955         /*
3956          * spapr_core_pre_plug() already buys us this is a brand new
3957          * core being plugged into a free slot. Nothing should already
3958          * be attached to the corresponding DRC.
3959          */
3960         spapr_drc_attach(drc, dev);
3961 
3962         if (hotplugged) {
3963             /*
3964              * Send hotplug notification interrupt to the guest only
3965              * in case of hotplugged CPUs.
3966              */
3967             spapr_hotplug_req_add_by_index(drc);
3968         } else {
3969             spapr_drc_reset(drc);
3970         }
3971     }
3972 
3973     core_slot->cpu = OBJECT(dev);
3974 
3975     /*
3976      * Set compatibility mode to match the boot CPU, which was either set
3977      * by the machine reset code or by CAS. This really shouldn't fail at
3978      * this point.
3979      */
3980     if (hotplugged) {
3981         for (i = 0; i < cc->nr_threads; i++) {
3982             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3983                            &error_abort);
3984         }
3985     }
3986 
3987     if (smc->pre_2_10_has_unused_icps) {
3988         for (i = 0; i < cc->nr_threads; i++) {
3989             cs = CPU(core->threads[i]);
3990             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3991         }
3992     }
3993 }
3994 
3995 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3996                                 Error **errp)
3997 {
3998     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3999     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4000     CPUCore *cc = CPU_CORE(dev);
4001     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4002     const char *type = object_get_typename(OBJECT(dev));
4003     CPUArchId *core_slot;
4004     int index;
4005     unsigned int smp_threads = machine->smp.threads;
4006 
4007     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4008         error_setg(errp, "CPU hotplug not supported for this machine");
4009         return;
4010     }
4011 
4012     if (strcmp(base_core_type, type)) {
4013         error_setg(errp, "CPU core type should be %s", base_core_type);
4014         return;
4015     }
4016 
4017     if (cc->core_id % smp_threads) {
4018         error_setg(errp, "invalid core id %d", cc->core_id);
4019         return;
4020     }
4021 
4022     /*
4023      * In general we should have homogeneous threads-per-core, but old
4024      * (pre hotplug support) machine types allow the last core to have
4025      * reduced threads as a compatibility hack for when we allowed
4026      * total vcpus not a multiple of threads-per-core.
4027      */
4028     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4029         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4030                    smp_threads);
4031         return;
4032     }
4033 
4034     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4035     if (!core_slot) {
4036         error_setg(errp, "core id %d out of range", cc->core_id);
4037         return;
4038     }
4039 
4040     if (core_slot->cpu) {
4041         error_setg(errp, "core %d already populated", cc->core_id);
4042         return;
4043     }
4044 
4045     numa_cpu_pre_plug(core_slot, dev, errp);
4046 }
4047 
4048 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4049                           void *fdt, int *fdt_start_offset, Error **errp)
4050 {
4051     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4052     int intc_phandle;
4053 
4054     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4055     if (intc_phandle <= 0) {
4056         return -1;
4057     }
4058 
4059     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4060         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4061         return -1;
4062     }
4063 
4064     /* generally SLOF creates these, for hotplug it's up to QEMU */
4065     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4066 
4067     return 0;
4068 }
4069 
4070 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4071                                Error **errp)
4072 {
4073     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4074     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4075     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4076     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4077     SpaprDrc *drc;
4078 
4079     if (dev->hotplugged && !smc->dr_phb_enabled) {
4080         error_setg(errp, "PHB hotplug not supported for this machine");
4081         return false;
4082     }
4083 
4084     if (sphb->index == (uint32_t)-1) {
4085         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4086         return false;
4087     }
4088 
4089     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4090     if (drc && drc->dev) {
4091         error_setg(errp, "PHB %d already attached", sphb->index);
4092         return false;
4093     }
4094 
4095     /*
4096      * This will check that sphb->index doesn't exceed the maximum number of
4097      * PHBs for the current machine type.
4098      */
4099     return
4100         smc->phb_placement(spapr, sphb->index,
4101                            &sphb->buid, &sphb->io_win_addr,
4102                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4103                            windows_supported, sphb->dma_liobn,
4104                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4105                            errp);
4106 }
4107 
4108 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4109 {
4110     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4111     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4112     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4113     SpaprDrc *drc;
4114     bool hotplugged = spapr_drc_hotplugged(dev);
4115 
4116     if (!smc->dr_phb_enabled) {
4117         return;
4118     }
4119 
4120     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4121     /* hotplug hooks should check it's enabled before getting this far */
4122     assert(drc);
4123 
4124     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4125     spapr_drc_attach(drc, dev);
4126 
4127     if (hotplugged) {
4128         spapr_hotplug_req_add_by_index(drc);
4129     } else {
4130         spapr_drc_reset(drc);
4131     }
4132 }
4133 
4134 void spapr_phb_release(DeviceState *dev)
4135 {
4136     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4137 
4138     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4139     object_unparent(OBJECT(dev));
4140 }
4141 
4142 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4143 {
4144     qdev_unrealize(dev);
4145 }
4146 
4147 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4148                                      DeviceState *dev, Error **errp)
4149 {
4150     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4151     SpaprDrc *drc;
4152 
4153     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4154     assert(drc);
4155 
4156     if (!spapr_drc_unplug_requested(drc)) {
4157         spapr_drc_unplug_request(drc);
4158         spapr_hotplug_req_remove_by_index(drc);
4159     } else {
4160         error_setg(errp,
4161                    "PCI Host Bridge unplug already in progress for device %s",
4162                    dev->id);
4163     }
4164 }
4165 
4166 static
4167 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4168                               Error **errp)
4169 {
4170     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4171 
4172     if (spapr->tpm_proxy != NULL) {
4173         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4174         return false;
4175     }
4176 
4177     return true;
4178 }
4179 
4180 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4181 {
4182     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4183     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4184 
4185     /* Already checked in spapr_tpm_proxy_pre_plug() */
4186     g_assert(spapr->tpm_proxy == NULL);
4187 
4188     spapr->tpm_proxy = tpm_proxy;
4189 }
4190 
4191 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4192 {
4193     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4194 
4195     qdev_unrealize(dev);
4196     object_unparent(OBJECT(dev));
4197     spapr->tpm_proxy = NULL;
4198 }
4199 
4200 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4201                                       DeviceState *dev, Error **errp)
4202 {
4203     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4204         spapr_memory_plug(hotplug_dev, dev);
4205     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4206         spapr_core_plug(hotplug_dev, dev);
4207     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4208         spapr_phb_plug(hotplug_dev, dev);
4209     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4210         spapr_tpm_proxy_plug(hotplug_dev, dev);
4211     }
4212 }
4213 
4214 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4215                                         DeviceState *dev, Error **errp)
4216 {
4217     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4218         spapr_memory_unplug(hotplug_dev, dev);
4219     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4220         spapr_core_unplug(hotplug_dev, dev);
4221     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4222         spapr_phb_unplug(hotplug_dev, dev);
4223     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4224         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4225     }
4226 }
4227 
4228 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4229 {
4230     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4231         /*
4232          * CAS will process all pending unplug requests.
4233          *
4234          * HACK: a guest could theoretically have cleared all bits in OV5,
4235          * but none of the guests we care for do.
4236          */
4237         spapr_ovec_empty(spapr->ov5_cas);
4238 }
4239 
4240 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4241                                                 DeviceState *dev, Error **errp)
4242 {
4243     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4244     MachineClass *mc = MACHINE_GET_CLASS(sms);
4245     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4246 
4247     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4248         if (spapr_memory_hot_unplug_supported(sms)) {
4249             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4250         } else {
4251             error_setg(errp, "Memory hot unplug not supported for this guest");
4252         }
4253     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4254         if (!mc->has_hotpluggable_cpus) {
4255             error_setg(errp, "CPU hot unplug not supported on this machine");
4256             return;
4257         }
4258         spapr_core_unplug_request(hotplug_dev, dev, errp);
4259     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4260         if (!smc->dr_phb_enabled) {
4261             error_setg(errp, "PHB hot unplug not supported on this machine");
4262             return;
4263         }
4264         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4265     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4266         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4267     }
4268 }
4269 
4270 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4271                                           DeviceState *dev, Error **errp)
4272 {
4273     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4274         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4275     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4276         spapr_core_pre_plug(hotplug_dev, dev, errp);
4277     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4278         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4279     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4280         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4281     }
4282 }
4283 
4284 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4285                                                  DeviceState *dev)
4286 {
4287     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4288         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4289         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4290         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4291         return HOTPLUG_HANDLER(machine);
4292     }
4293     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4294         PCIDevice *pcidev = PCI_DEVICE(dev);
4295         PCIBus *root = pci_device_root_bus(pcidev);
4296         SpaprPhbState *phb =
4297             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4298                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4299 
4300         if (phb) {
4301             return HOTPLUG_HANDLER(phb);
4302         }
4303     }
4304     return NULL;
4305 }
4306 
4307 static CpuInstanceProperties
4308 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4309 {
4310     CPUArchId *core_slot;
4311     MachineClass *mc = MACHINE_GET_CLASS(machine);
4312 
4313     /* make sure possible_cpu are intialized */
4314     mc->possible_cpu_arch_ids(machine);
4315     /* get CPU core slot containing thread that matches cpu_index */
4316     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4317     assert(core_slot);
4318     return core_slot->props;
4319 }
4320 
4321 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4322 {
4323     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4324 }
4325 
4326 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4327 {
4328     int i;
4329     unsigned int smp_threads = machine->smp.threads;
4330     unsigned int smp_cpus = machine->smp.cpus;
4331     const char *core_type;
4332     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4333     MachineClass *mc = MACHINE_GET_CLASS(machine);
4334 
4335     if (!mc->has_hotpluggable_cpus) {
4336         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4337     }
4338     if (machine->possible_cpus) {
4339         assert(machine->possible_cpus->len == spapr_max_cores);
4340         return machine->possible_cpus;
4341     }
4342 
4343     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4344     if (!core_type) {
4345         error_report("Unable to find sPAPR CPU Core definition");
4346         exit(1);
4347     }
4348 
4349     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4350                              sizeof(CPUArchId) * spapr_max_cores);
4351     machine->possible_cpus->len = spapr_max_cores;
4352     for (i = 0; i < machine->possible_cpus->len; i++) {
4353         int core_id = i * smp_threads;
4354 
4355         machine->possible_cpus->cpus[i].type = core_type;
4356         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4357         machine->possible_cpus->cpus[i].arch_id = core_id;
4358         machine->possible_cpus->cpus[i].props.has_core_id = true;
4359         machine->possible_cpus->cpus[i].props.core_id = core_id;
4360     }
4361     return machine->possible_cpus;
4362 }
4363 
4364 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4365                                 uint64_t *buid, hwaddr *pio,
4366                                 hwaddr *mmio32, hwaddr *mmio64,
4367                                 unsigned n_dma, uint32_t *liobns,
4368                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4369 {
4370     /*
4371      * New-style PHB window placement.
4372      *
4373      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4374      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4375      * windows.
4376      *
4377      * Some guest kernels can't work with MMIO windows above 1<<46
4378      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4379      *
4380      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4381      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4382      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4383      * 1TiB 64-bit MMIO windows for each PHB.
4384      */
4385     const uint64_t base_buid = 0x800000020000000ULL;
4386     int i;
4387 
4388     /* Sanity check natural alignments */
4389     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4390     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4391     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4392     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4393     /* Sanity check bounds */
4394     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4395                       SPAPR_PCI_MEM32_WIN_SIZE);
4396     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4397                       SPAPR_PCI_MEM64_WIN_SIZE);
4398 
4399     if (index >= SPAPR_MAX_PHBS) {
4400         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4401                    SPAPR_MAX_PHBS - 1);
4402         return false;
4403     }
4404 
4405     *buid = base_buid + index;
4406     for (i = 0; i < n_dma; ++i) {
4407         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4408     }
4409 
4410     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4411     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4412     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4413 
4414     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4415     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4416     return true;
4417 }
4418 
4419 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4420 {
4421     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4422 
4423     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4424 }
4425 
4426 static void spapr_ics_resend(XICSFabric *dev)
4427 {
4428     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4429 
4430     ics_resend(spapr->ics);
4431 }
4432 
4433 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4434 {
4435     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4436 
4437     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4438 }
4439 
4440 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4441                                  Monitor *mon)
4442 {
4443     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4444 
4445     spapr_irq_print_info(spapr, mon);
4446     monitor_printf(mon, "irqchip: %s\n",
4447                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4448 }
4449 
4450 /*
4451  * This is a XIVE only operation
4452  */
4453 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4454                            uint8_t nvt_blk, uint32_t nvt_idx,
4455                            bool cam_ignore, uint8_t priority,
4456                            uint32_t logic_serv, XiveTCTXMatch *match)
4457 {
4458     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4459     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4460     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4461     int count;
4462 
4463     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4464                            priority, logic_serv, match);
4465     if (count < 0) {
4466         return count;
4467     }
4468 
4469     /*
4470      * When we implement the save and restore of the thread interrupt
4471      * contexts in the enter/exit CPU handlers of the machine and the
4472      * escalations in QEMU, we should be able to handle non dispatched
4473      * vCPUs.
4474      *
4475      * Until this is done, the sPAPR machine should find at least one
4476      * matching context always.
4477      */
4478     if (count == 0) {
4479         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4480                       nvt_blk, nvt_idx);
4481     }
4482 
4483     return count;
4484 }
4485 
4486 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4487 {
4488     return cpu->vcpu_id;
4489 }
4490 
4491 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4492 {
4493     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4494     MachineState *ms = MACHINE(spapr);
4495     int vcpu_id;
4496 
4497     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4498 
4499     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4500         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4501         error_append_hint(errp, "Adjust the number of cpus to %d "
4502                           "or try to raise the number of threads per core\n",
4503                           vcpu_id * ms->smp.threads / spapr->vsmt);
4504         return false;
4505     }
4506 
4507     cpu->vcpu_id = vcpu_id;
4508     return true;
4509 }
4510 
4511 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4512 {
4513     CPUState *cs;
4514 
4515     CPU_FOREACH(cs) {
4516         PowerPCCPU *cpu = POWERPC_CPU(cs);
4517 
4518         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4519             return cpu;
4520         }
4521     }
4522 
4523     return NULL;
4524 }
4525 
4526 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4527 {
4528     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4529 
4530     return spapr_cpu->in_nested;
4531 }
4532 
4533 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4534 {
4535     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4536 
4537     /* These are only called by TCG, KVM maintains dispatch state */
4538 
4539     spapr_cpu->prod = false;
4540     if (spapr_cpu->vpa_addr) {
4541         CPUState *cs = CPU(cpu);
4542         uint32_t dispatch;
4543 
4544         dispatch = ldl_be_phys(cs->as,
4545                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4546         dispatch++;
4547         if ((dispatch & 1) != 0) {
4548             qemu_log_mask(LOG_GUEST_ERROR,
4549                           "VPA: incorrect dispatch counter value for "
4550                           "dispatched partition %u, correcting.\n", dispatch);
4551             dispatch++;
4552         }
4553         stl_be_phys(cs->as,
4554                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4555     }
4556 }
4557 
4558 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4559 {
4560     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4561 
4562     if (spapr_cpu->vpa_addr) {
4563         CPUState *cs = CPU(cpu);
4564         uint32_t dispatch;
4565 
4566         dispatch = ldl_be_phys(cs->as,
4567                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4568         dispatch++;
4569         if ((dispatch & 1) != 1) {
4570             qemu_log_mask(LOG_GUEST_ERROR,
4571                           "VPA: incorrect dispatch counter value for "
4572                           "preempted partition %u, correcting.\n", dispatch);
4573             dispatch++;
4574         }
4575         stl_be_phys(cs->as,
4576                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4577     }
4578 }
4579 
4580 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4581 {
4582     MachineClass *mc = MACHINE_CLASS(oc);
4583     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4584     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4585     NMIClass *nc = NMI_CLASS(oc);
4586     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4587     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4588     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4589     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4590     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4591     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4592 
4593     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4594     mc->ignore_boot_device_suffixes = true;
4595 
4596     /*
4597      * We set up the default / latest behaviour here.  The class_init
4598      * functions for the specific versioned machine types can override
4599      * these details for backwards compatibility
4600      */
4601     mc->init = spapr_machine_init;
4602     mc->reset = spapr_machine_reset;
4603     mc->block_default_type = IF_SCSI;
4604 
4605     /*
4606      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4607      * should be limited by the host capability instead of hardcoded.
4608      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4609      * guests are welcome to have as many CPUs as the host are capable
4610      * of emulate.
4611      */
4612     mc->max_cpus = INT32_MAX;
4613 
4614     mc->no_parallel = 1;
4615     mc->default_boot_order = "";
4616     mc->default_ram_size = 512 * MiB;
4617     mc->default_ram_id = "ppc_spapr.ram";
4618     mc->default_display = "std";
4619     mc->kvm_type = spapr_kvm_type;
4620     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4621     mc->pci_allow_0_address = true;
4622     assert(!mc->get_hotplug_handler);
4623     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4624     hc->pre_plug = spapr_machine_device_pre_plug;
4625     hc->plug = spapr_machine_device_plug;
4626     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4627     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4628     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4629     hc->unplug_request = spapr_machine_device_unplug_request;
4630     hc->unplug = spapr_machine_device_unplug;
4631 
4632     smc->dr_lmb_enabled = true;
4633     smc->update_dt_enabled = true;
4634     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.2");
4635     mc->has_hotpluggable_cpus = true;
4636     mc->nvdimm_supported = true;
4637     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4638     fwc->get_dev_path = spapr_get_fw_dev_path;
4639     nc->nmi_monitor_handler = spapr_nmi;
4640     smc->phb_placement = spapr_phb_placement;
4641     vhc->cpu_in_nested = spapr_cpu_in_nested;
4642     vhc->deliver_hv_excp = spapr_exit_nested;
4643     vhc->hypercall = emulate_spapr_hypercall;
4644     vhc->hpt_mask = spapr_hpt_mask;
4645     vhc->map_hptes = spapr_map_hptes;
4646     vhc->unmap_hptes = spapr_unmap_hptes;
4647     vhc->hpte_set_c = spapr_hpte_set_c;
4648     vhc->hpte_set_r = spapr_hpte_set_r;
4649     vhc->get_pate = spapr_get_pate;
4650     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4651     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4652     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4653     xic->ics_get = spapr_ics_get;
4654     xic->ics_resend = spapr_ics_resend;
4655     xic->icp_get = spapr_icp_get;
4656     ispc->print_info = spapr_pic_print_info;
4657     /* Force NUMA node memory size to be a multiple of
4658      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4659      * in which LMBs are represented and hot-added
4660      */
4661     mc->numa_mem_align_shift = 28;
4662     mc->auto_enable_numa = true;
4663 
4664     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4665     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4666     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4667     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4668     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4669     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4670     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4671     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4672     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4673     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4674     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4675     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4676 
4677     /*
4678      * This cap specifies whether the AIL 3 mode for
4679      * H_SET_RESOURCE is supported. The default is modified
4680      * by default_caps_with_cpu().
4681      */
4682     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4683     spapr_caps_add_properties(smc);
4684     smc->irq = &spapr_irq_dual;
4685     smc->dr_phb_enabled = true;
4686     smc->linux_pci_probe = true;
4687     smc->smp_threads_vsmt = true;
4688     smc->nr_xirqs = SPAPR_NR_XIRQS;
4689     xfc->match_nvt = spapr_match_nvt;
4690     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4691     vmc->quiesce = spapr_vof_quiesce;
4692     vmc->setprop = spapr_vof_setprop;
4693 }
4694 
4695 static const TypeInfo spapr_machine_info = {
4696     .name          = TYPE_SPAPR_MACHINE,
4697     .parent        = TYPE_MACHINE,
4698     .abstract      = true,
4699     .instance_size = sizeof(SpaprMachineState),
4700     .instance_init = spapr_instance_init,
4701     .instance_finalize = spapr_machine_finalizefn,
4702     .class_size    = sizeof(SpaprMachineClass),
4703     .class_init    = spapr_machine_class_init,
4704     .interfaces = (InterfaceInfo[]) {
4705         { TYPE_FW_PATH_PROVIDER },
4706         { TYPE_NMI },
4707         { TYPE_HOTPLUG_HANDLER },
4708         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4709         { TYPE_XICS_FABRIC },
4710         { TYPE_INTERRUPT_STATS_PROVIDER },
4711         { TYPE_XIVE_FABRIC },
4712         { TYPE_VOF_MACHINE_IF },
4713         { }
4714     },
4715 };
4716 
4717 static void spapr_machine_latest_class_options(MachineClass *mc)
4718 {
4719     mc->alias = "pseries";
4720     mc->is_default = true;
4721 }
4722 
4723 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4724     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4725                                                     void *data)      \
4726     {                                                                \
4727         MachineClass *mc = MACHINE_CLASS(oc);                        \
4728         spapr_machine_##suffix##_class_options(mc);                  \
4729         if (latest) {                                                \
4730             spapr_machine_latest_class_options(mc);                  \
4731         }                                                            \
4732     }                                                                \
4733     static const TypeInfo spapr_machine_##suffix##_info = {          \
4734         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4735         .parent = TYPE_SPAPR_MACHINE,                                \
4736         .class_init = spapr_machine_##suffix##_class_init,           \
4737     };                                                               \
4738     static void spapr_machine_register_##suffix(void)                \
4739     {                                                                \
4740         type_register(&spapr_machine_##suffix##_info);               \
4741     }                                                                \
4742     type_init(spapr_machine_register_##suffix)
4743 
4744 /*
4745  * pseries-8.1
4746  */
4747 static void spapr_machine_8_1_class_options(MachineClass *mc)
4748 {
4749     /* Defaults for the latest behaviour inherited from the base class */
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(8_1, "8.1", true);
4753 
4754 /*
4755  * pseries-8.0
4756  */
4757 static void spapr_machine_8_0_class_options(MachineClass *mc)
4758 {
4759     spapr_machine_8_1_class_options(mc);
4760     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4761 }
4762 
4763 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4764 
4765 /*
4766  * pseries-7.2
4767  */
4768 static void spapr_machine_7_2_class_options(MachineClass *mc)
4769 {
4770     spapr_machine_8_0_class_options(mc);
4771     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4772 }
4773 
4774 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4775 
4776 /*
4777  * pseries-7.1
4778  */
4779 static void spapr_machine_7_1_class_options(MachineClass *mc)
4780 {
4781     spapr_machine_7_2_class_options(mc);
4782     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4783 }
4784 
4785 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4786 
4787 /*
4788  * pseries-7.0
4789  */
4790 static void spapr_machine_7_0_class_options(MachineClass *mc)
4791 {
4792     spapr_machine_7_1_class_options(mc);
4793     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4794 }
4795 
4796 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4797 
4798 /*
4799  * pseries-6.2
4800  */
4801 static void spapr_machine_6_2_class_options(MachineClass *mc)
4802 {
4803     spapr_machine_7_0_class_options(mc);
4804     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4805 }
4806 
4807 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4808 
4809 /*
4810  * pseries-6.1
4811  */
4812 static void spapr_machine_6_1_class_options(MachineClass *mc)
4813 {
4814     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4815 
4816     spapr_machine_6_2_class_options(mc);
4817     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4818     smc->pre_6_2_numa_affinity = true;
4819     mc->smp_props.prefer_sockets = true;
4820 }
4821 
4822 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4823 
4824 /*
4825  * pseries-6.0
4826  */
4827 static void spapr_machine_6_0_class_options(MachineClass *mc)
4828 {
4829     spapr_machine_6_1_class_options(mc);
4830     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4831 }
4832 
4833 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4834 
4835 /*
4836  * pseries-5.2
4837  */
4838 static void spapr_machine_5_2_class_options(MachineClass *mc)
4839 {
4840     spapr_machine_6_0_class_options(mc);
4841     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4842 }
4843 
4844 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4845 
4846 /*
4847  * pseries-5.1
4848  */
4849 static void spapr_machine_5_1_class_options(MachineClass *mc)
4850 {
4851     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4852 
4853     spapr_machine_5_2_class_options(mc);
4854     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4855     smc->pre_5_2_numa_associativity = true;
4856 }
4857 
4858 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4859 
4860 /*
4861  * pseries-5.0
4862  */
4863 static void spapr_machine_5_0_class_options(MachineClass *mc)
4864 {
4865     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4866     static GlobalProperty compat[] = {
4867         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4868     };
4869 
4870     spapr_machine_5_1_class_options(mc);
4871     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4872     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4873     mc->numa_mem_supported = true;
4874     smc->pre_5_1_assoc_refpoints = true;
4875 }
4876 
4877 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4878 
4879 /*
4880  * pseries-4.2
4881  */
4882 static void spapr_machine_4_2_class_options(MachineClass *mc)
4883 {
4884     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4885 
4886     spapr_machine_5_0_class_options(mc);
4887     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4888     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4889     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4890     smc->rma_limit = 16 * GiB;
4891     mc->nvdimm_supported = false;
4892 }
4893 
4894 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4895 
4896 /*
4897  * pseries-4.1
4898  */
4899 static void spapr_machine_4_1_class_options(MachineClass *mc)
4900 {
4901     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4902     static GlobalProperty compat[] = {
4903         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4904         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4905     };
4906 
4907     spapr_machine_4_2_class_options(mc);
4908     smc->linux_pci_probe = false;
4909     smc->smp_threads_vsmt = false;
4910     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4911     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4912 }
4913 
4914 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4915 
4916 /*
4917  * pseries-4.0
4918  */
4919 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4920                               uint64_t *buid, hwaddr *pio,
4921                               hwaddr *mmio32, hwaddr *mmio64,
4922                               unsigned n_dma, uint32_t *liobns,
4923                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4924 {
4925     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4926                              liobns, nv2gpa, nv2atsd, errp)) {
4927         return false;
4928     }
4929 
4930     *nv2gpa = 0;
4931     *nv2atsd = 0;
4932     return true;
4933 }
4934 static void spapr_machine_4_0_class_options(MachineClass *mc)
4935 {
4936     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4937 
4938     spapr_machine_4_1_class_options(mc);
4939     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4940     smc->phb_placement = phb_placement_4_0;
4941     smc->irq = &spapr_irq_xics;
4942     smc->pre_4_1_migration = true;
4943 }
4944 
4945 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4946 
4947 /*
4948  * pseries-3.1
4949  */
4950 static void spapr_machine_3_1_class_options(MachineClass *mc)
4951 {
4952     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4953 
4954     spapr_machine_4_0_class_options(mc);
4955     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4956 
4957     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4958     smc->update_dt_enabled = false;
4959     smc->dr_phb_enabled = false;
4960     smc->broken_host_serial_model = true;
4961     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4962     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4963     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4964     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4965 }
4966 
4967 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4968 
4969 /*
4970  * pseries-3.0
4971  */
4972 
4973 static void spapr_machine_3_0_class_options(MachineClass *mc)
4974 {
4975     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4976 
4977     spapr_machine_3_1_class_options(mc);
4978     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4979 
4980     smc->legacy_irq_allocation = true;
4981     smc->nr_xirqs = 0x400;
4982     smc->irq = &spapr_irq_xics_legacy;
4983 }
4984 
4985 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4986 
4987 /*
4988  * pseries-2.12
4989  */
4990 static void spapr_machine_2_12_class_options(MachineClass *mc)
4991 {
4992     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4993     static GlobalProperty compat[] = {
4994         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4995         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4996     };
4997 
4998     spapr_machine_3_0_class_options(mc);
4999     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5000     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5001 
5002     /* We depend on kvm_enabled() to choose a default value for the
5003      * hpt-max-page-size capability. Of course we can't do it here
5004      * because this is too early and the HW accelerator isn't initialzed
5005      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5006      */
5007     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5008 }
5009 
5010 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5011 
5012 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5013 {
5014     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5015 
5016     spapr_machine_2_12_class_options(mc);
5017     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5018     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5019     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5020 }
5021 
5022 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5023 
5024 /*
5025  * pseries-2.11
5026  */
5027 
5028 static void spapr_machine_2_11_class_options(MachineClass *mc)
5029 {
5030     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5031 
5032     spapr_machine_2_12_class_options(mc);
5033     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5034     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5035 }
5036 
5037 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5038 
5039 /*
5040  * pseries-2.10
5041  */
5042 
5043 static void spapr_machine_2_10_class_options(MachineClass *mc)
5044 {
5045     spapr_machine_2_11_class_options(mc);
5046     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5047 }
5048 
5049 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5050 
5051 /*
5052  * pseries-2.9
5053  */
5054 
5055 static void spapr_machine_2_9_class_options(MachineClass *mc)
5056 {
5057     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5058     static GlobalProperty compat[] = {
5059         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5060     };
5061 
5062     spapr_machine_2_10_class_options(mc);
5063     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5064     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5065     smc->pre_2_10_has_unused_icps = true;
5066     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5067 }
5068 
5069 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5070 
5071 /*
5072  * pseries-2.8
5073  */
5074 
5075 static void spapr_machine_2_8_class_options(MachineClass *mc)
5076 {
5077     static GlobalProperty compat[] = {
5078         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5079     };
5080 
5081     spapr_machine_2_9_class_options(mc);
5082     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5083     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5084     mc->numa_mem_align_shift = 23;
5085 }
5086 
5087 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5088 
5089 /*
5090  * pseries-2.7
5091  */
5092 
5093 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5094                               uint64_t *buid, hwaddr *pio,
5095                               hwaddr *mmio32, hwaddr *mmio64,
5096                               unsigned n_dma, uint32_t *liobns,
5097                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
5098 {
5099     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5100     const uint64_t base_buid = 0x800000020000000ULL;
5101     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5102     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5103     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5104     const uint32_t max_index = 255;
5105     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5106 
5107     uint64_t ram_top = MACHINE(spapr)->ram_size;
5108     hwaddr phb0_base, phb_base;
5109     int i;
5110 
5111     /* Do we have device memory? */
5112     if (MACHINE(spapr)->maxram_size > ram_top) {
5113         /* Can't just use maxram_size, because there may be an
5114          * alignment gap between normal and device memory regions
5115          */
5116         ram_top = MACHINE(spapr)->device_memory->base +
5117             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5118     }
5119 
5120     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5121 
5122     if (index > max_index) {
5123         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5124                    max_index);
5125         return false;
5126     }
5127 
5128     *buid = base_buid + index;
5129     for (i = 0; i < n_dma; ++i) {
5130         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5131     }
5132 
5133     phb_base = phb0_base + index * phb_spacing;
5134     *pio = phb_base + pio_offset;
5135     *mmio32 = phb_base + mmio_offset;
5136     /*
5137      * We don't set the 64-bit MMIO window, relying on the PHB's
5138      * fallback behaviour of automatically splitting a large "32-bit"
5139      * window into contiguous 32-bit and 64-bit windows
5140      */
5141 
5142     *nv2gpa = 0;
5143     *nv2atsd = 0;
5144     return true;
5145 }
5146 
5147 static void spapr_machine_2_7_class_options(MachineClass *mc)
5148 {
5149     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5150     static GlobalProperty compat[] = {
5151         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5152         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5153         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5154         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5155     };
5156 
5157     spapr_machine_2_8_class_options(mc);
5158     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5159     mc->default_machine_opts = "modern-hotplug-events=off";
5160     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5161     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5162     smc->phb_placement = phb_placement_2_7;
5163 }
5164 
5165 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5166 
5167 /*
5168  * pseries-2.6
5169  */
5170 
5171 static void spapr_machine_2_6_class_options(MachineClass *mc)
5172 {
5173     static GlobalProperty compat[] = {
5174         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5175     };
5176 
5177     spapr_machine_2_7_class_options(mc);
5178     mc->has_hotpluggable_cpus = false;
5179     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5180     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5181 }
5182 
5183 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5184 
5185 /*
5186  * pseries-2.5
5187  */
5188 
5189 static void spapr_machine_2_5_class_options(MachineClass *mc)
5190 {
5191     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5192     static GlobalProperty compat[] = {
5193         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5194     };
5195 
5196     spapr_machine_2_6_class_options(mc);
5197     smc->use_ohci_by_default = true;
5198     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5199     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5200 }
5201 
5202 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5203 
5204 /*
5205  * pseries-2.4
5206  */
5207 
5208 static void spapr_machine_2_4_class_options(MachineClass *mc)
5209 {
5210     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5211 
5212     spapr_machine_2_5_class_options(mc);
5213     smc->dr_lmb_enabled = false;
5214     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5215 }
5216 
5217 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5218 
5219 /*
5220  * pseries-2.3
5221  */
5222 
5223 static void spapr_machine_2_3_class_options(MachineClass *mc)
5224 {
5225     static GlobalProperty compat[] = {
5226         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5227     };
5228     spapr_machine_2_4_class_options(mc);
5229     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5230     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5231 }
5232 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5233 
5234 /*
5235  * pseries-2.2
5236  */
5237 
5238 static void spapr_machine_2_2_class_options(MachineClass *mc)
5239 {
5240     static GlobalProperty compat[] = {
5241         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5242     };
5243 
5244     spapr_machine_2_3_class_options(mc);
5245     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5246     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5247     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5248 }
5249 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5250 
5251 /*
5252  * pseries-2.1
5253  */
5254 
5255 static void spapr_machine_2_1_class_options(MachineClass *mc)
5256 {
5257     spapr_machine_2_2_class_options(mc);
5258     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5259 }
5260 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5261 
5262 static void spapr_machine_register_types(void)
5263 {
5264     type_register_static(&spapr_machine_info);
5265 }
5266 
5267 type_init(spapr_machine_register_types)
5268