xref: /qemu/hw/ppc/spapr.c (revision 5ced78955fe3f74002ad27676cf7c65cc89d6660)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "mmu-hash64.h"
50 #include "mmu-book3s-v3.h"
51 #include "cpu-models.h"
52 #include "hw/core/cpu.h"
53 
54 #include "hw/boards.h"
55 #include "hw/ppc/ppc.h"
56 #include "hw/loader.h"
57 
58 #include "hw/ppc/fdt.h"
59 #include "hw/ppc/spapr.h"
60 #include "hw/ppc/spapr_vio.h"
61 #include "hw/qdev-properties.h"
62 #include "hw/pci-host/spapr.h"
63 #include "hw/pci/msi.h"
64 
65 #include "hw/pci/pci.h"
66 #include "hw/scsi/scsi.h"
67 #include "hw/virtio/virtio-scsi.h"
68 #include "hw/virtio/vhost-scsi-common.h"
69 
70 #include "exec/address-spaces.h"
71 #include "exec/ram_addr.h"
72 #include "hw/usb.h"
73 #include "qemu/config-file.h"
74 #include "qemu/error-report.h"
75 #include "trace.h"
76 #include "hw/nmi.h"
77 #include "hw/intc/intc.h"
78 
79 #include "qemu/cutils.h"
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 
84 #include "monitor/monitor.h"
85 
86 #include <libfdt.h>
87 
88 /* SLOF memory layout:
89  *
90  * SLOF raw image loaded at 0, copies its romfs right below the flat
91  * device-tree, then position SLOF itself 31M below that
92  *
93  * So we set FW_OVERHEAD to 40MB which should account for all of that
94  * and more
95  *
96  * We load our kernel at 4M, leaving space for SLOF initial image
97  */
98 #define FDT_MAX_SIZE            0x100000
99 #define RTAS_MAX_SIZE           0x10000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            128UL
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
222                                        PowerPCCPU *cpu,
223                                        void *fdt, int offset)
224 {
225     uint8_t pa_features_206[] = { 6, 0,
226         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227     uint8_t pa_features_207[] = { 24, 0,
228         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232     uint8_t pa_features_300[] = { 66, 0,
233         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
236         /* 6: DS207 */
237         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
238         /* 16: Vector */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248         /* 42: PM, 44: PC RA, 46: SC vec'd */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250         /* 48: SIMD, 50: QP BFP, 52: String */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252         /* 54: DecFP, 56: DecI, 58: SHA */
253         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254         /* 60: NM atomic, 62: RNG */
255         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
256     };
257     uint8_t *pa_features = NULL;
258     size_t pa_size;
259 
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_206;
262         pa_size = sizeof(pa_features_206);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_207;
266         pa_size = sizeof(pa_features_207);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_300;
270         pa_size = sizeof(pa_features_300);
271     }
272     if (!pa_features) {
273         return;
274     }
275 
276     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
277         /*
278          * Note: we keep CI large pages off by default because a 64K capable
279          * guest provisioned with large pages might otherwise try to map a qemu
280          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281          * even if that qemu runs on a 4k host.
282          * We dd this bit back here if we are confident this is not an issue
283          */
284         pa_features[3] |= 0x20;
285     }
286     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287         pa_features[24] |= 0x80;    /* Transactional memory support */
288     }
289     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290         /* Workaround for broken kernels that attempt (guest) radix
291          * mode when they can't handle it, if they see the radix bit set
292          * in pa-features. So hide it from them. */
293         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
294     }
295 
296     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
297 }
298 
299 static int spapr_fixup_cpu_dt(void *fdt, SpaprMachineState *spapr)
300 {
301     MachineState *ms = MACHINE(spapr);
302     int ret = 0, offset, cpus_offset;
303     CPUState *cs;
304     char cpu_model[32];
305     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
306 
307     CPU_FOREACH(cs) {
308         PowerPCCPU *cpu = POWERPC_CPU(cs);
309         DeviceClass *dc = DEVICE_GET_CLASS(cs);
310         int index = spapr_get_vcpu_id(cpu);
311         int compat_smt = MIN(ms->smp.threads, ppc_compat_max_vthreads(cpu));
312 
313         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
314             continue;
315         }
316 
317         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
318 
319         cpus_offset = fdt_path_offset(fdt, "/cpus");
320         if (cpus_offset < 0) {
321             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
322             if (cpus_offset < 0) {
323                 return cpus_offset;
324             }
325         }
326         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
327         if (offset < 0) {
328             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
329             if (offset < 0) {
330                 return offset;
331             }
332         }
333 
334         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
335                           pft_size_prop, sizeof(pft_size_prop));
336         if (ret < 0) {
337             return ret;
338         }
339 
340         if (ms->numa_state->num_nodes > 1) {
341             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
342             if (ret < 0) {
343                 return ret;
344             }
345         }
346 
347         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
348         if (ret < 0) {
349             return ret;
350         }
351 
352         spapr_populate_pa_features(spapr, cpu, fdt, offset);
353     }
354     return ret;
355 }
356 
357 static hwaddr spapr_node0_size(MachineState *machine)
358 {
359     if (machine->numa_state->num_nodes) {
360         int i;
361         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
362             if (machine->numa_state->nodes[i].node_mem) {
363                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
364                            machine->ram_size);
365             }
366         }
367     }
368     return machine->ram_size;
369 }
370 
371 static void add_str(GString *s, const gchar *s1)
372 {
373     g_string_append_len(s, s1, strlen(s1) + 1);
374 }
375 
376 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
377                                        hwaddr size)
378 {
379     uint32_t associativity[] = {
380         cpu_to_be32(0x4), /* length */
381         cpu_to_be32(0x0), cpu_to_be32(0x0),
382         cpu_to_be32(0x0), cpu_to_be32(nodeid)
383     };
384     char mem_name[32];
385     uint64_t mem_reg_property[2];
386     int off;
387 
388     mem_reg_property[0] = cpu_to_be64(start);
389     mem_reg_property[1] = cpu_to_be64(size);
390 
391     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
392     off = fdt_add_subnode(fdt, 0, mem_name);
393     _FDT(off);
394     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
395     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
396                       sizeof(mem_reg_property))));
397     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
398                       sizeof(associativity))));
399     return off;
400 }
401 
402 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
403 {
404     MachineState *machine = MACHINE(spapr);
405     hwaddr mem_start, node_size;
406     int i, nb_nodes = machine->numa_state->num_nodes;
407     NodeInfo *nodes = machine->numa_state->nodes;
408     NodeInfo ramnode;
409 
410     /* No NUMA nodes, assume there is just one node with whole RAM */
411     if (!nb_nodes) {
412         nb_nodes = 1;
413         ramnode.node_mem = machine->ram_size;
414         nodes = &ramnode;
415     }
416 
417     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
418         if (!nodes[i].node_mem) {
419             continue;
420         }
421         if (mem_start >= machine->ram_size) {
422             node_size = 0;
423         } else {
424             node_size = nodes[i].node_mem;
425             if (node_size > machine->ram_size - mem_start) {
426                 node_size = machine->ram_size - mem_start;
427             }
428         }
429         if (!mem_start) {
430             /* spapr_machine_init() checks for rma_size <= node0_size
431              * already */
432             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
433             mem_start += spapr->rma_size;
434             node_size -= spapr->rma_size;
435         }
436         for ( ; node_size; ) {
437             hwaddr sizetmp = pow2floor(node_size);
438 
439             /* mem_start != 0 here */
440             if (ctzl(mem_start) < ctzl(sizetmp)) {
441                 sizetmp = 1ULL << ctzl(mem_start);
442             }
443 
444             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
445             node_size -= sizetmp;
446             mem_start += sizetmp;
447         }
448     }
449 
450     return 0;
451 }
452 
453 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
454                                   SpaprMachineState *spapr)
455 {
456     MachineState *ms = MACHINE(spapr);
457     PowerPCCPU *cpu = POWERPC_CPU(cs);
458     CPUPPCState *env = &cpu->env;
459     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
460     int index = spapr_get_vcpu_id(cpu);
461     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
462                        0xffffffff, 0xffffffff};
463     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
464         : SPAPR_TIMEBASE_FREQ;
465     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
466     uint32_t page_sizes_prop[64];
467     size_t page_sizes_prop_size;
468     unsigned int smp_threads = ms->smp.threads;
469     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
470     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
471     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
472     SpaprDrc *drc;
473     int drc_index;
474     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
475     int i;
476 
477     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
478     if (drc) {
479         drc_index = spapr_drc_index(drc);
480         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
481     }
482 
483     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
484     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
485 
486     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
487     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
488                            env->dcache_line_size)));
489     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
490                            env->dcache_line_size)));
491     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
492                            env->icache_line_size)));
493     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
494                            env->icache_line_size)));
495 
496     if (pcc->l1_dcache_size) {
497         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
498                                pcc->l1_dcache_size)));
499     } else {
500         warn_report("Unknown L1 dcache size for cpu");
501     }
502     if (pcc->l1_icache_size) {
503         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
504                                pcc->l1_icache_size)));
505     } else {
506         warn_report("Unknown L1 icache size for cpu");
507     }
508 
509     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
510     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
511     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
512     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
513     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
514     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
515 
516     if (env->spr_cb[SPR_PURR].oea_read) {
517         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
518     }
519     if (env->spr_cb[SPR_SPURR].oea_read) {
520         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
521     }
522 
523     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
524         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
525                           segs, sizeof(segs))));
526     }
527 
528     /* Advertise VSX (vector extensions) if available
529      *   1               == VMX / Altivec available
530      *   2               == VSX available
531      *
532      * Only CPUs for which we create core types in spapr_cpu_core.c
533      * are possible, and all of those have VMX */
534     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
535         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
536     } else {
537         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
538     }
539 
540     /* Advertise DFP (Decimal Floating Point) if available
541      *   0 / no property == no DFP
542      *   1               == DFP available */
543     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
544         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
545     }
546 
547     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
548                                                       sizeof(page_sizes_prop));
549     if (page_sizes_prop_size) {
550         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
551                           page_sizes_prop, page_sizes_prop_size)));
552     }
553 
554     spapr_populate_pa_features(spapr, cpu, fdt, offset);
555 
556     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
557                            cs->cpu_index / vcpus_per_socket)));
558 
559     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
560                       pft_size_prop, sizeof(pft_size_prop))));
561 
562     if (ms->numa_state->num_nodes > 1) {
563         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
564     }
565 
566     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
567 
568     if (pcc->radix_page_info) {
569         for (i = 0; i < pcc->radix_page_info->count; i++) {
570             radix_AP_encodings[i] =
571                 cpu_to_be32(pcc->radix_page_info->entries[i]);
572         }
573         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
574                           radix_AP_encodings,
575                           pcc->radix_page_info->count *
576                           sizeof(radix_AP_encodings[0]))));
577     }
578 
579     /*
580      * We set this property to let the guest know that it can use the large
581      * decrementer and its width in bits.
582      */
583     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
584         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
585                               pcc->lrg_decr_bits)));
586 }
587 
588 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
589 {
590     CPUState **rev;
591     CPUState *cs;
592     int n_cpus;
593     int cpus_offset;
594     char *nodename;
595     int i;
596 
597     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
598     _FDT(cpus_offset);
599     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
600     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
601 
602     /*
603      * We walk the CPUs in reverse order to ensure that CPU DT nodes
604      * created by fdt_add_subnode() end up in the right order in FDT
605      * for the guest kernel the enumerate the CPUs correctly.
606      *
607      * The CPU list cannot be traversed in reverse order, so we need
608      * to do extra work.
609      */
610     n_cpus = 0;
611     rev = NULL;
612     CPU_FOREACH(cs) {
613         rev = g_renew(CPUState *, rev, n_cpus + 1);
614         rev[n_cpus++] = cs;
615     }
616 
617     for (i = n_cpus - 1; i >= 0; i--) {
618         CPUState *cs = rev[i];
619         PowerPCCPU *cpu = POWERPC_CPU(cs);
620         int index = spapr_get_vcpu_id(cpu);
621         DeviceClass *dc = DEVICE_GET_CLASS(cs);
622         int offset;
623 
624         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
625             continue;
626         }
627 
628         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
629         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
630         g_free(nodename);
631         _FDT(offset);
632         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
633     }
634 
635     g_free(rev);
636 }
637 
638 static int spapr_rng_populate_dt(void *fdt)
639 {
640     int node;
641     int ret;
642 
643     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
644     if (node <= 0) {
645         return -1;
646     }
647     ret = fdt_setprop_string(fdt, node, "device_type",
648                              "ibm,platform-facilities");
649     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
650     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
651 
652     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
653     if (node <= 0) {
654         return -1;
655     }
656     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
657 
658     return ret ? -1 : 0;
659 }
660 
661 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
662 {
663     MemoryDeviceInfoList *info;
664 
665     for (info = list; info; info = info->next) {
666         MemoryDeviceInfo *value = info->value;
667 
668         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
669             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
670 
671             if (addr >= pcdimm_info->addr &&
672                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
673                 return pcdimm_info->node;
674             }
675         }
676     }
677 
678     return -1;
679 }
680 
681 struct sPAPRDrconfCellV2 {
682      uint32_t seq_lmbs;
683      uint64_t base_addr;
684      uint32_t drc_index;
685      uint32_t aa_index;
686      uint32_t flags;
687 } QEMU_PACKED;
688 
689 typedef struct DrconfCellQueue {
690     struct sPAPRDrconfCellV2 cell;
691     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
692 } DrconfCellQueue;
693 
694 static DrconfCellQueue *
695 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
696                       uint32_t drc_index, uint32_t aa_index,
697                       uint32_t flags)
698 {
699     DrconfCellQueue *elem;
700 
701     elem = g_malloc0(sizeof(*elem));
702     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
703     elem->cell.base_addr = cpu_to_be64(base_addr);
704     elem->cell.drc_index = cpu_to_be32(drc_index);
705     elem->cell.aa_index = cpu_to_be32(aa_index);
706     elem->cell.flags = cpu_to_be32(flags);
707 
708     return elem;
709 }
710 
711 /* ibm,dynamic-memory-v2 */
712 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
713                                    int offset, MemoryDeviceInfoList *dimms)
714 {
715     MachineState *machine = MACHINE(spapr);
716     uint8_t *int_buf, *cur_index;
717     int ret;
718     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
719     uint64_t addr, cur_addr, size;
720     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
721     uint64_t mem_end = machine->device_memory->base +
722                        memory_region_size(&machine->device_memory->mr);
723     uint32_t node, buf_len, nr_entries = 0;
724     SpaprDrc *drc;
725     DrconfCellQueue *elem, *next;
726     MemoryDeviceInfoList *info;
727     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
728         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
729 
730     /* Entry to cover RAM and the gap area */
731     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
732                                  SPAPR_LMB_FLAGS_RESERVED |
733                                  SPAPR_LMB_FLAGS_DRC_INVALID);
734     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
735     nr_entries++;
736 
737     cur_addr = machine->device_memory->base;
738     for (info = dimms; info; info = info->next) {
739         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
740 
741         addr = di->addr;
742         size = di->size;
743         node = di->node;
744 
745         /* Entry for hot-pluggable area */
746         if (cur_addr < addr) {
747             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
748             g_assert(drc);
749             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
750                                          cur_addr, spapr_drc_index(drc), -1, 0);
751             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
752             nr_entries++;
753         }
754 
755         /* Entry for DIMM */
756         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
757         g_assert(drc);
758         elem = spapr_get_drconf_cell(size / lmb_size, addr,
759                                      spapr_drc_index(drc), node,
760                                      SPAPR_LMB_FLAGS_ASSIGNED);
761         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
762         nr_entries++;
763         cur_addr = addr + size;
764     }
765 
766     /* Entry for remaining hotpluggable area */
767     if (cur_addr < mem_end) {
768         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
769         g_assert(drc);
770         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
771                                      cur_addr, spapr_drc_index(drc), -1, 0);
772         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
773         nr_entries++;
774     }
775 
776     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
777     int_buf = cur_index = g_malloc0(buf_len);
778     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
779     cur_index += sizeof(nr_entries);
780 
781     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
782         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
783         cur_index += sizeof(elem->cell);
784         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
785         g_free(elem);
786     }
787 
788     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
789     g_free(int_buf);
790     if (ret < 0) {
791         return -1;
792     }
793     return 0;
794 }
795 
796 /* ibm,dynamic-memory */
797 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
798                                    int offset, MemoryDeviceInfoList *dimms)
799 {
800     MachineState *machine = MACHINE(spapr);
801     int i, ret;
802     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
803     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
804     uint32_t nr_lmbs = (machine->device_memory->base +
805                        memory_region_size(&machine->device_memory->mr)) /
806                        lmb_size;
807     uint32_t *int_buf, *cur_index, buf_len;
808 
809     /*
810      * Allocate enough buffer size to fit in ibm,dynamic-memory
811      */
812     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
813     cur_index = int_buf = g_malloc0(buf_len);
814     int_buf[0] = cpu_to_be32(nr_lmbs);
815     cur_index++;
816     for (i = 0; i < nr_lmbs; i++) {
817         uint64_t addr = i * lmb_size;
818         uint32_t *dynamic_memory = cur_index;
819 
820         if (i >= device_lmb_start) {
821             SpaprDrc *drc;
822 
823             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
824             g_assert(drc);
825 
826             dynamic_memory[0] = cpu_to_be32(addr >> 32);
827             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
828             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
829             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
830             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
831             if (memory_region_present(get_system_memory(), addr)) {
832                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
833             } else {
834                 dynamic_memory[5] = cpu_to_be32(0);
835             }
836         } else {
837             /*
838              * LMB information for RMA, boot time RAM and gap b/n RAM and
839              * device memory region -- all these are marked as reserved
840              * and as having no valid DRC.
841              */
842             dynamic_memory[0] = cpu_to_be32(addr >> 32);
843             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
844             dynamic_memory[2] = cpu_to_be32(0);
845             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
846             dynamic_memory[4] = cpu_to_be32(-1);
847             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
848                                             SPAPR_LMB_FLAGS_DRC_INVALID);
849         }
850 
851         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
852     }
853     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
854     g_free(int_buf);
855     if (ret < 0) {
856         return -1;
857     }
858     return 0;
859 }
860 
861 /*
862  * Adds ibm,dynamic-reconfiguration-memory node.
863  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
864  * of this device tree node.
865  */
866 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
867 {
868     MachineState *machine = MACHINE(spapr);
869     int nb_numa_nodes = machine->numa_state->num_nodes;
870     int ret, i, offset;
871     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
872     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
873     uint32_t *int_buf, *cur_index, buf_len;
874     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
875     MemoryDeviceInfoList *dimms = NULL;
876 
877     /*
878      * Don't create the node if there is no device memory
879      */
880     if (machine->ram_size == machine->maxram_size) {
881         return 0;
882     }
883 
884     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
885 
886     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
887                     sizeof(prop_lmb_size));
888     if (ret < 0) {
889         return ret;
890     }
891 
892     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
893     if (ret < 0) {
894         return ret;
895     }
896 
897     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
898     if (ret < 0) {
899         return ret;
900     }
901 
902     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
903     dimms = qmp_memory_device_list();
904     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
905         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
906     } else {
907         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
908     }
909     qapi_free_MemoryDeviceInfoList(dimms);
910 
911     if (ret < 0) {
912         return ret;
913     }
914 
915     /* ibm,associativity-lookup-arrays */
916     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
917     cur_index = int_buf = g_malloc0(buf_len);
918     int_buf[0] = cpu_to_be32(nr_nodes);
919     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
920     cur_index += 2;
921     for (i = 0; i < nr_nodes; i++) {
922         uint32_t associativity[] = {
923             cpu_to_be32(0x0),
924             cpu_to_be32(0x0),
925             cpu_to_be32(0x0),
926             cpu_to_be32(i)
927         };
928         memcpy(cur_index, associativity, sizeof(associativity));
929         cur_index += 4;
930     }
931     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
932             (cur_index - int_buf) * sizeof(uint32_t));
933     g_free(int_buf);
934 
935     return ret;
936 }
937 
938 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
939                                 SpaprOptionVector *ov5_updates)
940 {
941     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
942     int ret = 0, offset;
943 
944     /* Generate ibm,dynamic-reconfiguration-memory node if required */
945     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
946         g_assert(smc->dr_lmb_enabled);
947         ret = spapr_populate_drconf_memory(spapr, fdt);
948         if (ret) {
949             goto out;
950         }
951     }
952 
953     offset = fdt_path_offset(fdt, "/chosen");
954     if (offset < 0) {
955         offset = fdt_add_subnode(fdt, 0, "chosen");
956         if (offset < 0) {
957             return offset;
958         }
959     }
960     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
961                                  "ibm,architecture-vec-5");
962 
963 out:
964     return ret;
965 }
966 
967 static bool spapr_hotplugged_dev_before_cas(void)
968 {
969     Object *drc_container, *obj;
970     ObjectProperty *prop;
971     ObjectPropertyIterator iter;
972 
973     drc_container = container_get(object_get_root(), "/dr-connector");
974     object_property_iter_init(&iter, drc_container);
975     while ((prop = object_property_iter_next(&iter))) {
976         if (!strstart(prop->type, "link<", NULL)) {
977             continue;
978         }
979         obj = object_property_get_link(drc_container, prop->name, NULL);
980         if (spapr_drc_needed(obj)) {
981             return true;
982         }
983     }
984     return false;
985 }
986 
987 int spapr_h_cas_compose_response(SpaprMachineState *spapr,
988                                  target_ulong addr, target_ulong size,
989                                  SpaprOptionVector *ov5_updates)
990 {
991     void *fdt, *fdt_skel;
992     SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
993 
994     if (spapr_hotplugged_dev_before_cas()) {
995         return 1;
996     }
997 
998     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
999         error_report("SLOF provided an unexpected CAS buffer size "
1000                      TARGET_FMT_lu " (min: %zu, max: %u)",
1001                      size, sizeof(hdr), FW_MAX_SIZE);
1002         exit(EXIT_FAILURE);
1003     }
1004 
1005     size -= sizeof(hdr);
1006 
1007     /* Create skeleton */
1008     fdt_skel = g_malloc0(size);
1009     _FDT((fdt_create(fdt_skel, size)));
1010     _FDT((fdt_finish_reservemap(fdt_skel)));
1011     _FDT((fdt_begin_node(fdt_skel, "")));
1012     _FDT((fdt_end_node(fdt_skel)));
1013     _FDT((fdt_finish(fdt_skel)));
1014     fdt = g_malloc0(size);
1015     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1016     g_free(fdt_skel);
1017 
1018     /* Fixup cpu nodes */
1019     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1020 
1021     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1022         return -1;
1023     }
1024 
1025     /* Pack resulting tree */
1026     _FDT((fdt_pack(fdt)));
1027 
1028     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1029         g_free(fdt);
1030         trace_spapr_cas_failed(size);
1031         return -1;
1032     }
1033 
1034     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1035     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1036     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1037     g_free(fdt);
1038 
1039     return 0;
1040 }
1041 
1042 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
1043 {
1044     MachineState *ms = MACHINE(spapr);
1045     int rtas;
1046     GString *hypertas = g_string_sized_new(256);
1047     GString *qemu_hypertas = g_string_sized_new(256);
1048     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1049     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1050         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1051     uint32_t lrdr_capacity[] = {
1052         cpu_to_be32(max_device_addr >> 32),
1053         cpu_to_be32(max_device_addr & 0xffffffff),
1054         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1055         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
1056     };
1057     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
1058     uint32_t maxdomains[] = {
1059         cpu_to_be32(4),
1060         maxdomain,
1061         maxdomain,
1062         maxdomain,
1063         cpu_to_be32(spapr->gpu_numa_id),
1064     };
1065 
1066     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1067 
1068     /* hypertas */
1069     add_str(hypertas, "hcall-pft");
1070     add_str(hypertas, "hcall-term");
1071     add_str(hypertas, "hcall-dabr");
1072     add_str(hypertas, "hcall-interrupt");
1073     add_str(hypertas, "hcall-tce");
1074     add_str(hypertas, "hcall-vio");
1075     add_str(hypertas, "hcall-splpar");
1076     add_str(hypertas, "hcall-join");
1077     add_str(hypertas, "hcall-bulk");
1078     add_str(hypertas, "hcall-set-mode");
1079     add_str(hypertas, "hcall-sprg0");
1080     add_str(hypertas, "hcall-copy");
1081     add_str(hypertas, "hcall-debug");
1082     add_str(hypertas, "hcall-vphn");
1083     add_str(qemu_hypertas, "hcall-memop1");
1084 
1085     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1086         add_str(hypertas, "hcall-multi-tce");
1087     }
1088 
1089     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1090         add_str(hypertas, "hcall-hpt-resize");
1091     }
1092 
1093     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1094                      hypertas->str, hypertas->len));
1095     g_string_free(hypertas, TRUE);
1096     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1097                      qemu_hypertas->str, qemu_hypertas->len));
1098     g_string_free(qemu_hypertas, TRUE);
1099 
1100     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1101                      refpoints, sizeof(refpoints)));
1102 
1103     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1104                      maxdomains, sizeof(maxdomains)));
1105 
1106     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1107                           RTAS_ERROR_LOG_MAX));
1108     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1109                           RTAS_EVENT_SCAN_RATE));
1110 
1111     g_assert(msi_nonbroken);
1112     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1113 
1114     /*
1115      * According to PAPR, rtas ibm,os-term does not guarantee a return
1116      * back to the guest cpu.
1117      *
1118      * While an additional ibm,extended-os-term property indicates
1119      * that rtas call return will always occur. Set this property.
1120      */
1121     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1122 
1123     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1124                      lrdr_capacity, sizeof(lrdr_capacity)));
1125 
1126     spapr_dt_rtas_tokens(fdt, rtas);
1127 }
1128 
1129 /*
1130  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1131  * and the XIVE features that the guest may request and thus the valid
1132  * values for bytes 23..26 of option vector 5:
1133  */
1134 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1135                                           int chosen)
1136 {
1137     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1138 
1139     char val[2 * 4] = {
1140         23, spapr->irq->ov5, /* Xive mode. */
1141         24, 0x00, /* Hash/Radix, filled in below. */
1142         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1143         26, 0x40, /* Radix options: GTSE == yes. */
1144     };
1145 
1146     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1147                           first_ppc_cpu->compat_pvr)) {
1148         /*
1149          * If we're in a pre POWER9 compat mode then the guest should
1150          * do hash and use the legacy interrupt mode
1151          */
1152         val[1] = 0x00; /* XICS */
1153         val[3] = 0x00; /* Hash */
1154     } else if (kvm_enabled()) {
1155         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1156             val[3] = 0x80; /* OV5_MMU_BOTH */
1157         } else if (kvmppc_has_cap_mmu_radix()) {
1158             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1159         } else {
1160             val[3] = 0x00; /* Hash */
1161         }
1162     } else {
1163         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1164         val[3] = 0xC0;
1165     }
1166     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1167                      val, sizeof(val)));
1168 }
1169 
1170 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1171 {
1172     MachineState *machine = MACHINE(spapr);
1173     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1174     int chosen;
1175     const char *boot_device = machine->boot_order;
1176     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1177     size_t cb = 0;
1178     char *bootlist = get_boot_devices_list(&cb);
1179 
1180     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1181 
1182     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1183         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1184                                 machine->kernel_cmdline));
1185     }
1186     if (spapr->initrd_size) {
1187         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1188                               spapr->initrd_base));
1189         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1190                               spapr->initrd_base + spapr->initrd_size));
1191     }
1192 
1193     if (spapr->kernel_size) {
1194         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1195                               cpu_to_be64(spapr->kernel_size) };
1196 
1197         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1198                          &kprop, sizeof(kprop)));
1199         if (spapr->kernel_le) {
1200             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1201         }
1202     }
1203     if (boot_menu) {
1204         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1205     }
1206     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1207     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1208     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1209 
1210     if (cb && bootlist) {
1211         int i;
1212 
1213         for (i = 0; i < cb; i++) {
1214             if (bootlist[i] == '\n') {
1215                 bootlist[i] = ' ';
1216             }
1217         }
1218         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1219     }
1220 
1221     if (boot_device && strlen(boot_device)) {
1222         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1223     }
1224 
1225     if (!spapr->has_graphics && stdout_path) {
1226         /*
1227          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1228          * kernel. New platforms should only use the "stdout-path" property. Set
1229          * the new property and continue using older property to remain
1230          * compatible with the existing firmware.
1231          */
1232         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1233         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1234     }
1235 
1236     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1237     if (smc->linux_pci_probe) {
1238         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1239     }
1240 
1241     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1242 
1243     g_free(stdout_path);
1244     g_free(bootlist);
1245 }
1246 
1247 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1248 {
1249     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1250      * KVM to work under pHyp with some guest co-operation */
1251     int hypervisor;
1252     uint8_t hypercall[16];
1253 
1254     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1255     /* indicate KVM hypercall interface */
1256     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1257     if (kvmppc_has_cap_fixup_hcalls()) {
1258         /*
1259          * Older KVM versions with older guest kernels were broken
1260          * with the magic page, don't allow the guest to map it.
1261          */
1262         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1263                                   sizeof(hypercall))) {
1264             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1265                              hypercall, sizeof(hypercall)));
1266         }
1267     }
1268 }
1269 
1270 static void *spapr_build_fdt(SpaprMachineState *spapr)
1271 {
1272     MachineState *machine = MACHINE(spapr);
1273     MachineClass *mc = MACHINE_GET_CLASS(machine);
1274     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1275     int ret;
1276     void *fdt;
1277     SpaprPhbState *phb;
1278     char *buf;
1279 
1280     fdt = g_malloc0(FDT_MAX_SIZE);
1281     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1282 
1283     /* Root node */
1284     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1285     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1286     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1287 
1288     /* Guest UUID & Name*/
1289     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1290     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1291     if (qemu_uuid_set) {
1292         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1293     }
1294     g_free(buf);
1295 
1296     if (qemu_get_vm_name()) {
1297         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1298                                 qemu_get_vm_name()));
1299     }
1300 
1301     /* Host Model & Serial Number */
1302     if (spapr->host_model) {
1303         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1304     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1305         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1306         g_free(buf);
1307     }
1308 
1309     if (spapr->host_serial) {
1310         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1311     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1312         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1313         g_free(buf);
1314     }
1315 
1316     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1317     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1318 
1319     /* /interrupt controller */
1320     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1321                           PHANDLE_INTC);
1322 
1323     ret = spapr_populate_memory(spapr, fdt);
1324     if (ret < 0) {
1325         error_report("couldn't setup memory nodes in fdt");
1326         exit(1);
1327     }
1328 
1329     /* /vdevice */
1330     spapr_dt_vdevice(spapr->vio_bus, fdt);
1331 
1332     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1333         ret = spapr_rng_populate_dt(fdt);
1334         if (ret < 0) {
1335             error_report("could not set up rng device in the fdt");
1336             exit(1);
1337         }
1338     }
1339 
1340     QLIST_FOREACH(phb, &spapr->phbs, list) {
1341         ret = spapr_dt_phb(phb, PHANDLE_INTC, fdt, spapr->irq->nr_msis, NULL);
1342         if (ret < 0) {
1343             error_report("couldn't setup PCI devices in fdt");
1344             exit(1);
1345         }
1346     }
1347 
1348     /* cpus */
1349     spapr_populate_cpus_dt_node(fdt, spapr);
1350 
1351     if (smc->dr_lmb_enabled) {
1352         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1353     }
1354 
1355     if (mc->has_hotpluggable_cpus) {
1356         int offset = fdt_path_offset(fdt, "/cpus");
1357         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1358         if (ret < 0) {
1359             error_report("Couldn't set up CPU DR device tree properties");
1360             exit(1);
1361         }
1362     }
1363 
1364     /* /event-sources */
1365     spapr_dt_events(spapr, fdt);
1366 
1367     /* /rtas */
1368     spapr_dt_rtas(spapr, fdt);
1369 
1370     /* /chosen */
1371     spapr_dt_chosen(spapr, fdt);
1372 
1373     /* /hypervisor */
1374     if (kvm_enabled()) {
1375         spapr_dt_hypervisor(spapr, fdt);
1376     }
1377 
1378     /* Build memory reserve map */
1379     if (spapr->kernel_size) {
1380         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1381     }
1382     if (spapr->initrd_size) {
1383         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1384     }
1385 
1386     /* ibm,client-architecture-support updates */
1387     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1388     if (ret < 0) {
1389         error_report("couldn't setup CAS properties fdt");
1390         exit(1);
1391     }
1392 
1393     if (smc->dr_phb_enabled) {
1394         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1395         if (ret < 0) {
1396             error_report("Couldn't set up PHB DR device tree properties");
1397             exit(1);
1398         }
1399     }
1400 
1401     return fdt;
1402 }
1403 
1404 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1405 {
1406     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1407 }
1408 
1409 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1410                                     PowerPCCPU *cpu)
1411 {
1412     CPUPPCState *env = &cpu->env;
1413 
1414     /* The TCG path should also be holding the BQL at this point */
1415     g_assert(qemu_mutex_iothread_locked());
1416 
1417     if (msr_pr) {
1418         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1419         env->gpr[3] = H_PRIVILEGE;
1420     } else {
1421         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1422     }
1423 }
1424 
1425 struct LPCRSyncState {
1426     target_ulong value;
1427     target_ulong mask;
1428 };
1429 
1430 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1431 {
1432     struct LPCRSyncState *s = arg.host_ptr;
1433     PowerPCCPU *cpu = POWERPC_CPU(cs);
1434     CPUPPCState *env = &cpu->env;
1435     target_ulong lpcr;
1436 
1437     cpu_synchronize_state(cs);
1438     lpcr = env->spr[SPR_LPCR];
1439     lpcr &= ~s->mask;
1440     lpcr |= s->value;
1441     ppc_store_lpcr(cpu, lpcr);
1442 }
1443 
1444 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1445 {
1446     CPUState *cs;
1447     struct LPCRSyncState s = {
1448         .value = value,
1449         .mask = mask
1450     };
1451     CPU_FOREACH(cs) {
1452         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1453     }
1454 }
1455 
1456 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1457 {
1458     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1459 
1460     /* Copy PATE1:GR into PATE0:HR */
1461     entry->dw0 = spapr->patb_entry & PATE0_HR;
1462     entry->dw1 = spapr->patb_entry;
1463 }
1464 
1465 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1466 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1467 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1468 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1469 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1470 
1471 /*
1472  * Get the fd to access the kernel htab, re-opening it if necessary
1473  */
1474 static int get_htab_fd(SpaprMachineState *spapr)
1475 {
1476     Error *local_err = NULL;
1477 
1478     if (spapr->htab_fd >= 0) {
1479         return spapr->htab_fd;
1480     }
1481 
1482     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1483     if (spapr->htab_fd < 0) {
1484         error_report_err(local_err);
1485     }
1486 
1487     return spapr->htab_fd;
1488 }
1489 
1490 void close_htab_fd(SpaprMachineState *spapr)
1491 {
1492     if (spapr->htab_fd >= 0) {
1493         close(spapr->htab_fd);
1494     }
1495     spapr->htab_fd = -1;
1496 }
1497 
1498 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1499 {
1500     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1501 
1502     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1503 }
1504 
1505 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1506 {
1507     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1508 
1509     assert(kvm_enabled());
1510 
1511     if (!spapr->htab) {
1512         return 0;
1513     }
1514 
1515     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1516 }
1517 
1518 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1519                                                 hwaddr ptex, int n)
1520 {
1521     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1522     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1523 
1524     if (!spapr->htab) {
1525         /*
1526          * HTAB is controlled by KVM. Fetch into temporary buffer
1527          */
1528         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1529         kvmppc_read_hptes(hptes, ptex, n);
1530         return hptes;
1531     }
1532 
1533     /*
1534      * HTAB is controlled by QEMU. Just point to the internally
1535      * accessible PTEG.
1536      */
1537     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1538 }
1539 
1540 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1541                               const ppc_hash_pte64_t *hptes,
1542                               hwaddr ptex, int n)
1543 {
1544     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1545 
1546     if (!spapr->htab) {
1547         g_free((void *)hptes);
1548     }
1549 
1550     /* Nothing to do for qemu managed HPT */
1551 }
1552 
1553 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1554                       uint64_t pte0, uint64_t pte1)
1555 {
1556     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1557     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1558 
1559     if (!spapr->htab) {
1560         kvmppc_write_hpte(ptex, pte0, pte1);
1561     } else {
1562         if (pte0 & HPTE64_V_VALID) {
1563             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1564             /*
1565              * When setting valid, we write PTE1 first. This ensures
1566              * proper synchronization with the reading code in
1567              * ppc_hash64_pteg_search()
1568              */
1569             smp_wmb();
1570             stq_p(spapr->htab + offset, pte0);
1571         } else {
1572             stq_p(spapr->htab + offset, pte0);
1573             /*
1574              * When clearing it we set PTE0 first. This ensures proper
1575              * synchronization with the reading code in
1576              * ppc_hash64_pteg_search()
1577              */
1578             smp_wmb();
1579             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1580         }
1581     }
1582 }
1583 
1584 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1585                              uint64_t pte1)
1586 {
1587     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1588     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1589 
1590     if (!spapr->htab) {
1591         /* There should always be a hash table when this is called */
1592         error_report("spapr_hpte_set_c called with no hash table !");
1593         return;
1594     }
1595 
1596     /* The HW performs a non-atomic byte update */
1597     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1598 }
1599 
1600 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1601                              uint64_t pte1)
1602 {
1603     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1604     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1605 
1606     if (!spapr->htab) {
1607         /* There should always be a hash table when this is called */
1608         error_report("spapr_hpte_set_r called with no hash table !");
1609         return;
1610     }
1611 
1612     /* The HW performs a non-atomic byte update */
1613     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1614 }
1615 
1616 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1617 {
1618     int shift;
1619 
1620     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1621      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1622      * that's much more than is needed for Linux guests */
1623     shift = ctz64(pow2ceil(ramsize)) - 7;
1624     shift = MAX(shift, 18); /* Minimum architected size */
1625     shift = MIN(shift, 46); /* Maximum architected size */
1626     return shift;
1627 }
1628 
1629 void spapr_free_hpt(SpaprMachineState *spapr)
1630 {
1631     g_free(spapr->htab);
1632     spapr->htab = NULL;
1633     spapr->htab_shift = 0;
1634     close_htab_fd(spapr);
1635 }
1636 
1637 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1638                           Error **errp)
1639 {
1640     long rc;
1641 
1642     /* Clean up any HPT info from a previous boot */
1643     spapr_free_hpt(spapr);
1644 
1645     rc = kvmppc_reset_htab(shift);
1646     if (rc < 0) {
1647         /* kernel-side HPT needed, but couldn't allocate one */
1648         error_setg_errno(errp, errno,
1649                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1650                          shift);
1651         /* This is almost certainly fatal, but if the caller really
1652          * wants to carry on with shift == 0, it's welcome to try */
1653     } else if (rc > 0) {
1654         /* kernel-side HPT allocated */
1655         if (rc != shift) {
1656             error_setg(errp,
1657                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1658                        shift, rc);
1659         }
1660 
1661         spapr->htab_shift = shift;
1662         spapr->htab = NULL;
1663     } else {
1664         /* kernel-side HPT not needed, allocate in userspace instead */
1665         size_t size = 1ULL << shift;
1666         int i;
1667 
1668         spapr->htab = qemu_memalign(size, size);
1669         if (!spapr->htab) {
1670             error_setg_errno(errp, errno,
1671                              "Could not allocate HPT of order %d", shift);
1672             return;
1673         }
1674 
1675         memset(spapr->htab, 0, size);
1676         spapr->htab_shift = shift;
1677 
1678         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1679             DIRTY_HPTE(HPTE(spapr->htab, i));
1680         }
1681     }
1682     /* We're setting up a hash table, so that means we're not radix */
1683     spapr->patb_entry = 0;
1684     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1685 }
1686 
1687 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1688 {
1689     int hpt_shift;
1690 
1691     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1692         || (spapr->cas_reboot
1693             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1694         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1695     } else {
1696         uint64_t current_ram_size;
1697 
1698         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1699         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1700     }
1701     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1702 
1703     if (spapr->vrma_adjust) {
1704         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1705                                           spapr->htab_shift);
1706     }
1707 }
1708 
1709 static int spapr_reset_drcs(Object *child, void *opaque)
1710 {
1711     SpaprDrc *drc =
1712         (SpaprDrc *) object_dynamic_cast(child,
1713                                                  TYPE_SPAPR_DR_CONNECTOR);
1714 
1715     if (drc) {
1716         spapr_drc_reset(drc);
1717     }
1718 
1719     return 0;
1720 }
1721 
1722 static void spapr_machine_reset(MachineState *machine)
1723 {
1724     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1725     PowerPCCPU *first_ppc_cpu;
1726     uint32_t rtas_limit;
1727     hwaddr rtas_addr, fdt_addr;
1728     void *fdt;
1729     int rc;
1730 
1731     spapr_caps_apply(spapr);
1732 
1733     first_ppc_cpu = POWERPC_CPU(first_cpu);
1734     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1735         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1736                               spapr->max_compat_pvr)) {
1737         /*
1738          * If using KVM with radix mode available, VCPUs can be started
1739          * without a HPT because KVM will start them in radix mode.
1740          * Set the GR bit in PATE so that we know there is no HPT.
1741          */
1742         spapr->patb_entry = PATE1_GR;
1743         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1744     } else {
1745         spapr_setup_hpt_and_vrma(spapr);
1746     }
1747 
1748     qemu_devices_reset();
1749 
1750     /*
1751      * If this reset wasn't generated by CAS, we should reset our
1752      * negotiated options and start from scratch
1753      */
1754     if (!spapr->cas_reboot) {
1755         spapr_ovec_cleanup(spapr->ov5_cas);
1756         spapr->ov5_cas = spapr_ovec_new();
1757 
1758         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1759     }
1760 
1761     /*
1762      * This is fixing some of the default configuration of the XIVE
1763      * devices. To be called after the reset of the machine devices.
1764      */
1765     spapr_irq_reset(spapr, &error_fatal);
1766 
1767     /*
1768      * There is no CAS under qtest. Simulate one to please the code that
1769      * depends on spapr->ov5_cas. This is especially needed to test device
1770      * unplug, so we do that before resetting the DRCs.
1771      */
1772     if (qtest_enabled()) {
1773         spapr_ovec_cleanup(spapr->ov5_cas);
1774         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1775     }
1776 
1777     /* DRC reset may cause a device to be unplugged. This will cause troubles
1778      * if this device is used by another device (eg, a running vhost backend
1779      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1780      * situations, we reset DRCs after all devices have been reset.
1781      */
1782     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1783 
1784     spapr_clear_pending_events(spapr);
1785 
1786     /*
1787      * We place the device tree and RTAS just below either the top of the RMA,
1788      * or just below 2GB, whichever is lower, so that it can be
1789      * processed with 32-bit real mode code if necessary
1790      */
1791     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1792     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1793     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1794 
1795     fdt = spapr_build_fdt(spapr);
1796 
1797     spapr_load_rtas(spapr, fdt, rtas_addr);
1798 
1799     rc = fdt_pack(fdt);
1800 
1801     /* Should only fail if we've built a corrupted tree */
1802     assert(rc == 0);
1803 
1804     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1805         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1806                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1807         exit(1);
1808     }
1809 
1810     /* Load the fdt */
1811     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1812     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1813     g_free(spapr->fdt_blob);
1814     spapr->fdt_size = fdt_totalsize(fdt);
1815     spapr->fdt_initial_size = spapr->fdt_size;
1816     spapr->fdt_blob = fdt;
1817 
1818     /* Set up the entry state */
1819     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1820     first_ppc_cpu->env.gpr[5] = 0;
1821 
1822     spapr->cas_reboot = false;
1823 }
1824 
1825 static void spapr_create_nvram(SpaprMachineState *spapr)
1826 {
1827     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1828     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1829 
1830     if (dinfo) {
1831         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1832                             &error_fatal);
1833     }
1834 
1835     qdev_init_nofail(dev);
1836 
1837     spapr->nvram = (struct SpaprNvram *)dev;
1838 }
1839 
1840 static void spapr_rtc_create(SpaprMachineState *spapr)
1841 {
1842     object_initialize_child(OBJECT(spapr), "rtc",
1843                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1844                             &error_fatal, NULL);
1845     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1846                               &error_fatal);
1847     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1848                               "date", &error_fatal);
1849 }
1850 
1851 /* Returns whether we want to use VGA or not */
1852 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1853 {
1854     switch (vga_interface_type) {
1855     case VGA_NONE:
1856         return false;
1857     case VGA_DEVICE:
1858         return true;
1859     case VGA_STD:
1860     case VGA_VIRTIO:
1861     case VGA_CIRRUS:
1862         return pci_vga_init(pci_bus) != NULL;
1863     default:
1864         error_setg(errp,
1865                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1866         return false;
1867     }
1868 }
1869 
1870 static int spapr_pre_load(void *opaque)
1871 {
1872     int rc;
1873 
1874     rc = spapr_caps_pre_load(opaque);
1875     if (rc) {
1876         return rc;
1877     }
1878 
1879     return 0;
1880 }
1881 
1882 static int spapr_post_load(void *opaque, int version_id)
1883 {
1884     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1885     int err = 0;
1886 
1887     err = spapr_caps_post_migration(spapr);
1888     if (err) {
1889         return err;
1890     }
1891 
1892     /*
1893      * In earlier versions, there was no separate qdev for the PAPR
1894      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1895      * So when migrating from those versions, poke the incoming offset
1896      * value into the RTC device
1897      */
1898     if (version_id < 3) {
1899         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1900         if (err) {
1901             return err;
1902         }
1903     }
1904 
1905     if (kvm_enabled() && spapr->patb_entry) {
1906         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1907         bool radix = !!(spapr->patb_entry & PATE1_GR);
1908         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1909 
1910         /*
1911          * Update LPCR:HR and UPRT as they may not be set properly in
1912          * the stream
1913          */
1914         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1915                             LPCR_HR | LPCR_UPRT);
1916 
1917         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1918         if (err) {
1919             error_report("Process table config unsupported by the host");
1920             return -EINVAL;
1921         }
1922     }
1923 
1924     err = spapr_irq_post_load(spapr, version_id);
1925     if (err) {
1926         return err;
1927     }
1928 
1929     return err;
1930 }
1931 
1932 static int spapr_pre_save(void *opaque)
1933 {
1934     int rc;
1935 
1936     rc = spapr_caps_pre_save(opaque);
1937     if (rc) {
1938         return rc;
1939     }
1940 
1941     return 0;
1942 }
1943 
1944 static bool version_before_3(void *opaque, int version_id)
1945 {
1946     return version_id < 3;
1947 }
1948 
1949 static bool spapr_pending_events_needed(void *opaque)
1950 {
1951     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1952     return !QTAILQ_EMPTY(&spapr->pending_events);
1953 }
1954 
1955 static const VMStateDescription vmstate_spapr_event_entry = {
1956     .name = "spapr_event_log_entry",
1957     .version_id = 1,
1958     .minimum_version_id = 1,
1959     .fields = (VMStateField[]) {
1960         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1961         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1962         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1963                                      NULL, extended_length),
1964         VMSTATE_END_OF_LIST()
1965     },
1966 };
1967 
1968 static const VMStateDescription vmstate_spapr_pending_events = {
1969     .name = "spapr_pending_events",
1970     .version_id = 1,
1971     .minimum_version_id = 1,
1972     .needed = spapr_pending_events_needed,
1973     .fields = (VMStateField[]) {
1974         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1975                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1976         VMSTATE_END_OF_LIST()
1977     },
1978 };
1979 
1980 static bool spapr_ov5_cas_needed(void *opaque)
1981 {
1982     SpaprMachineState *spapr = opaque;
1983     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1984     SpaprOptionVector *ov5_legacy = spapr_ovec_new();
1985     SpaprOptionVector *ov5_removed = spapr_ovec_new();
1986     bool cas_needed;
1987 
1988     /* Prior to the introduction of SpaprOptionVector, we had two option
1989      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1990      * Both of these options encode machine topology into the device-tree
1991      * in such a way that the now-booted OS should still be able to interact
1992      * appropriately with QEMU regardless of what options were actually
1993      * negotiatied on the source side.
1994      *
1995      * As such, we can avoid migrating the CAS-negotiated options if these
1996      * are the only options available on the current machine/platform.
1997      * Since these are the only options available for pseries-2.7 and
1998      * earlier, this allows us to maintain old->new/new->old migration
1999      * compatibility.
2000      *
2001      * For QEMU 2.8+, there are additional CAS-negotiatable options available
2002      * via default pseries-2.8 machines and explicit command-line parameters.
2003      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
2004      * of the actual CAS-negotiated values to continue working properly. For
2005      * example, availability of memory unplug depends on knowing whether
2006      * OV5_HP_EVT was negotiated via CAS.
2007      *
2008      * Thus, for any cases where the set of available CAS-negotiatable
2009      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
2010      * include the CAS-negotiated options in the migration stream, unless
2011      * if they affect boot time behaviour only.
2012      */
2013     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
2014     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
2015     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
2016 
2017     /* spapr_ovec_diff returns true if bits were removed. we avoid using
2018      * the mask itself since in the future it's possible "legacy" bits may be
2019      * removed via machine options, which could generate a false positive
2020      * that breaks migration.
2021      */
2022     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
2023     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
2024 
2025     spapr_ovec_cleanup(ov5_mask);
2026     spapr_ovec_cleanup(ov5_legacy);
2027     spapr_ovec_cleanup(ov5_removed);
2028 
2029     return cas_needed;
2030 }
2031 
2032 static const VMStateDescription vmstate_spapr_ov5_cas = {
2033     .name = "spapr_option_vector_ov5_cas",
2034     .version_id = 1,
2035     .minimum_version_id = 1,
2036     .needed = spapr_ov5_cas_needed,
2037     .fields = (VMStateField[]) {
2038         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2039                                  vmstate_spapr_ovec, SpaprOptionVector),
2040         VMSTATE_END_OF_LIST()
2041     },
2042 };
2043 
2044 static bool spapr_patb_entry_needed(void *opaque)
2045 {
2046     SpaprMachineState *spapr = opaque;
2047 
2048     return !!spapr->patb_entry;
2049 }
2050 
2051 static const VMStateDescription vmstate_spapr_patb_entry = {
2052     .name = "spapr_patb_entry",
2053     .version_id = 1,
2054     .minimum_version_id = 1,
2055     .needed = spapr_patb_entry_needed,
2056     .fields = (VMStateField[]) {
2057         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2058         VMSTATE_END_OF_LIST()
2059     },
2060 };
2061 
2062 static bool spapr_irq_map_needed(void *opaque)
2063 {
2064     SpaprMachineState *spapr = opaque;
2065 
2066     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2067 }
2068 
2069 static const VMStateDescription vmstate_spapr_irq_map = {
2070     .name = "spapr_irq_map",
2071     .version_id = 1,
2072     .minimum_version_id = 1,
2073     .needed = spapr_irq_map_needed,
2074     .fields = (VMStateField[]) {
2075         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2076         VMSTATE_END_OF_LIST()
2077     },
2078 };
2079 
2080 static bool spapr_dtb_needed(void *opaque)
2081 {
2082     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2083 
2084     return smc->update_dt_enabled;
2085 }
2086 
2087 static int spapr_dtb_pre_load(void *opaque)
2088 {
2089     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2090 
2091     g_free(spapr->fdt_blob);
2092     spapr->fdt_blob = NULL;
2093     spapr->fdt_size = 0;
2094 
2095     return 0;
2096 }
2097 
2098 static const VMStateDescription vmstate_spapr_dtb = {
2099     .name = "spapr_dtb",
2100     .version_id = 1,
2101     .minimum_version_id = 1,
2102     .needed = spapr_dtb_needed,
2103     .pre_load = spapr_dtb_pre_load,
2104     .fields = (VMStateField[]) {
2105         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2106         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2107         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2108                                      fdt_size),
2109         VMSTATE_END_OF_LIST()
2110     },
2111 };
2112 
2113 static const VMStateDescription vmstate_spapr = {
2114     .name = "spapr",
2115     .version_id = 3,
2116     .minimum_version_id = 1,
2117     .pre_load = spapr_pre_load,
2118     .post_load = spapr_post_load,
2119     .pre_save = spapr_pre_save,
2120     .fields = (VMStateField[]) {
2121         /* used to be @next_irq */
2122         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2123 
2124         /* RTC offset */
2125         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2126 
2127         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2128         VMSTATE_END_OF_LIST()
2129     },
2130     .subsections = (const VMStateDescription*[]) {
2131         &vmstate_spapr_ov5_cas,
2132         &vmstate_spapr_patb_entry,
2133         &vmstate_spapr_pending_events,
2134         &vmstate_spapr_cap_htm,
2135         &vmstate_spapr_cap_vsx,
2136         &vmstate_spapr_cap_dfp,
2137         &vmstate_spapr_cap_cfpc,
2138         &vmstate_spapr_cap_sbbc,
2139         &vmstate_spapr_cap_ibs,
2140         &vmstate_spapr_cap_hpt_maxpagesize,
2141         &vmstate_spapr_irq_map,
2142         &vmstate_spapr_cap_nested_kvm_hv,
2143         &vmstate_spapr_dtb,
2144         &vmstate_spapr_cap_large_decr,
2145         &vmstate_spapr_cap_ccf_assist,
2146         NULL
2147     }
2148 };
2149 
2150 static int htab_save_setup(QEMUFile *f, void *opaque)
2151 {
2152     SpaprMachineState *spapr = opaque;
2153 
2154     /* "Iteration" header */
2155     if (!spapr->htab_shift) {
2156         qemu_put_be32(f, -1);
2157     } else {
2158         qemu_put_be32(f, spapr->htab_shift);
2159     }
2160 
2161     if (spapr->htab) {
2162         spapr->htab_save_index = 0;
2163         spapr->htab_first_pass = true;
2164     } else {
2165         if (spapr->htab_shift) {
2166             assert(kvm_enabled());
2167         }
2168     }
2169 
2170 
2171     return 0;
2172 }
2173 
2174 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2175                             int chunkstart, int n_valid, int n_invalid)
2176 {
2177     qemu_put_be32(f, chunkstart);
2178     qemu_put_be16(f, n_valid);
2179     qemu_put_be16(f, n_invalid);
2180     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2181                     HASH_PTE_SIZE_64 * n_valid);
2182 }
2183 
2184 static void htab_save_end_marker(QEMUFile *f)
2185 {
2186     qemu_put_be32(f, 0);
2187     qemu_put_be16(f, 0);
2188     qemu_put_be16(f, 0);
2189 }
2190 
2191 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2192                                  int64_t max_ns)
2193 {
2194     bool has_timeout = max_ns != -1;
2195     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2196     int index = spapr->htab_save_index;
2197     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2198 
2199     assert(spapr->htab_first_pass);
2200 
2201     do {
2202         int chunkstart;
2203 
2204         /* Consume invalid HPTEs */
2205         while ((index < htabslots)
2206                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2207             CLEAN_HPTE(HPTE(spapr->htab, index));
2208             index++;
2209         }
2210 
2211         /* Consume valid HPTEs */
2212         chunkstart = index;
2213         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2214                && HPTE_VALID(HPTE(spapr->htab, index))) {
2215             CLEAN_HPTE(HPTE(spapr->htab, index));
2216             index++;
2217         }
2218 
2219         if (index > chunkstart) {
2220             int n_valid = index - chunkstart;
2221 
2222             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2223 
2224             if (has_timeout &&
2225                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2226                 break;
2227             }
2228         }
2229     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2230 
2231     if (index >= htabslots) {
2232         assert(index == htabslots);
2233         index = 0;
2234         spapr->htab_first_pass = false;
2235     }
2236     spapr->htab_save_index = index;
2237 }
2238 
2239 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2240                                 int64_t max_ns)
2241 {
2242     bool final = max_ns < 0;
2243     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2244     int examined = 0, sent = 0;
2245     int index = spapr->htab_save_index;
2246     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2247 
2248     assert(!spapr->htab_first_pass);
2249 
2250     do {
2251         int chunkstart, invalidstart;
2252 
2253         /* Consume non-dirty HPTEs */
2254         while ((index < htabslots)
2255                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2256             index++;
2257             examined++;
2258         }
2259 
2260         chunkstart = index;
2261         /* Consume valid dirty HPTEs */
2262         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2263                && HPTE_DIRTY(HPTE(spapr->htab, index))
2264                && HPTE_VALID(HPTE(spapr->htab, index))) {
2265             CLEAN_HPTE(HPTE(spapr->htab, index));
2266             index++;
2267             examined++;
2268         }
2269 
2270         invalidstart = index;
2271         /* Consume invalid dirty HPTEs */
2272         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2273                && HPTE_DIRTY(HPTE(spapr->htab, index))
2274                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2275             CLEAN_HPTE(HPTE(spapr->htab, index));
2276             index++;
2277             examined++;
2278         }
2279 
2280         if (index > chunkstart) {
2281             int n_valid = invalidstart - chunkstart;
2282             int n_invalid = index - invalidstart;
2283 
2284             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2285             sent += index - chunkstart;
2286 
2287             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2288                 break;
2289             }
2290         }
2291 
2292         if (examined >= htabslots) {
2293             break;
2294         }
2295 
2296         if (index >= htabslots) {
2297             assert(index == htabslots);
2298             index = 0;
2299         }
2300     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2301 
2302     if (index >= htabslots) {
2303         assert(index == htabslots);
2304         index = 0;
2305     }
2306 
2307     spapr->htab_save_index = index;
2308 
2309     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2310 }
2311 
2312 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2313 #define MAX_KVM_BUF_SIZE    2048
2314 
2315 static int htab_save_iterate(QEMUFile *f, void *opaque)
2316 {
2317     SpaprMachineState *spapr = opaque;
2318     int fd;
2319     int rc = 0;
2320 
2321     /* Iteration header */
2322     if (!spapr->htab_shift) {
2323         qemu_put_be32(f, -1);
2324         return 1;
2325     } else {
2326         qemu_put_be32(f, 0);
2327     }
2328 
2329     if (!spapr->htab) {
2330         assert(kvm_enabled());
2331 
2332         fd = get_htab_fd(spapr);
2333         if (fd < 0) {
2334             return fd;
2335         }
2336 
2337         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2338         if (rc < 0) {
2339             return rc;
2340         }
2341     } else  if (spapr->htab_first_pass) {
2342         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2343     } else {
2344         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2345     }
2346 
2347     htab_save_end_marker(f);
2348 
2349     return rc;
2350 }
2351 
2352 static int htab_save_complete(QEMUFile *f, void *opaque)
2353 {
2354     SpaprMachineState *spapr = opaque;
2355     int fd;
2356 
2357     /* Iteration header */
2358     if (!spapr->htab_shift) {
2359         qemu_put_be32(f, -1);
2360         return 0;
2361     } else {
2362         qemu_put_be32(f, 0);
2363     }
2364 
2365     if (!spapr->htab) {
2366         int rc;
2367 
2368         assert(kvm_enabled());
2369 
2370         fd = get_htab_fd(spapr);
2371         if (fd < 0) {
2372             return fd;
2373         }
2374 
2375         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2376         if (rc < 0) {
2377             return rc;
2378         }
2379     } else {
2380         if (spapr->htab_first_pass) {
2381             htab_save_first_pass(f, spapr, -1);
2382         }
2383         htab_save_later_pass(f, spapr, -1);
2384     }
2385 
2386     /* End marker */
2387     htab_save_end_marker(f);
2388 
2389     return 0;
2390 }
2391 
2392 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2393 {
2394     SpaprMachineState *spapr = opaque;
2395     uint32_t section_hdr;
2396     int fd = -1;
2397     Error *local_err = NULL;
2398 
2399     if (version_id < 1 || version_id > 1) {
2400         error_report("htab_load() bad version");
2401         return -EINVAL;
2402     }
2403 
2404     section_hdr = qemu_get_be32(f);
2405 
2406     if (section_hdr == -1) {
2407         spapr_free_hpt(spapr);
2408         return 0;
2409     }
2410 
2411     if (section_hdr) {
2412         /* First section gives the htab size */
2413         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2414         if (local_err) {
2415             error_report_err(local_err);
2416             return -EINVAL;
2417         }
2418         return 0;
2419     }
2420 
2421     if (!spapr->htab) {
2422         assert(kvm_enabled());
2423 
2424         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2425         if (fd < 0) {
2426             error_report_err(local_err);
2427             return fd;
2428         }
2429     }
2430 
2431     while (true) {
2432         uint32_t index;
2433         uint16_t n_valid, n_invalid;
2434 
2435         index = qemu_get_be32(f);
2436         n_valid = qemu_get_be16(f);
2437         n_invalid = qemu_get_be16(f);
2438 
2439         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2440             /* End of Stream */
2441             break;
2442         }
2443 
2444         if ((index + n_valid + n_invalid) >
2445             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2446             /* Bad index in stream */
2447             error_report(
2448                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2449                 index, n_valid, n_invalid, spapr->htab_shift);
2450             return -EINVAL;
2451         }
2452 
2453         if (spapr->htab) {
2454             if (n_valid) {
2455                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2456                                 HASH_PTE_SIZE_64 * n_valid);
2457             }
2458             if (n_invalid) {
2459                 memset(HPTE(spapr->htab, index + n_valid), 0,
2460                        HASH_PTE_SIZE_64 * n_invalid);
2461             }
2462         } else {
2463             int rc;
2464 
2465             assert(fd >= 0);
2466 
2467             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2468             if (rc < 0) {
2469                 return rc;
2470             }
2471         }
2472     }
2473 
2474     if (!spapr->htab) {
2475         assert(fd >= 0);
2476         close(fd);
2477     }
2478 
2479     return 0;
2480 }
2481 
2482 static void htab_save_cleanup(void *opaque)
2483 {
2484     SpaprMachineState *spapr = opaque;
2485 
2486     close_htab_fd(spapr);
2487 }
2488 
2489 static SaveVMHandlers savevm_htab_handlers = {
2490     .save_setup = htab_save_setup,
2491     .save_live_iterate = htab_save_iterate,
2492     .save_live_complete_precopy = htab_save_complete,
2493     .save_cleanup = htab_save_cleanup,
2494     .load_state = htab_load,
2495 };
2496 
2497 static void spapr_boot_set(void *opaque, const char *boot_device,
2498                            Error **errp)
2499 {
2500     MachineState *machine = MACHINE(opaque);
2501     machine->boot_order = g_strdup(boot_device);
2502 }
2503 
2504 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2505 {
2506     MachineState *machine = MACHINE(spapr);
2507     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2508     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2509     int i;
2510 
2511     for (i = 0; i < nr_lmbs; i++) {
2512         uint64_t addr;
2513 
2514         addr = i * lmb_size + machine->device_memory->base;
2515         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2516                                addr / lmb_size);
2517     }
2518 }
2519 
2520 /*
2521  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2522  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2523  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2524  */
2525 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2526 {
2527     int i;
2528 
2529     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2530         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2531                    " is not aligned to %" PRIu64 " MiB",
2532                    machine->ram_size,
2533                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2534         return;
2535     }
2536 
2537     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2538         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2539                    " is not aligned to %" PRIu64 " MiB",
2540                    machine->ram_size,
2541                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2542         return;
2543     }
2544 
2545     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2546         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2547             error_setg(errp,
2548                        "Node %d memory size 0x%" PRIx64
2549                        " is not aligned to %" PRIu64 " MiB",
2550                        i, machine->numa_state->nodes[i].node_mem,
2551                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2552             return;
2553         }
2554     }
2555 }
2556 
2557 /* find cpu slot in machine->possible_cpus by core_id */
2558 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2559 {
2560     int index = id / ms->smp.threads;
2561 
2562     if (index >= ms->possible_cpus->len) {
2563         return NULL;
2564     }
2565     if (idx) {
2566         *idx = index;
2567     }
2568     return &ms->possible_cpus->cpus[index];
2569 }
2570 
2571 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2572 {
2573     MachineState *ms = MACHINE(spapr);
2574     Error *local_err = NULL;
2575     bool vsmt_user = !!spapr->vsmt;
2576     int kvm_smt = kvmppc_smt_threads();
2577     int ret;
2578     unsigned int smp_threads = ms->smp.threads;
2579 
2580     if (!kvm_enabled() && (smp_threads > 1)) {
2581         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2582                      "on a pseries machine");
2583         goto out;
2584     }
2585     if (!is_power_of_2(smp_threads)) {
2586         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2587                      "machine because it must be a power of 2", smp_threads);
2588         goto out;
2589     }
2590 
2591     /* Detemine the VSMT mode to use: */
2592     if (vsmt_user) {
2593         if (spapr->vsmt < smp_threads) {
2594             error_setg(&local_err, "Cannot support VSMT mode %d"
2595                          " because it must be >= threads/core (%d)",
2596                          spapr->vsmt, smp_threads);
2597             goto out;
2598         }
2599         /* In this case, spapr->vsmt has been set by the command line */
2600     } else {
2601         /*
2602          * Default VSMT value is tricky, because we need it to be as
2603          * consistent as possible (for migration), but this requires
2604          * changing it for at least some existing cases.  We pick 8 as
2605          * the value that we'd get with KVM on POWER8, the
2606          * overwhelmingly common case in production systems.
2607          */
2608         spapr->vsmt = MAX(8, smp_threads);
2609     }
2610 
2611     /* KVM: If necessary, set the SMT mode: */
2612     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2613         ret = kvmppc_set_smt_threads(spapr->vsmt);
2614         if (ret) {
2615             /* Looks like KVM isn't able to change VSMT mode */
2616             error_setg(&local_err,
2617                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2618                        spapr->vsmt, ret);
2619             /* We can live with that if the default one is big enough
2620              * for the number of threads, and a submultiple of the one
2621              * we want.  In this case we'll waste some vcpu ids, but
2622              * behaviour will be correct */
2623             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2624                 warn_report_err(local_err);
2625                 local_err = NULL;
2626                 goto out;
2627             } else {
2628                 if (!vsmt_user) {
2629                     error_append_hint(&local_err,
2630                                       "On PPC, a VM with %d threads/core"
2631                                       " on a host with %d threads/core"
2632                                       " requires the use of VSMT mode %d.\n",
2633                                       smp_threads, kvm_smt, spapr->vsmt);
2634                 }
2635                 kvmppc_hint_smt_possible(&local_err);
2636                 goto out;
2637             }
2638         }
2639     }
2640     /* else TCG: nothing to do currently */
2641 out:
2642     error_propagate(errp, local_err);
2643 }
2644 
2645 static void spapr_init_cpus(SpaprMachineState *spapr)
2646 {
2647     MachineState *machine = MACHINE(spapr);
2648     MachineClass *mc = MACHINE_GET_CLASS(machine);
2649     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2650     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2651     const CPUArchIdList *possible_cpus;
2652     unsigned int smp_cpus = machine->smp.cpus;
2653     unsigned int smp_threads = machine->smp.threads;
2654     unsigned int max_cpus = machine->smp.max_cpus;
2655     int boot_cores_nr = smp_cpus / smp_threads;
2656     int i;
2657 
2658     possible_cpus = mc->possible_cpu_arch_ids(machine);
2659     if (mc->has_hotpluggable_cpus) {
2660         if (smp_cpus % smp_threads) {
2661             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2662                          smp_cpus, smp_threads);
2663             exit(1);
2664         }
2665         if (max_cpus % smp_threads) {
2666             error_report("max_cpus (%u) must be multiple of threads (%u)",
2667                          max_cpus, smp_threads);
2668             exit(1);
2669         }
2670     } else {
2671         if (max_cpus != smp_cpus) {
2672             error_report("This machine version does not support CPU hotplug");
2673             exit(1);
2674         }
2675         boot_cores_nr = possible_cpus->len;
2676     }
2677 
2678     if (smc->pre_2_10_has_unused_icps) {
2679         int i;
2680 
2681         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2682             /* Dummy entries get deregistered when real ICPState objects
2683              * are registered during CPU core hotplug.
2684              */
2685             pre_2_10_vmstate_register_dummy_icp(i);
2686         }
2687     }
2688 
2689     for (i = 0; i < possible_cpus->len; i++) {
2690         int core_id = i * smp_threads;
2691 
2692         if (mc->has_hotpluggable_cpus) {
2693             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2694                                    spapr_vcpu_id(spapr, core_id));
2695         }
2696 
2697         if (i < boot_cores_nr) {
2698             Object *core  = object_new(type);
2699             int nr_threads = smp_threads;
2700 
2701             /* Handle the partially filled core for older machine types */
2702             if ((i + 1) * smp_threads >= smp_cpus) {
2703                 nr_threads = smp_cpus - i * smp_threads;
2704             }
2705 
2706             object_property_set_int(core, nr_threads, "nr-threads",
2707                                     &error_fatal);
2708             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2709                                     &error_fatal);
2710             object_property_set_bool(core, true, "realized", &error_fatal);
2711 
2712             object_unref(core);
2713         }
2714     }
2715 }
2716 
2717 static PCIHostState *spapr_create_default_phb(void)
2718 {
2719     DeviceState *dev;
2720 
2721     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2722     qdev_prop_set_uint32(dev, "index", 0);
2723     qdev_init_nofail(dev);
2724 
2725     return PCI_HOST_BRIDGE(dev);
2726 }
2727 
2728 /* pSeries LPAR / sPAPR hardware init */
2729 static void spapr_machine_init(MachineState *machine)
2730 {
2731     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2732     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2733     const char *kernel_filename = machine->kernel_filename;
2734     const char *initrd_filename = machine->initrd_filename;
2735     PCIHostState *phb;
2736     int i;
2737     MemoryRegion *sysmem = get_system_memory();
2738     MemoryRegion *ram = g_new(MemoryRegion, 1);
2739     hwaddr node0_size = spapr_node0_size(machine);
2740     long load_limit, fw_size;
2741     char *filename;
2742     Error *resize_hpt_err = NULL;
2743 
2744     msi_nonbroken = true;
2745 
2746     QLIST_INIT(&spapr->phbs);
2747     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2748 
2749     /* Determine capabilities to run with */
2750     spapr_caps_init(spapr);
2751 
2752     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2753     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2754         /*
2755          * If the user explicitly requested a mode we should either
2756          * supply it, or fail completely (which we do below).  But if
2757          * it's not set explicitly, we reset our mode to something
2758          * that works
2759          */
2760         if (resize_hpt_err) {
2761             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2762             error_free(resize_hpt_err);
2763             resize_hpt_err = NULL;
2764         } else {
2765             spapr->resize_hpt = smc->resize_hpt_default;
2766         }
2767     }
2768 
2769     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2770 
2771     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2772         /*
2773          * User requested HPT resize, but this host can't supply it.  Bail out
2774          */
2775         error_report_err(resize_hpt_err);
2776         exit(1);
2777     }
2778 
2779     spapr->rma_size = node0_size;
2780 
2781     /* With KVM, we don't actually know whether KVM supports an
2782      * unbounded RMA (PR KVM) or is limited by the hash table size
2783      * (HV KVM using VRMA), so we always assume the latter
2784      *
2785      * In that case, we also limit the initial allocations for RTAS
2786      * etc... to 256M since we have no way to know what the VRMA size
2787      * is going to be as it depends on the size of the hash table
2788      * which isn't determined yet.
2789      */
2790     if (kvm_enabled()) {
2791         spapr->vrma_adjust = 1;
2792         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2793     }
2794 
2795     /* Actually we don't support unbounded RMA anymore since we added
2796      * proper emulation of HV mode. The max we can get is 16G which
2797      * also happens to be what we configure for PAPR mode so make sure
2798      * we don't do anything bigger than that
2799      */
2800     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2801 
2802     if (spapr->rma_size > node0_size) {
2803         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2804                      spapr->rma_size);
2805         exit(1);
2806     }
2807 
2808     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2809     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2810 
2811     /*
2812      * VSMT must be set in order to be able to compute VCPU ids, ie to
2813      * call spapr_max_server_number() or spapr_vcpu_id().
2814      */
2815     spapr_set_vsmt_mode(spapr, &error_fatal);
2816 
2817     /* Set up Interrupt Controller before we create the VCPUs */
2818     spapr_irq_init(spapr, &error_fatal);
2819 
2820     /* Set up containers for ibm,client-architecture-support negotiated options
2821      */
2822     spapr->ov5 = spapr_ovec_new();
2823     spapr->ov5_cas = spapr_ovec_new();
2824 
2825     if (smc->dr_lmb_enabled) {
2826         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2827         spapr_validate_node_memory(machine, &error_fatal);
2828     }
2829 
2830     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2831 
2832     /* advertise support for dedicated HP event source to guests */
2833     if (spapr->use_hotplug_event_source) {
2834         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2835     }
2836 
2837     /* advertise support for HPT resizing */
2838     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2839         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2840     }
2841 
2842     /* advertise support for ibm,dyamic-memory-v2 */
2843     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2844 
2845     /* advertise XIVE on POWER9 machines */
2846     if (spapr->irq->ov5 & (SPAPR_OV5_XIVE_EXPLOIT | SPAPR_OV5_XIVE_BOTH)) {
2847         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2848     }
2849 
2850     /* init CPUs */
2851     spapr_init_cpus(spapr);
2852 
2853     /*
2854      * check we don't have a memory-less/cpu-less NUMA node
2855      * Firmware relies on the existing memory/cpu topology to provide the
2856      * NUMA topology to the kernel.
2857      * And the linux kernel needs to know the NUMA topology at start
2858      * to be able to hotplug CPUs later.
2859      */
2860     if (machine->numa_state->num_nodes) {
2861         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2862             /* check for memory-less node */
2863             if (machine->numa_state->nodes[i].node_mem == 0) {
2864                 CPUState *cs;
2865                 int found = 0;
2866                 /* check for cpu-less node */
2867                 CPU_FOREACH(cs) {
2868                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2869                     if (cpu->node_id == i) {
2870                         found = 1;
2871                         break;
2872                     }
2873                 }
2874                 /* memory-less and cpu-less node */
2875                 if (!found) {
2876                     error_report(
2877                        "Memory-less/cpu-less nodes are not supported (node %d)",
2878                                  i);
2879                     exit(1);
2880                 }
2881             }
2882         }
2883 
2884     }
2885 
2886     /*
2887      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2888      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2889      * called from vPHB reset handler so we initialize the counter here.
2890      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2891      * must be equally distant from any other node.
2892      * The final value of spapr->gpu_numa_id is going to be written to
2893      * max-associativity-domains in spapr_build_fdt().
2894      */
2895     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2896 
2897     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2898         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2899                               spapr->max_compat_pvr)) {
2900         /* KVM and TCG always allow GTSE with radix... */
2901         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2902     }
2903     /* ... but not with hash (currently). */
2904 
2905     if (kvm_enabled()) {
2906         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2907         kvmppc_enable_logical_ci_hcalls();
2908         kvmppc_enable_set_mode_hcall();
2909 
2910         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2911         kvmppc_enable_clear_ref_mod_hcalls();
2912 
2913         /* Enable H_PAGE_INIT */
2914         kvmppc_enable_h_page_init();
2915     }
2916 
2917     /* allocate RAM */
2918     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2919                                          machine->ram_size);
2920     memory_region_add_subregion(sysmem, 0, ram);
2921 
2922     /* always allocate the device memory information */
2923     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2924 
2925     /* initialize hotplug memory address space */
2926     if (machine->ram_size < machine->maxram_size) {
2927         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2928         /*
2929          * Limit the number of hotpluggable memory slots to half the number
2930          * slots that KVM supports, leaving the other half for PCI and other
2931          * devices. However ensure that number of slots doesn't drop below 32.
2932          */
2933         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2934                            SPAPR_MAX_RAM_SLOTS;
2935 
2936         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2937             max_memslots = SPAPR_MAX_RAM_SLOTS;
2938         }
2939         if (machine->ram_slots > max_memslots) {
2940             error_report("Specified number of memory slots %"
2941                          PRIu64" exceeds max supported %d",
2942                          machine->ram_slots, max_memslots);
2943             exit(1);
2944         }
2945 
2946         machine->device_memory->base = ROUND_UP(machine->ram_size,
2947                                                 SPAPR_DEVICE_MEM_ALIGN);
2948         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2949                            "device-memory", device_mem_size);
2950         memory_region_add_subregion(sysmem, machine->device_memory->base,
2951                                     &machine->device_memory->mr);
2952     }
2953 
2954     if (smc->dr_lmb_enabled) {
2955         spapr_create_lmb_dr_connectors(spapr);
2956     }
2957 
2958     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2959     if (!filename) {
2960         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2961         exit(1);
2962     }
2963     spapr->rtas_size = get_image_size(filename);
2964     if (spapr->rtas_size < 0) {
2965         error_report("Could not get size of LPAR rtas '%s'", filename);
2966         exit(1);
2967     }
2968     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2969     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2970         error_report("Could not load LPAR rtas '%s'", filename);
2971         exit(1);
2972     }
2973     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2974         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2975                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2976         exit(1);
2977     }
2978     g_free(filename);
2979 
2980     /* Set up RTAS event infrastructure */
2981     spapr_events_init(spapr);
2982 
2983     /* Set up the RTC RTAS interfaces */
2984     spapr_rtc_create(spapr);
2985 
2986     /* Set up VIO bus */
2987     spapr->vio_bus = spapr_vio_bus_init();
2988 
2989     for (i = 0; i < serial_max_hds(); i++) {
2990         if (serial_hd(i)) {
2991             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2992         }
2993     }
2994 
2995     /* We always have at least the nvram device on VIO */
2996     spapr_create_nvram(spapr);
2997 
2998     /*
2999      * Setup hotplug / dynamic-reconfiguration connectors. top-level
3000      * connectors (described in root DT node's "ibm,drc-types" property)
3001      * are pre-initialized here. additional child connectors (such as
3002      * connectors for a PHBs PCI slots) are added as needed during their
3003      * parent's realization.
3004      */
3005     if (smc->dr_phb_enabled) {
3006         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3007             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3008         }
3009     }
3010 
3011     /* Set up PCI */
3012     spapr_pci_rtas_init();
3013 
3014     phb = spapr_create_default_phb();
3015 
3016     for (i = 0; i < nb_nics; i++) {
3017         NICInfo *nd = &nd_table[i];
3018 
3019         if (!nd->model) {
3020             nd->model = g_strdup("spapr-vlan");
3021         }
3022 
3023         if (g_str_equal(nd->model, "spapr-vlan") ||
3024             g_str_equal(nd->model, "ibmveth")) {
3025             spapr_vlan_create(spapr->vio_bus, nd);
3026         } else {
3027             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
3028         }
3029     }
3030 
3031     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3032         spapr_vscsi_create(spapr->vio_bus);
3033     }
3034 
3035     /* Graphics */
3036     if (spapr_vga_init(phb->bus, &error_fatal)) {
3037         spapr->has_graphics = true;
3038         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3039     }
3040 
3041     if (machine->usb) {
3042         if (smc->use_ohci_by_default) {
3043             pci_create_simple(phb->bus, -1, "pci-ohci");
3044         } else {
3045             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3046         }
3047 
3048         if (spapr->has_graphics) {
3049             USBBus *usb_bus = usb_bus_find(-1);
3050 
3051             usb_create_simple(usb_bus, "usb-kbd");
3052             usb_create_simple(usb_bus, "usb-mouse");
3053         }
3054     }
3055 
3056     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
3057         error_report(
3058             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
3059             MIN_RMA_SLOF);
3060         exit(1);
3061     }
3062 
3063     if (kernel_filename) {
3064         uint64_t lowaddr = 0;
3065 
3066         spapr->kernel_size = load_elf(kernel_filename, NULL,
3067                                       translate_kernel_address, NULL,
3068                                       NULL, &lowaddr, NULL, 1,
3069                                       PPC_ELF_MACHINE, 0, 0);
3070         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3071             spapr->kernel_size = load_elf(kernel_filename, NULL,
3072                                           translate_kernel_address, NULL, NULL,
3073                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
3074                                           0, 0);
3075             spapr->kernel_le = spapr->kernel_size > 0;
3076         }
3077         if (spapr->kernel_size < 0) {
3078             error_report("error loading %s: %s", kernel_filename,
3079                          load_elf_strerror(spapr->kernel_size));
3080             exit(1);
3081         }
3082 
3083         /* load initrd */
3084         if (initrd_filename) {
3085             /* Try to locate the initrd in the gap between the kernel
3086              * and the firmware. Add a bit of space just in case
3087              */
3088             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
3089                                   + 0x1ffff) & ~0xffff;
3090             spapr->initrd_size = load_image_targphys(initrd_filename,
3091                                                      spapr->initrd_base,
3092                                                      load_limit
3093                                                      - spapr->initrd_base);
3094             if (spapr->initrd_size < 0) {
3095                 error_report("could not load initial ram disk '%s'",
3096                              initrd_filename);
3097                 exit(1);
3098             }
3099         }
3100     }
3101 
3102     if (bios_name == NULL) {
3103         bios_name = FW_FILE_NAME;
3104     }
3105     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3106     if (!filename) {
3107         error_report("Could not find LPAR firmware '%s'", bios_name);
3108         exit(1);
3109     }
3110     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3111     if (fw_size <= 0) {
3112         error_report("Could not load LPAR firmware '%s'", filename);
3113         exit(1);
3114     }
3115     g_free(filename);
3116 
3117     /* FIXME: Should register things through the MachineState's qdev
3118      * interface, this is a legacy from the sPAPREnvironment structure
3119      * which predated MachineState but had a similar function */
3120     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3121     register_savevm_live("spapr/htab", -1, 1,
3122                          &savevm_htab_handlers, spapr);
3123 
3124     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3125                              &error_fatal);
3126 
3127     qemu_register_boot_set(spapr_boot_set, spapr);
3128 
3129     /*
3130      * Nothing needs to be done to resume a suspended guest because
3131      * suspending does not change the machine state, so no need for
3132      * a ->wakeup method.
3133      */
3134     qemu_register_wakeup_support();
3135 
3136     if (kvm_enabled()) {
3137         /* to stop and start vmclock */
3138         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3139                                          &spapr->tb);
3140 
3141         kvmppc_spapr_enable_inkernel_multitce();
3142     }
3143 }
3144 
3145 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3146 {
3147     if (!vm_type) {
3148         return 0;
3149     }
3150 
3151     if (!strcmp(vm_type, "HV")) {
3152         return 1;
3153     }
3154 
3155     if (!strcmp(vm_type, "PR")) {
3156         return 2;
3157     }
3158 
3159     error_report("Unknown kvm-type specified '%s'", vm_type);
3160     exit(1);
3161 }
3162 
3163 /*
3164  * Implementation of an interface to adjust firmware path
3165  * for the bootindex property handling.
3166  */
3167 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3168                                    DeviceState *dev)
3169 {
3170 #define CAST(type, obj, name) \
3171     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3172     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3173     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3174     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3175 
3176     if (d) {
3177         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3178         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3179         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3180 
3181         if (spapr) {
3182             /*
3183              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3184              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3185              * 0x8000 | (target << 8) | (bus << 5) | lun
3186              * (see the "Logical unit addressing format" table in SAM5)
3187              */
3188             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3189             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3190                                    (uint64_t)id << 48);
3191         } else if (virtio) {
3192             /*
3193              * We use SRP luns of the form 01000000 | (target << 8) | lun
3194              * in the top 32 bits of the 64-bit LUN
3195              * Note: the quote above is from SLOF and it is wrong,
3196              * the actual binding is:
3197              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3198              */
3199             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3200             if (d->lun >= 256) {
3201                 /* Use the LUN "flat space addressing method" */
3202                 id |= 0x4000;
3203             }
3204             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3205                                    (uint64_t)id << 32);
3206         } else if (usb) {
3207             /*
3208              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3209              * in the top 32 bits of the 64-bit LUN
3210              */
3211             unsigned usb_port = atoi(usb->port->path);
3212             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3213             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3214                                    (uint64_t)id << 32);
3215         }
3216     }
3217 
3218     /*
3219      * SLOF probes the USB devices, and if it recognizes that the device is a
3220      * storage device, it changes its name to "storage" instead of "usb-host",
3221      * and additionally adds a child node for the SCSI LUN, so the correct
3222      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3223      */
3224     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3225         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3226         if (usb_host_dev_is_scsi_storage(usbdev)) {
3227             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3228         }
3229     }
3230 
3231     if (phb) {
3232         /* Replace "pci" with "pci@800000020000000" */
3233         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3234     }
3235 
3236     if (vsc) {
3237         /* Same logic as virtio above */
3238         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3239         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3240     }
3241 
3242     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3243         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3244         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3245         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3246     }
3247 
3248     return NULL;
3249 }
3250 
3251 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3252 {
3253     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3254 
3255     return g_strdup(spapr->kvm_type);
3256 }
3257 
3258 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3259 {
3260     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3261 
3262     g_free(spapr->kvm_type);
3263     spapr->kvm_type = g_strdup(value);
3264 }
3265 
3266 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3267 {
3268     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3269 
3270     return spapr->use_hotplug_event_source;
3271 }
3272 
3273 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3274                                             Error **errp)
3275 {
3276     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3277 
3278     spapr->use_hotplug_event_source = value;
3279 }
3280 
3281 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3282 {
3283     return true;
3284 }
3285 
3286 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3287 {
3288     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3289 
3290     switch (spapr->resize_hpt) {
3291     case SPAPR_RESIZE_HPT_DEFAULT:
3292         return g_strdup("default");
3293     case SPAPR_RESIZE_HPT_DISABLED:
3294         return g_strdup("disabled");
3295     case SPAPR_RESIZE_HPT_ENABLED:
3296         return g_strdup("enabled");
3297     case SPAPR_RESIZE_HPT_REQUIRED:
3298         return g_strdup("required");
3299     }
3300     g_assert_not_reached();
3301 }
3302 
3303 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3304 {
3305     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3306 
3307     if (strcmp(value, "default") == 0) {
3308         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3309     } else if (strcmp(value, "disabled") == 0) {
3310         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3311     } else if (strcmp(value, "enabled") == 0) {
3312         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3313     } else if (strcmp(value, "required") == 0) {
3314         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3315     } else {
3316         error_setg(errp, "Bad value for \"resize-hpt\" property");
3317     }
3318 }
3319 
3320 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3321                                    void *opaque, Error **errp)
3322 {
3323     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3324 }
3325 
3326 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3327                                    void *opaque, Error **errp)
3328 {
3329     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3330 }
3331 
3332 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3333 {
3334     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3335 
3336     if (spapr->irq == &spapr_irq_xics_legacy) {
3337         return g_strdup("legacy");
3338     } else if (spapr->irq == &spapr_irq_xics) {
3339         return g_strdup("xics");
3340     } else if (spapr->irq == &spapr_irq_xive) {
3341         return g_strdup("xive");
3342     } else if (spapr->irq == &spapr_irq_dual) {
3343         return g_strdup("dual");
3344     }
3345     g_assert_not_reached();
3346 }
3347 
3348 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3349 {
3350     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3351 
3352     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3353         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3354         return;
3355     }
3356 
3357     /* The legacy IRQ backend can not be set */
3358     if (strcmp(value, "xics") == 0) {
3359         spapr->irq = &spapr_irq_xics;
3360     } else if (strcmp(value, "xive") == 0) {
3361         spapr->irq = &spapr_irq_xive;
3362     } else if (strcmp(value, "dual") == 0) {
3363         spapr->irq = &spapr_irq_dual;
3364     } else {
3365         error_setg(errp, "Bad value for \"ic-mode\" property");
3366     }
3367 }
3368 
3369 static char *spapr_get_host_model(Object *obj, Error **errp)
3370 {
3371     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3372 
3373     return g_strdup(spapr->host_model);
3374 }
3375 
3376 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3377 {
3378     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3379 
3380     g_free(spapr->host_model);
3381     spapr->host_model = g_strdup(value);
3382 }
3383 
3384 static char *spapr_get_host_serial(Object *obj, Error **errp)
3385 {
3386     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3387 
3388     return g_strdup(spapr->host_serial);
3389 }
3390 
3391 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3392 {
3393     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3394 
3395     g_free(spapr->host_serial);
3396     spapr->host_serial = g_strdup(value);
3397 }
3398 
3399 static void spapr_instance_init(Object *obj)
3400 {
3401     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3402     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3403 
3404     spapr->htab_fd = -1;
3405     spapr->use_hotplug_event_source = true;
3406     object_property_add_str(obj, "kvm-type",
3407                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3408     object_property_set_description(obj, "kvm-type",
3409                                     "Specifies the KVM virtualization mode (HV, PR)",
3410                                     NULL);
3411     object_property_add_bool(obj, "modern-hotplug-events",
3412                             spapr_get_modern_hotplug_events,
3413                             spapr_set_modern_hotplug_events,
3414                             NULL);
3415     object_property_set_description(obj, "modern-hotplug-events",
3416                                     "Use dedicated hotplug event mechanism in"
3417                                     " place of standard EPOW events when possible"
3418                                     " (required for memory hot-unplug support)",
3419                                     NULL);
3420     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3421                             "Maximum permitted CPU compatibility mode",
3422                             &error_fatal);
3423 
3424     object_property_add_str(obj, "resize-hpt",
3425                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3426     object_property_set_description(obj, "resize-hpt",
3427                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3428                                     NULL);
3429     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3430                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3431     object_property_set_description(obj, "vsmt",
3432                                     "Virtual SMT: KVM behaves as if this were"
3433                                     " the host's SMT mode", &error_abort);
3434     object_property_add_bool(obj, "vfio-no-msix-emulation",
3435                              spapr_get_msix_emulation, NULL, NULL);
3436 
3437     /* The machine class defines the default interrupt controller mode */
3438     spapr->irq = smc->irq;
3439     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3440                             spapr_set_ic_mode, NULL);
3441     object_property_set_description(obj, "ic-mode",
3442                  "Specifies the interrupt controller mode (xics, xive, dual)",
3443                  NULL);
3444 
3445     object_property_add_str(obj, "host-model",
3446         spapr_get_host_model, spapr_set_host_model,
3447         &error_abort);
3448     object_property_set_description(obj, "host-model",
3449         "Host model to advertise in guest device tree", &error_abort);
3450     object_property_add_str(obj, "host-serial",
3451         spapr_get_host_serial, spapr_set_host_serial,
3452         &error_abort);
3453     object_property_set_description(obj, "host-serial",
3454         "Host serial number to advertise in guest device tree", &error_abort);
3455 }
3456 
3457 static void spapr_machine_finalizefn(Object *obj)
3458 {
3459     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3460 
3461     g_free(spapr->kvm_type);
3462 }
3463 
3464 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3465 {
3466     cpu_synchronize_state(cs);
3467     ppc_cpu_do_system_reset(cs);
3468 }
3469 
3470 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3471 {
3472     CPUState *cs;
3473 
3474     CPU_FOREACH(cs) {
3475         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3476     }
3477 }
3478 
3479 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3480                           void *fdt, int *fdt_start_offset, Error **errp)
3481 {
3482     uint64_t addr;
3483     uint32_t node;
3484 
3485     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3486     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3487                                     &error_abort);
3488     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3489                                                    SPAPR_MEMORY_BLOCK_SIZE);
3490     return 0;
3491 }
3492 
3493 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3494                            bool dedicated_hp_event_source, Error **errp)
3495 {
3496     SpaprDrc *drc;
3497     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3498     int i;
3499     uint64_t addr = addr_start;
3500     bool hotplugged = spapr_drc_hotplugged(dev);
3501     Error *local_err = NULL;
3502 
3503     for (i = 0; i < nr_lmbs; i++) {
3504         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3505                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3506         g_assert(drc);
3507 
3508         spapr_drc_attach(drc, dev, &local_err);
3509         if (local_err) {
3510             while (addr > addr_start) {
3511                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3512                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3513                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3514                 spapr_drc_detach(drc);
3515             }
3516             error_propagate(errp, local_err);
3517             return;
3518         }
3519         if (!hotplugged) {
3520             spapr_drc_reset(drc);
3521         }
3522         addr += SPAPR_MEMORY_BLOCK_SIZE;
3523     }
3524     /* send hotplug notification to the
3525      * guest only in case of hotplugged memory
3526      */
3527     if (hotplugged) {
3528         if (dedicated_hp_event_source) {
3529             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3530                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3531             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3532                                                    nr_lmbs,
3533                                                    spapr_drc_index(drc));
3534         } else {
3535             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3536                                            nr_lmbs);
3537         }
3538     }
3539 }
3540 
3541 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3542                               Error **errp)
3543 {
3544     Error *local_err = NULL;
3545     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3546     PCDIMMDevice *dimm = PC_DIMM(dev);
3547     uint64_t size, addr;
3548 
3549     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3550 
3551     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3552     if (local_err) {
3553         goto out;
3554     }
3555 
3556     addr = object_property_get_uint(OBJECT(dimm),
3557                                     PC_DIMM_ADDR_PROP, &local_err);
3558     if (local_err) {
3559         goto out_unplug;
3560     }
3561 
3562     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3563                    &local_err);
3564     if (local_err) {
3565         goto out_unplug;
3566     }
3567 
3568     return;
3569 
3570 out_unplug:
3571     pc_dimm_unplug(dimm, MACHINE(ms));
3572 out:
3573     error_propagate(errp, local_err);
3574 }
3575 
3576 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3577                                   Error **errp)
3578 {
3579     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3580     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3581     PCDIMMDevice *dimm = PC_DIMM(dev);
3582     Error *local_err = NULL;
3583     uint64_t size;
3584     Object *memdev;
3585     hwaddr pagesize;
3586 
3587     if (!smc->dr_lmb_enabled) {
3588         error_setg(errp, "Memory hotplug not supported for this machine");
3589         return;
3590     }
3591 
3592     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3593     if (local_err) {
3594         error_propagate(errp, local_err);
3595         return;
3596     }
3597 
3598     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3599         error_setg(errp, "Hotplugged memory size must be a multiple of "
3600                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3601         return;
3602     }
3603 
3604     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3605                                       &error_abort);
3606     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3607     spapr_check_pagesize(spapr, pagesize, &local_err);
3608     if (local_err) {
3609         error_propagate(errp, local_err);
3610         return;
3611     }
3612 
3613     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3614 }
3615 
3616 struct SpaprDimmState {
3617     PCDIMMDevice *dimm;
3618     uint32_t nr_lmbs;
3619     QTAILQ_ENTRY(SpaprDimmState) next;
3620 };
3621 
3622 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3623                                                        PCDIMMDevice *dimm)
3624 {
3625     SpaprDimmState *dimm_state = NULL;
3626 
3627     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3628         if (dimm_state->dimm == dimm) {
3629             break;
3630         }
3631     }
3632     return dimm_state;
3633 }
3634 
3635 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3636                                                       uint32_t nr_lmbs,
3637                                                       PCDIMMDevice *dimm)
3638 {
3639     SpaprDimmState *ds = NULL;
3640 
3641     /*
3642      * If this request is for a DIMM whose removal had failed earlier
3643      * (due to guest's refusal to remove the LMBs), we would have this
3644      * dimm already in the pending_dimm_unplugs list. In that
3645      * case don't add again.
3646      */
3647     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3648     if (!ds) {
3649         ds = g_malloc0(sizeof(SpaprDimmState));
3650         ds->nr_lmbs = nr_lmbs;
3651         ds->dimm = dimm;
3652         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3653     }
3654     return ds;
3655 }
3656 
3657 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3658                                               SpaprDimmState *dimm_state)
3659 {
3660     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3661     g_free(dimm_state);
3662 }
3663 
3664 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3665                                                         PCDIMMDevice *dimm)
3666 {
3667     SpaprDrc *drc;
3668     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3669                                                   &error_abort);
3670     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3671     uint32_t avail_lmbs = 0;
3672     uint64_t addr_start, addr;
3673     int i;
3674 
3675     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3676                                          &error_abort);
3677 
3678     addr = addr_start;
3679     for (i = 0; i < nr_lmbs; i++) {
3680         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3681                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3682         g_assert(drc);
3683         if (drc->dev) {
3684             avail_lmbs++;
3685         }
3686         addr += SPAPR_MEMORY_BLOCK_SIZE;
3687     }
3688 
3689     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3690 }
3691 
3692 /* Callback to be called during DRC release. */
3693 void spapr_lmb_release(DeviceState *dev)
3694 {
3695     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3696     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3697     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3698 
3699     /* This information will get lost if a migration occurs
3700      * during the unplug process. In this case recover it. */
3701     if (ds == NULL) {
3702         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3703         g_assert(ds);
3704         /* The DRC being examined by the caller at least must be counted */
3705         g_assert(ds->nr_lmbs);
3706     }
3707 
3708     if (--ds->nr_lmbs) {
3709         return;
3710     }
3711 
3712     /*
3713      * Now that all the LMBs have been removed by the guest, call the
3714      * unplug handler chain. This can never fail.
3715      */
3716     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3717     object_unparent(OBJECT(dev));
3718 }
3719 
3720 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3721 {
3722     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3723     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3724 
3725     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3726     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3727     spapr_pending_dimm_unplugs_remove(spapr, ds);
3728 }
3729 
3730 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3731                                         DeviceState *dev, Error **errp)
3732 {
3733     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3734     Error *local_err = NULL;
3735     PCDIMMDevice *dimm = PC_DIMM(dev);
3736     uint32_t nr_lmbs;
3737     uint64_t size, addr_start, addr;
3738     int i;
3739     SpaprDrc *drc;
3740 
3741     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3742     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3743 
3744     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3745                                          &local_err);
3746     if (local_err) {
3747         goto out;
3748     }
3749 
3750     /*
3751      * An existing pending dimm state for this DIMM means that there is an
3752      * unplug operation in progress, waiting for the spapr_lmb_release
3753      * callback to complete the job (BQL can't cover that far). In this case,
3754      * bail out to avoid detaching DRCs that were already released.
3755      */
3756     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3757         error_setg(&local_err,
3758                    "Memory unplug already in progress for device %s",
3759                    dev->id);
3760         goto out;
3761     }
3762 
3763     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3764 
3765     addr = addr_start;
3766     for (i = 0; i < nr_lmbs; i++) {
3767         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3768                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3769         g_assert(drc);
3770 
3771         spapr_drc_detach(drc);
3772         addr += SPAPR_MEMORY_BLOCK_SIZE;
3773     }
3774 
3775     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3776                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3777     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3778                                               nr_lmbs, spapr_drc_index(drc));
3779 out:
3780     error_propagate(errp, local_err);
3781 }
3782 
3783 /* Callback to be called during DRC release. */
3784 void spapr_core_release(DeviceState *dev)
3785 {
3786     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3787 
3788     /* Call the unplug handler chain. This can never fail. */
3789     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3790     object_unparent(OBJECT(dev));
3791 }
3792 
3793 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3794 {
3795     MachineState *ms = MACHINE(hotplug_dev);
3796     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3797     CPUCore *cc = CPU_CORE(dev);
3798     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3799 
3800     if (smc->pre_2_10_has_unused_icps) {
3801         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3802         int i;
3803 
3804         for (i = 0; i < cc->nr_threads; i++) {
3805             CPUState *cs = CPU(sc->threads[i]);
3806 
3807             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3808         }
3809     }
3810 
3811     assert(core_slot);
3812     core_slot->cpu = NULL;
3813     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3814 }
3815 
3816 static
3817 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3818                                Error **errp)
3819 {
3820     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3821     int index;
3822     SpaprDrc *drc;
3823     CPUCore *cc = CPU_CORE(dev);
3824 
3825     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3826         error_setg(errp, "Unable to find CPU core with core-id: %d",
3827                    cc->core_id);
3828         return;
3829     }
3830     if (index == 0) {
3831         error_setg(errp, "Boot CPU core may not be unplugged");
3832         return;
3833     }
3834 
3835     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3836                           spapr_vcpu_id(spapr, cc->core_id));
3837     g_assert(drc);
3838 
3839     spapr_drc_detach(drc);
3840 
3841     spapr_hotplug_req_remove_by_index(drc);
3842 }
3843 
3844 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3845                            void *fdt, int *fdt_start_offset, Error **errp)
3846 {
3847     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3848     CPUState *cs = CPU(core->threads[0]);
3849     PowerPCCPU *cpu = POWERPC_CPU(cs);
3850     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3851     int id = spapr_get_vcpu_id(cpu);
3852     char *nodename;
3853     int offset;
3854 
3855     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3856     offset = fdt_add_subnode(fdt, 0, nodename);
3857     g_free(nodename);
3858 
3859     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3860 
3861     *fdt_start_offset = offset;
3862     return 0;
3863 }
3864 
3865 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3866                             Error **errp)
3867 {
3868     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3869     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3870     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3871     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3872     CPUCore *cc = CPU_CORE(dev);
3873     CPUState *cs;
3874     SpaprDrc *drc;
3875     Error *local_err = NULL;
3876     CPUArchId *core_slot;
3877     int index;
3878     bool hotplugged = spapr_drc_hotplugged(dev);
3879     int i;
3880 
3881     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3882     if (!core_slot) {
3883         error_setg(errp, "Unable to find CPU core with core-id: %d",
3884                    cc->core_id);
3885         return;
3886     }
3887     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3888                           spapr_vcpu_id(spapr, cc->core_id));
3889 
3890     g_assert(drc || !mc->has_hotpluggable_cpus);
3891 
3892     if (drc) {
3893         spapr_drc_attach(drc, dev, &local_err);
3894         if (local_err) {
3895             error_propagate(errp, local_err);
3896             return;
3897         }
3898 
3899         if (hotplugged) {
3900             /*
3901              * Send hotplug notification interrupt to the guest only
3902              * in case of hotplugged CPUs.
3903              */
3904             spapr_hotplug_req_add_by_index(drc);
3905         } else {
3906             spapr_drc_reset(drc);
3907         }
3908     }
3909 
3910     core_slot->cpu = OBJECT(dev);
3911 
3912     if (smc->pre_2_10_has_unused_icps) {
3913         for (i = 0; i < cc->nr_threads; i++) {
3914             cs = CPU(core->threads[i]);
3915             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3916         }
3917     }
3918 
3919     /*
3920      * Set compatibility mode to match the boot CPU, which was either set
3921      * by the machine reset code or by CAS.
3922      */
3923     if (hotplugged) {
3924         for (i = 0; i < cc->nr_threads; i++) {
3925             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3926                            &local_err);
3927             if (local_err) {
3928                 error_propagate(errp, local_err);
3929                 return;
3930             }
3931         }
3932     }
3933 }
3934 
3935 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3936                                 Error **errp)
3937 {
3938     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3939     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3940     Error *local_err = NULL;
3941     CPUCore *cc = CPU_CORE(dev);
3942     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3943     const char *type = object_get_typename(OBJECT(dev));
3944     CPUArchId *core_slot;
3945     int index;
3946     unsigned int smp_threads = machine->smp.threads;
3947 
3948     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3949         error_setg(&local_err, "CPU hotplug not supported for this machine");
3950         goto out;
3951     }
3952 
3953     if (strcmp(base_core_type, type)) {
3954         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3955         goto out;
3956     }
3957 
3958     if (cc->core_id % smp_threads) {
3959         error_setg(&local_err, "invalid core id %d", cc->core_id);
3960         goto out;
3961     }
3962 
3963     /*
3964      * In general we should have homogeneous threads-per-core, but old
3965      * (pre hotplug support) machine types allow the last core to have
3966      * reduced threads as a compatibility hack for when we allowed
3967      * total vcpus not a multiple of threads-per-core.
3968      */
3969     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3970         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3971                    cc->nr_threads, smp_threads);
3972         goto out;
3973     }
3974 
3975     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3976     if (!core_slot) {
3977         error_setg(&local_err, "core id %d out of range", cc->core_id);
3978         goto out;
3979     }
3980 
3981     if (core_slot->cpu) {
3982         error_setg(&local_err, "core %d already populated", cc->core_id);
3983         goto out;
3984     }
3985 
3986     numa_cpu_pre_plug(core_slot, dev, &local_err);
3987 
3988 out:
3989     error_propagate(errp, local_err);
3990 }
3991 
3992 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3993                           void *fdt, int *fdt_start_offset, Error **errp)
3994 {
3995     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3996     int intc_phandle;
3997 
3998     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3999     if (intc_phandle <= 0) {
4000         return -1;
4001     }
4002 
4003     if (spapr_dt_phb(sphb, intc_phandle, fdt, spapr->irq->nr_msis,
4004                      fdt_start_offset)) {
4005         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4006         return -1;
4007     }
4008 
4009     /* generally SLOF creates these, for hotplug it's up to QEMU */
4010     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4011 
4012     return 0;
4013 }
4014 
4015 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4016                                Error **errp)
4017 {
4018     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4019     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4020     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4021     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4022 
4023     if (dev->hotplugged && !smc->dr_phb_enabled) {
4024         error_setg(errp, "PHB hotplug not supported for this machine");
4025         return;
4026     }
4027 
4028     if (sphb->index == (uint32_t)-1) {
4029         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4030         return;
4031     }
4032 
4033     /*
4034      * This will check that sphb->index doesn't exceed the maximum number of
4035      * PHBs for the current machine type.
4036      */
4037     smc->phb_placement(spapr, sphb->index,
4038                        &sphb->buid, &sphb->io_win_addr,
4039                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
4040                        windows_supported, sphb->dma_liobn,
4041                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4042                        errp);
4043 }
4044 
4045 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4046                            Error **errp)
4047 {
4048     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4049     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4050     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4051     SpaprDrc *drc;
4052     bool hotplugged = spapr_drc_hotplugged(dev);
4053     Error *local_err = NULL;
4054 
4055     if (!smc->dr_phb_enabled) {
4056         return;
4057     }
4058 
4059     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4060     /* hotplug hooks should check it's enabled before getting this far */
4061     assert(drc);
4062 
4063     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4064     if (local_err) {
4065         error_propagate(errp, local_err);
4066         return;
4067     }
4068 
4069     if (hotplugged) {
4070         spapr_hotplug_req_add_by_index(drc);
4071     } else {
4072         spapr_drc_reset(drc);
4073     }
4074 }
4075 
4076 void spapr_phb_release(DeviceState *dev)
4077 {
4078     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4079 
4080     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4081     object_unparent(OBJECT(dev));
4082 }
4083 
4084 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4085 {
4086     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4087 }
4088 
4089 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4090                                      DeviceState *dev, Error **errp)
4091 {
4092     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4093     SpaprDrc *drc;
4094 
4095     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4096     assert(drc);
4097 
4098     if (!spapr_drc_unplug_requested(drc)) {
4099         spapr_drc_detach(drc);
4100         spapr_hotplug_req_remove_by_index(drc);
4101     }
4102 }
4103 
4104 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4105                                  Error **errp)
4106 {
4107     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4108     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4109 
4110     if (spapr->tpm_proxy != NULL) {
4111         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4112         return;
4113     }
4114 
4115     spapr->tpm_proxy = tpm_proxy;
4116 }
4117 
4118 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4119 {
4120     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4121 
4122     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4123     object_unparent(OBJECT(dev));
4124     spapr->tpm_proxy = NULL;
4125 }
4126 
4127 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4128                                       DeviceState *dev, Error **errp)
4129 {
4130     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4131         spapr_memory_plug(hotplug_dev, dev, errp);
4132     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4133         spapr_core_plug(hotplug_dev, dev, errp);
4134     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4135         spapr_phb_plug(hotplug_dev, dev, errp);
4136     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4137         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4138     }
4139 }
4140 
4141 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4142                                         DeviceState *dev, Error **errp)
4143 {
4144     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4145         spapr_memory_unplug(hotplug_dev, dev);
4146     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4147         spapr_core_unplug(hotplug_dev, dev);
4148     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4149         spapr_phb_unplug(hotplug_dev, dev);
4150     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4151         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4152     }
4153 }
4154 
4155 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4156                                                 DeviceState *dev, Error **errp)
4157 {
4158     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4159     MachineClass *mc = MACHINE_GET_CLASS(sms);
4160     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4161 
4162     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4163         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4164             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4165         } else {
4166             /* NOTE: this means there is a window after guest reset, prior to
4167              * CAS negotiation, where unplug requests will fail due to the
4168              * capability not being detected yet. This is a bit different than
4169              * the case with PCI unplug, where the events will be queued and
4170              * eventually handled by the guest after boot
4171              */
4172             error_setg(errp, "Memory hot unplug not supported for this guest");
4173         }
4174     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4175         if (!mc->has_hotpluggable_cpus) {
4176             error_setg(errp, "CPU hot unplug not supported on this machine");
4177             return;
4178         }
4179         spapr_core_unplug_request(hotplug_dev, dev, errp);
4180     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4181         if (!smc->dr_phb_enabled) {
4182             error_setg(errp, "PHB hot unplug not supported on this machine");
4183             return;
4184         }
4185         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4186     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4187         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4188     }
4189 }
4190 
4191 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4192                                           DeviceState *dev, Error **errp)
4193 {
4194     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4195         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4196     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4197         spapr_core_pre_plug(hotplug_dev, dev, errp);
4198     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4199         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4200     }
4201 }
4202 
4203 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4204                                                  DeviceState *dev)
4205 {
4206     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4207         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4208         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4209         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4210         return HOTPLUG_HANDLER(machine);
4211     }
4212     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4213         PCIDevice *pcidev = PCI_DEVICE(dev);
4214         PCIBus *root = pci_device_root_bus(pcidev);
4215         SpaprPhbState *phb =
4216             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4217                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4218 
4219         if (phb) {
4220             return HOTPLUG_HANDLER(phb);
4221         }
4222     }
4223     return NULL;
4224 }
4225 
4226 static CpuInstanceProperties
4227 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4228 {
4229     CPUArchId *core_slot;
4230     MachineClass *mc = MACHINE_GET_CLASS(machine);
4231 
4232     /* make sure possible_cpu are intialized */
4233     mc->possible_cpu_arch_ids(machine);
4234     /* get CPU core slot containing thread that matches cpu_index */
4235     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4236     assert(core_slot);
4237     return core_slot->props;
4238 }
4239 
4240 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4241 {
4242     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4243 }
4244 
4245 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4246 {
4247     int i;
4248     unsigned int smp_threads = machine->smp.threads;
4249     unsigned int smp_cpus = machine->smp.cpus;
4250     const char *core_type;
4251     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4252     MachineClass *mc = MACHINE_GET_CLASS(machine);
4253 
4254     if (!mc->has_hotpluggable_cpus) {
4255         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4256     }
4257     if (machine->possible_cpus) {
4258         assert(machine->possible_cpus->len == spapr_max_cores);
4259         return machine->possible_cpus;
4260     }
4261 
4262     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4263     if (!core_type) {
4264         error_report("Unable to find sPAPR CPU Core definition");
4265         exit(1);
4266     }
4267 
4268     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4269                              sizeof(CPUArchId) * spapr_max_cores);
4270     machine->possible_cpus->len = spapr_max_cores;
4271     for (i = 0; i < machine->possible_cpus->len; i++) {
4272         int core_id = i * smp_threads;
4273 
4274         machine->possible_cpus->cpus[i].type = core_type;
4275         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4276         machine->possible_cpus->cpus[i].arch_id = core_id;
4277         machine->possible_cpus->cpus[i].props.has_core_id = true;
4278         machine->possible_cpus->cpus[i].props.core_id = core_id;
4279     }
4280     return machine->possible_cpus;
4281 }
4282 
4283 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4284                                 uint64_t *buid, hwaddr *pio,
4285                                 hwaddr *mmio32, hwaddr *mmio64,
4286                                 unsigned n_dma, uint32_t *liobns,
4287                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4288 {
4289     /*
4290      * New-style PHB window placement.
4291      *
4292      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4293      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4294      * windows.
4295      *
4296      * Some guest kernels can't work with MMIO windows above 1<<46
4297      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4298      *
4299      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4300      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4301      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4302      * 1TiB 64-bit MMIO windows for each PHB.
4303      */
4304     const uint64_t base_buid = 0x800000020000000ULL;
4305     int i;
4306 
4307     /* Sanity check natural alignments */
4308     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4309     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4310     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4311     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4312     /* Sanity check bounds */
4313     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4314                       SPAPR_PCI_MEM32_WIN_SIZE);
4315     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4316                       SPAPR_PCI_MEM64_WIN_SIZE);
4317 
4318     if (index >= SPAPR_MAX_PHBS) {
4319         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4320                    SPAPR_MAX_PHBS - 1);
4321         return;
4322     }
4323 
4324     *buid = base_buid + index;
4325     for (i = 0; i < n_dma; ++i) {
4326         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4327     }
4328 
4329     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4330     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4331     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4332 
4333     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4334     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4335 }
4336 
4337 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4338 {
4339     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4340 
4341     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4342 }
4343 
4344 static void spapr_ics_resend(XICSFabric *dev)
4345 {
4346     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4347 
4348     ics_resend(spapr->ics);
4349 }
4350 
4351 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4352 {
4353     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4354 
4355     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4356 }
4357 
4358 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4359                                  Monitor *mon)
4360 {
4361     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4362 
4363     spapr->irq->print_info(spapr, mon);
4364     monitor_printf(mon, "irqchip: %s\n",
4365                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4366 }
4367 
4368 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4369 {
4370     return cpu->vcpu_id;
4371 }
4372 
4373 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4374 {
4375     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4376     MachineState *ms = MACHINE(spapr);
4377     int vcpu_id;
4378 
4379     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4380 
4381     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4382         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4383         error_append_hint(errp, "Adjust the number of cpus to %d "
4384                           "or try to raise the number of threads per core\n",
4385                           vcpu_id * ms->smp.threads / spapr->vsmt);
4386         return;
4387     }
4388 
4389     cpu->vcpu_id = vcpu_id;
4390 }
4391 
4392 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4393 {
4394     CPUState *cs;
4395 
4396     CPU_FOREACH(cs) {
4397         PowerPCCPU *cpu = POWERPC_CPU(cs);
4398 
4399         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4400             return cpu;
4401         }
4402     }
4403 
4404     return NULL;
4405 }
4406 
4407 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4408 {
4409     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4410 
4411     /* These are only called by TCG, KVM maintains dispatch state */
4412 
4413     spapr_cpu->prod = false;
4414     if (spapr_cpu->vpa_addr) {
4415         CPUState *cs = CPU(cpu);
4416         uint32_t dispatch;
4417 
4418         dispatch = ldl_be_phys(cs->as,
4419                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4420         dispatch++;
4421         if ((dispatch & 1) != 0) {
4422             qemu_log_mask(LOG_GUEST_ERROR,
4423                           "VPA: incorrect dispatch counter value for "
4424                           "dispatched partition %u, correcting.\n", dispatch);
4425             dispatch++;
4426         }
4427         stl_be_phys(cs->as,
4428                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4429     }
4430 }
4431 
4432 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4433 {
4434     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4435 
4436     if (spapr_cpu->vpa_addr) {
4437         CPUState *cs = CPU(cpu);
4438         uint32_t dispatch;
4439 
4440         dispatch = ldl_be_phys(cs->as,
4441                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4442         dispatch++;
4443         if ((dispatch & 1) != 1) {
4444             qemu_log_mask(LOG_GUEST_ERROR,
4445                           "VPA: incorrect dispatch counter value for "
4446                           "preempted partition %u, correcting.\n", dispatch);
4447             dispatch++;
4448         }
4449         stl_be_phys(cs->as,
4450                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4451     }
4452 }
4453 
4454 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4455 {
4456     MachineClass *mc = MACHINE_CLASS(oc);
4457     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4458     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4459     NMIClass *nc = NMI_CLASS(oc);
4460     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4461     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4462     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4463     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4464 
4465     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4466     mc->ignore_boot_device_suffixes = true;
4467 
4468     /*
4469      * We set up the default / latest behaviour here.  The class_init
4470      * functions for the specific versioned machine types can override
4471      * these details for backwards compatibility
4472      */
4473     mc->init = spapr_machine_init;
4474     mc->reset = spapr_machine_reset;
4475     mc->block_default_type = IF_SCSI;
4476     mc->max_cpus = 1024;
4477     mc->no_parallel = 1;
4478     mc->default_boot_order = "";
4479     mc->default_ram_size = 512 * MiB;
4480     mc->default_display = "std";
4481     mc->kvm_type = spapr_kvm_type;
4482     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4483     mc->pci_allow_0_address = true;
4484     assert(!mc->get_hotplug_handler);
4485     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4486     hc->pre_plug = spapr_machine_device_pre_plug;
4487     hc->plug = spapr_machine_device_plug;
4488     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4489     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4490     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4491     hc->unplug_request = spapr_machine_device_unplug_request;
4492     hc->unplug = spapr_machine_device_unplug;
4493 
4494     smc->dr_lmb_enabled = true;
4495     smc->update_dt_enabled = true;
4496     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4497     mc->has_hotpluggable_cpus = true;
4498     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4499     fwc->get_dev_path = spapr_get_fw_dev_path;
4500     nc->nmi_monitor_handler = spapr_nmi;
4501     smc->phb_placement = spapr_phb_placement;
4502     vhc->hypercall = emulate_spapr_hypercall;
4503     vhc->hpt_mask = spapr_hpt_mask;
4504     vhc->map_hptes = spapr_map_hptes;
4505     vhc->unmap_hptes = spapr_unmap_hptes;
4506     vhc->hpte_set_c = spapr_hpte_set_c;
4507     vhc->hpte_set_r = spapr_hpte_set_r;
4508     vhc->get_pate = spapr_get_pate;
4509     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4510     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4511     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4512     xic->ics_get = spapr_ics_get;
4513     xic->ics_resend = spapr_ics_resend;
4514     xic->icp_get = spapr_icp_get;
4515     ispc->print_info = spapr_pic_print_info;
4516     /* Force NUMA node memory size to be a multiple of
4517      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4518      * in which LMBs are represented and hot-added
4519      */
4520     mc->numa_mem_align_shift = 28;
4521     mc->numa_mem_supported = true;
4522 
4523     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4524     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4525     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4526     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4527     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4528     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4529     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4530     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4531     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4532     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4533     spapr_caps_add_properties(smc, &error_abort);
4534     smc->irq = &spapr_irq_dual;
4535     smc->dr_phb_enabled = true;
4536     smc->linux_pci_probe = true;
4537 }
4538 
4539 static const TypeInfo spapr_machine_info = {
4540     .name          = TYPE_SPAPR_MACHINE,
4541     .parent        = TYPE_MACHINE,
4542     .abstract      = true,
4543     .instance_size = sizeof(SpaprMachineState),
4544     .instance_init = spapr_instance_init,
4545     .instance_finalize = spapr_machine_finalizefn,
4546     .class_size    = sizeof(SpaprMachineClass),
4547     .class_init    = spapr_machine_class_init,
4548     .interfaces = (InterfaceInfo[]) {
4549         { TYPE_FW_PATH_PROVIDER },
4550         { TYPE_NMI },
4551         { TYPE_HOTPLUG_HANDLER },
4552         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4553         { TYPE_XICS_FABRIC },
4554         { TYPE_INTERRUPT_STATS_PROVIDER },
4555         { }
4556     },
4557 };
4558 
4559 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4560     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4561                                                     void *data)      \
4562     {                                                                \
4563         MachineClass *mc = MACHINE_CLASS(oc);                        \
4564         spapr_machine_##suffix##_class_options(mc);                  \
4565         if (latest) {                                                \
4566             mc->alias = "pseries";                                   \
4567             mc->is_default = 1;                                      \
4568         }                                                            \
4569     }                                                                \
4570     static const TypeInfo spapr_machine_##suffix##_info = {          \
4571         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4572         .parent = TYPE_SPAPR_MACHINE,                                \
4573         .class_init = spapr_machine_##suffix##_class_init,           \
4574     };                                                               \
4575     static void spapr_machine_register_##suffix(void)                \
4576     {                                                                \
4577         type_register(&spapr_machine_##suffix##_info);               \
4578     }                                                                \
4579     type_init(spapr_machine_register_##suffix)
4580 
4581 /*
4582  * pseries-4.2
4583  */
4584 static void spapr_machine_4_2_class_options(MachineClass *mc)
4585 {
4586     /* Defaults for the latest behaviour inherited from the base class */
4587 }
4588 
4589 DEFINE_SPAPR_MACHINE(4_2, "4.2", true);
4590 
4591 /*
4592  * pseries-4.1
4593  */
4594 static void spapr_machine_4_1_class_options(MachineClass *mc)
4595 {
4596     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4597     static GlobalProperty compat[] = {
4598         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4599         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4600     };
4601 
4602     spapr_machine_4_2_class_options(mc);
4603     smc->linux_pci_probe = false;
4604     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4605     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4606 }
4607 
4608 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4609 
4610 /*
4611  * pseries-4.0
4612  */
4613 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4614                               uint64_t *buid, hwaddr *pio,
4615                               hwaddr *mmio32, hwaddr *mmio64,
4616                               unsigned n_dma, uint32_t *liobns,
4617                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4618 {
4619     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4620                         nv2gpa, nv2atsd, errp);
4621     *nv2gpa = 0;
4622     *nv2atsd = 0;
4623 }
4624 
4625 static void spapr_machine_4_0_class_options(MachineClass *mc)
4626 {
4627     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4628 
4629     spapr_machine_4_1_class_options(mc);
4630     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4631     smc->phb_placement = phb_placement_4_0;
4632     smc->irq = &spapr_irq_xics;
4633     smc->pre_4_1_migration = true;
4634 }
4635 
4636 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4637 
4638 /*
4639  * pseries-3.1
4640  */
4641 static void spapr_machine_3_1_class_options(MachineClass *mc)
4642 {
4643     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4644 
4645     spapr_machine_4_0_class_options(mc);
4646     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4647 
4648     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4649     smc->update_dt_enabled = false;
4650     smc->dr_phb_enabled = false;
4651     smc->broken_host_serial_model = true;
4652     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4653     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4654     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4655     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4656 }
4657 
4658 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4659 
4660 /*
4661  * pseries-3.0
4662  */
4663 
4664 static void spapr_machine_3_0_class_options(MachineClass *mc)
4665 {
4666     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4667 
4668     spapr_machine_3_1_class_options(mc);
4669     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4670 
4671     smc->legacy_irq_allocation = true;
4672     smc->irq = &spapr_irq_xics_legacy;
4673 }
4674 
4675 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4676 
4677 /*
4678  * pseries-2.12
4679  */
4680 static void spapr_machine_2_12_class_options(MachineClass *mc)
4681 {
4682     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4683     static GlobalProperty compat[] = {
4684         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4685         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4686     };
4687 
4688     spapr_machine_3_0_class_options(mc);
4689     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4690     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4691 
4692     /* We depend on kvm_enabled() to choose a default value for the
4693      * hpt-max-page-size capability. Of course we can't do it here
4694      * because this is too early and the HW accelerator isn't initialzed
4695      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4696      */
4697     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4698 }
4699 
4700 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4701 
4702 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4703 {
4704     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4705 
4706     spapr_machine_2_12_class_options(mc);
4707     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4708     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4709     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4710 }
4711 
4712 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4713 
4714 /*
4715  * pseries-2.11
4716  */
4717 
4718 static void spapr_machine_2_11_class_options(MachineClass *mc)
4719 {
4720     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4721 
4722     spapr_machine_2_12_class_options(mc);
4723     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4724     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4725 }
4726 
4727 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4728 
4729 /*
4730  * pseries-2.10
4731  */
4732 
4733 static void spapr_machine_2_10_class_options(MachineClass *mc)
4734 {
4735     spapr_machine_2_11_class_options(mc);
4736     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4737 }
4738 
4739 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4740 
4741 /*
4742  * pseries-2.9
4743  */
4744 
4745 static void spapr_machine_2_9_class_options(MachineClass *mc)
4746 {
4747     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4748     static GlobalProperty compat[] = {
4749         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4750     };
4751 
4752     spapr_machine_2_10_class_options(mc);
4753     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4754     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4755     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4756     smc->pre_2_10_has_unused_icps = true;
4757     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4758 }
4759 
4760 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4761 
4762 /*
4763  * pseries-2.8
4764  */
4765 
4766 static void spapr_machine_2_8_class_options(MachineClass *mc)
4767 {
4768     static GlobalProperty compat[] = {
4769         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4770     };
4771 
4772     spapr_machine_2_9_class_options(mc);
4773     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4774     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4775     mc->numa_mem_align_shift = 23;
4776 }
4777 
4778 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4779 
4780 /*
4781  * pseries-2.7
4782  */
4783 
4784 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4785                               uint64_t *buid, hwaddr *pio,
4786                               hwaddr *mmio32, hwaddr *mmio64,
4787                               unsigned n_dma, uint32_t *liobns,
4788                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4789 {
4790     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4791     const uint64_t base_buid = 0x800000020000000ULL;
4792     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4793     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4794     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4795     const uint32_t max_index = 255;
4796     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4797 
4798     uint64_t ram_top = MACHINE(spapr)->ram_size;
4799     hwaddr phb0_base, phb_base;
4800     int i;
4801 
4802     /* Do we have device memory? */
4803     if (MACHINE(spapr)->maxram_size > ram_top) {
4804         /* Can't just use maxram_size, because there may be an
4805          * alignment gap between normal and device memory regions
4806          */
4807         ram_top = MACHINE(spapr)->device_memory->base +
4808             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4809     }
4810 
4811     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4812 
4813     if (index > max_index) {
4814         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4815                    max_index);
4816         return;
4817     }
4818 
4819     *buid = base_buid + index;
4820     for (i = 0; i < n_dma; ++i) {
4821         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4822     }
4823 
4824     phb_base = phb0_base + index * phb_spacing;
4825     *pio = phb_base + pio_offset;
4826     *mmio32 = phb_base + mmio_offset;
4827     /*
4828      * We don't set the 64-bit MMIO window, relying on the PHB's
4829      * fallback behaviour of automatically splitting a large "32-bit"
4830      * window into contiguous 32-bit and 64-bit windows
4831      */
4832 
4833     *nv2gpa = 0;
4834     *nv2atsd = 0;
4835 }
4836 
4837 static void spapr_machine_2_7_class_options(MachineClass *mc)
4838 {
4839     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4840     static GlobalProperty compat[] = {
4841         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4842         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4843         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4844         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4845     };
4846 
4847     spapr_machine_2_8_class_options(mc);
4848     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4849     mc->default_machine_opts = "modern-hotplug-events=off";
4850     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4851     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4852     smc->phb_placement = phb_placement_2_7;
4853 }
4854 
4855 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4856 
4857 /*
4858  * pseries-2.6
4859  */
4860 
4861 static void spapr_machine_2_6_class_options(MachineClass *mc)
4862 {
4863     static GlobalProperty compat[] = {
4864         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4865     };
4866 
4867     spapr_machine_2_7_class_options(mc);
4868     mc->has_hotpluggable_cpus = false;
4869     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4870     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4871 }
4872 
4873 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4874 
4875 /*
4876  * pseries-2.5
4877  */
4878 
4879 static void spapr_machine_2_5_class_options(MachineClass *mc)
4880 {
4881     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4882     static GlobalProperty compat[] = {
4883         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4884     };
4885 
4886     spapr_machine_2_6_class_options(mc);
4887     smc->use_ohci_by_default = true;
4888     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4889     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4890 }
4891 
4892 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4893 
4894 /*
4895  * pseries-2.4
4896  */
4897 
4898 static void spapr_machine_2_4_class_options(MachineClass *mc)
4899 {
4900     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4901 
4902     spapr_machine_2_5_class_options(mc);
4903     smc->dr_lmb_enabled = false;
4904     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4905 }
4906 
4907 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4908 
4909 /*
4910  * pseries-2.3
4911  */
4912 
4913 static void spapr_machine_2_3_class_options(MachineClass *mc)
4914 {
4915     static GlobalProperty compat[] = {
4916         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4917     };
4918     spapr_machine_2_4_class_options(mc);
4919     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4920     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4921 }
4922 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4923 
4924 /*
4925  * pseries-2.2
4926  */
4927 
4928 static void spapr_machine_2_2_class_options(MachineClass *mc)
4929 {
4930     static GlobalProperty compat[] = {
4931         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4932     };
4933 
4934     spapr_machine_2_3_class_options(mc);
4935     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4936     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4937     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4938 }
4939 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4940 
4941 /*
4942  * pseries-2.1
4943  */
4944 
4945 static void spapr_machine_2_1_class_options(MachineClass *mc)
4946 {
4947     spapr_machine_2_2_class_options(mc);
4948     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4949 }
4950 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4951 
4952 static void spapr_machine_register_types(void)
4953 {
4954     type_register_static(&spapr_machine_info);
4955 }
4956 
4957 type_init(spapr_machine_register_types)
4958