xref: /openbmc/qemu/hw/ppc/spapr.c (revision 3eb21fe9e5a06e485dbb27838422ef85f4ae7967)
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 "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
47 #include "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/pci/msi.h"
58 
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
63 
64 #include "exec/address-spaces.h"
65 #include "exec/ram_addr.h"
66 #include "hw/usb.h"
67 #include "qemu/config-file.h"
68 #include "qemu/error-report.h"
69 #include "trace.h"
70 #include "hw/nmi.h"
71 #include "hw/intc/intc.h"
72 
73 #include "hw/compat.h"
74 #include "qemu/cutils.h"
75 #include "hw/ppc/spapr_cpu_core.h"
76 #include "hw/mem/memory-device.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 /* These two functions implement the VCPU id numbering: one to compute them
103  * all and one to identify thread 0 of a VCORE. Any change to the first one
104  * is likely to have an impact on the second one, so let's keep them close.
105  */
106 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
107 {
108     assert(spapr->vsmt);
109     return
110         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
111 }
112 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
113                                       PowerPCCPU *cpu)
114 {
115     assert(spapr->vsmt);
116     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
117 }
118 
119 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
120 {
121     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
122      * and newer QEMUs don't even have them. In both cases, we don't want
123      * to send anything on the wire.
124      */
125     return false;
126 }
127 
128 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
129     .name = "icp/server",
130     .version_id = 1,
131     .minimum_version_id = 1,
132     .needed = pre_2_10_vmstate_dummy_icp_needed,
133     .fields = (VMStateField[]) {
134         VMSTATE_UNUSED(4), /* uint32_t xirr */
135         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
136         VMSTATE_UNUSED(1), /* uint8_t mfrr */
137         VMSTATE_END_OF_LIST()
138     },
139 };
140 
141 static void pre_2_10_vmstate_register_dummy_icp(int i)
142 {
143     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
144                      (void *)(uintptr_t) i);
145 }
146 
147 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
148 {
149     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
150                        (void *)(uintptr_t) i);
151 }
152 
153 static int xics_max_server_number(sPAPRMachineState *spapr)
154 {
155     assert(spapr->vsmt);
156     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
157 }
158 
159 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
160                                   int smt_threads)
161 {
162     int i, ret = 0;
163     uint32_t servers_prop[smt_threads];
164     uint32_t gservers_prop[smt_threads * 2];
165     int index = spapr_get_vcpu_id(cpu);
166 
167     if (cpu->compat_pvr) {
168         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
169         if (ret < 0) {
170             return ret;
171         }
172     }
173 
174     /* Build interrupt servers and gservers properties */
175     for (i = 0; i < smt_threads; i++) {
176         servers_prop[i] = cpu_to_be32(index + i);
177         /* Hack, direct the group queues back to cpu 0 */
178         gservers_prop[i*2] = cpu_to_be32(index + i);
179         gservers_prop[i*2 + 1] = 0;
180     }
181     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
182                       servers_prop, sizeof(servers_prop));
183     if (ret < 0) {
184         return ret;
185     }
186     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
187                       gservers_prop, sizeof(gservers_prop));
188 
189     return ret;
190 }
191 
192 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
193 {
194     int index = spapr_get_vcpu_id(cpu);
195     uint32_t associativity[] = {cpu_to_be32(0x5),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(0x0),
198                                 cpu_to_be32(0x0),
199                                 cpu_to_be32(cpu->node_id),
200                                 cpu_to_be32(index)};
201 
202     /* Advertise NUMA via ibm,associativity */
203     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
204                           sizeof(associativity));
205 }
206 
207 /* Populate the "ibm,pa-features" property */
208 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
209                                        PowerPCCPU *cpu,
210                                        void *fdt, int offset,
211                                        bool legacy_guest)
212 {
213     uint8_t pa_features_206[] = { 6, 0,
214         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
215     uint8_t pa_features_207[] = { 24, 0,
216         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
217         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
218         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
219         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
220     uint8_t pa_features_300[] = { 66, 0,
221         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
222         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
223         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
224         /* 6: DS207 */
225         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
226         /* 16: Vector */
227         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
228         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
230         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
232         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
233         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
234         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
236         /* 42: PM, 44: PC RA, 46: SC vec'd */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
238         /* 48: SIMD, 50: QP BFP, 52: String */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
240         /* 54: DecFP, 56: DecI, 58: SHA */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
242         /* 60: NM atomic, 62: RNG */
243         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
244     };
245     uint8_t *pa_features = NULL;
246     size_t pa_size;
247 
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_206;
250         pa_size = sizeof(pa_features_206);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_207;
254         pa_size = sizeof(pa_features_207);
255     }
256     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
257         pa_features = pa_features_300;
258         pa_size = sizeof(pa_features_300);
259     }
260     if (!pa_features) {
261         return;
262     }
263 
264     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
265         /*
266          * Note: we keep CI large pages off by default because a 64K capable
267          * guest provisioned with large pages might otherwise try to map a qemu
268          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
269          * even if that qemu runs on a 4k host.
270          * We dd this bit back here if we are confident this is not an issue
271          */
272         pa_features[3] |= 0x20;
273     }
274     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
275         pa_features[24] |= 0x80;    /* Transactional memory support */
276     }
277     if (legacy_guest && pa_size > 40) {
278         /* Workaround for broken kernels that attempt (guest) radix
279          * mode when they can't handle it, if they see the radix bit set
280          * in pa-features. So hide it from them. */
281         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
282     }
283 
284     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
285 }
286 
287 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
288 {
289     int ret = 0, offset, cpus_offset;
290     CPUState *cs;
291     char cpu_model[32];
292     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
293 
294     CPU_FOREACH(cs) {
295         PowerPCCPU *cpu = POWERPC_CPU(cs);
296         DeviceClass *dc = DEVICE_GET_CLASS(cs);
297         int index = spapr_get_vcpu_id(cpu);
298         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
299 
300         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
301             continue;
302         }
303 
304         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
305 
306         cpus_offset = fdt_path_offset(fdt, "/cpus");
307         if (cpus_offset < 0) {
308             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
309             if (cpus_offset < 0) {
310                 return cpus_offset;
311             }
312         }
313         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
314         if (offset < 0) {
315             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
316             if (offset < 0) {
317                 return offset;
318             }
319         }
320 
321         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
322                           pft_size_prop, sizeof(pft_size_prop));
323         if (ret < 0) {
324             return ret;
325         }
326 
327         if (nb_numa_nodes > 1) {
328             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
329             if (ret < 0) {
330                 return ret;
331             }
332         }
333 
334         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
335         if (ret < 0) {
336             return ret;
337         }
338 
339         spapr_populate_pa_features(spapr, cpu, fdt, offset,
340                                    spapr->cas_legacy_guest_workaround);
341     }
342     return ret;
343 }
344 
345 static hwaddr spapr_node0_size(MachineState *machine)
346 {
347     if (nb_numa_nodes) {
348         int i;
349         for (i = 0; i < nb_numa_nodes; ++i) {
350             if (numa_info[i].node_mem) {
351                 return MIN(pow2floor(numa_info[i].node_mem),
352                            machine->ram_size);
353             }
354         }
355     }
356     return machine->ram_size;
357 }
358 
359 static void add_str(GString *s, const gchar *s1)
360 {
361     g_string_append_len(s, s1, strlen(s1) + 1);
362 }
363 
364 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
365                                        hwaddr size)
366 {
367     uint32_t associativity[] = {
368         cpu_to_be32(0x4), /* length */
369         cpu_to_be32(0x0), cpu_to_be32(0x0),
370         cpu_to_be32(0x0), cpu_to_be32(nodeid)
371     };
372     char mem_name[32];
373     uint64_t mem_reg_property[2];
374     int off;
375 
376     mem_reg_property[0] = cpu_to_be64(start);
377     mem_reg_property[1] = cpu_to_be64(size);
378 
379     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
380     off = fdt_add_subnode(fdt, 0, mem_name);
381     _FDT(off);
382     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
383     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
384                       sizeof(mem_reg_property))));
385     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
386                       sizeof(associativity))));
387     return off;
388 }
389 
390 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
391 {
392     MachineState *machine = MACHINE(spapr);
393     hwaddr mem_start, node_size;
394     int i, nb_nodes = nb_numa_nodes;
395     NodeInfo *nodes = numa_info;
396     NodeInfo ramnode;
397 
398     /* No NUMA nodes, assume there is just one node with whole RAM */
399     if (!nb_numa_nodes) {
400         nb_nodes = 1;
401         ramnode.node_mem = machine->ram_size;
402         nodes = &ramnode;
403     }
404 
405     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
406         if (!nodes[i].node_mem) {
407             continue;
408         }
409         if (mem_start >= machine->ram_size) {
410             node_size = 0;
411         } else {
412             node_size = nodes[i].node_mem;
413             if (node_size > machine->ram_size - mem_start) {
414                 node_size = machine->ram_size - mem_start;
415             }
416         }
417         if (!mem_start) {
418             /* spapr_machine_init() checks for rma_size <= node0_size
419              * already */
420             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
421             mem_start += spapr->rma_size;
422             node_size -= spapr->rma_size;
423         }
424         for ( ; node_size; ) {
425             hwaddr sizetmp = pow2floor(node_size);
426 
427             /* mem_start != 0 here */
428             if (ctzl(mem_start) < ctzl(sizetmp)) {
429                 sizetmp = 1ULL << ctzl(mem_start);
430             }
431 
432             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
433             node_size -= sizetmp;
434             mem_start += sizetmp;
435         }
436     }
437 
438     return 0;
439 }
440 
441 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
442                                   sPAPRMachineState *spapr)
443 {
444     PowerPCCPU *cpu = POWERPC_CPU(cs);
445     CPUPPCState *env = &cpu->env;
446     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
447     int index = spapr_get_vcpu_id(cpu);
448     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
449                        0xffffffff, 0xffffffff};
450     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
451         : SPAPR_TIMEBASE_FREQ;
452     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
453     uint32_t page_sizes_prop[64];
454     size_t page_sizes_prop_size;
455     uint32_t vcpus_per_socket = smp_threads * smp_cores;
456     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
457     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
458     sPAPRDRConnector *drc;
459     int drc_index;
460     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
461     int i;
462 
463     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
464     if (drc) {
465         drc_index = spapr_drc_index(drc);
466         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
467     }
468 
469     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
470     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
471 
472     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
476                            env->dcache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
478                            env->icache_line_size)));
479     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
480                            env->icache_line_size)));
481 
482     if (pcc->l1_dcache_size) {
483         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
484                                pcc->l1_dcache_size)));
485     } else {
486         warn_report("Unknown L1 dcache size for cpu");
487     }
488     if (pcc->l1_icache_size) {
489         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
490                                pcc->l1_icache_size)));
491     } else {
492         warn_report("Unknown L1 icache size for cpu");
493     }
494 
495     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
497     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
498     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
499     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
500     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
501 
502     if (env->spr_cb[SPR_PURR].oea_read) {
503         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
504     }
505 
506     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
508                           segs, sizeof(segs))));
509     }
510 
511     /* Advertise VSX (vector extensions) if available
512      *   1               == VMX / Altivec available
513      *   2               == VSX available
514      *
515      * Only CPUs for which we create core types in spapr_cpu_core.c
516      * are possible, and all of those have VMX */
517     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
518         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
519     } else {
520         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
521     }
522 
523     /* Advertise DFP (Decimal Floating Point) if available
524      *   0 / no property == no DFP
525      *   1               == DFP available */
526     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
527         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
528     }
529 
530     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
531                                                       sizeof(page_sizes_prop));
532     if (page_sizes_prop_size) {
533         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
534                           page_sizes_prop, page_sizes_prop_size)));
535     }
536 
537     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
538 
539     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
540                            cs->cpu_index / vcpus_per_socket)));
541 
542     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
543                       pft_size_prop, sizeof(pft_size_prop))));
544 
545     if (nb_numa_nodes > 1) {
546         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
547     }
548 
549     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
550 
551     if (pcc->radix_page_info) {
552         for (i = 0; i < pcc->radix_page_info->count; i++) {
553             radix_AP_encodings[i] =
554                 cpu_to_be32(pcc->radix_page_info->entries[i]);
555         }
556         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
557                           radix_AP_encodings,
558                           pcc->radix_page_info->count *
559                           sizeof(radix_AP_encodings[0]))));
560     }
561 }
562 
563 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
564 {
565     CPUState *cs;
566     int cpus_offset;
567     char *nodename;
568 
569     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
570     _FDT(cpus_offset);
571     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
572     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
573 
574     /*
575      * We walk the CPUs in reverse order to ensure that CPU DT nodes
576      * created by fdt_add_subnode() end up in the right order in FDT
577      * for the guest kernel the enumerate the CPUs correctly.
578      */
579     CPU_FOREACH_REVERSE(cs) {
580         PowerPCCPU *cpu = POWERPC_CPU(cs);
581         int index = spapr_get_vcpu_id(cpu);
582         DeviceClass *dc = DEVICE_GET_CLASS(cs);
583         int offset;
584 
585         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
586             continue;
587         }
588 
589         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
590         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
591         g_free(nodename);
592         _FDT(offset);
593         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
594     }
595 
596 }
597 
598 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
599 {
600     MemoryDeviceInfoList *info;
601 
602     for (info = list; info; info = info->next) {
603         MemoryDeviceInfo *value = info->value;
604 
605         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
606             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
607 
608             if (addr >= pcdimm_info->addr &&
609                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
610                 return pcdimm_info->node;
611             }
612         }
613     }
614 
615     return -1;
616 }
617 
618 struct sPAPRDrconfCellV2 {
619      uint32_t seq_lmbs;
620      uint64_t base_addr;
621      uint32_t drc_index;
622      uint32_t aa_index;
623      uint32_t flags;
624 } QEMU_PACKED;
625 
626 typedef struct DrconfCellQueue {
627     struct sPAPRDrconfCellV2 cell;
628     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
629 } DrconfCellQueue;
630 
631 static DrconfCellQueue *
632 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
633                       uint32_t drc_index, uint32_t aa_index,
634                       uint32_t flags)
635 {
636     DrconfCellQueue *elem;
637 
638     elem = g_malloc0(sizeof(*elem));
639     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
640     elem->cell.base_addr = cpu_to_be64(base_addr);
641     elem->cell.drc_index = cpu_to_be32(drc_index);
642     elem->cell.aa_index = cpu_to_be32(aa_index);
643     elem->cell.flags = cpu_to_be32(flags);
644 
645     return elem;
646 }
647 
648 /* ibm,dynamic-memory-v2 */
649 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
650                                    int offset, MemoryDeviceInfoList *dimms)
651 {
652     MachineState *machine = MACHINE(spapr);
653     uint8_t *int_buf, *cur_index, buf_len;
654     int ret;
655     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
656     uint64_t addr, cur_addr, size;
657     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
658     uint64_t mem_end = machine->device_memory->base +
659                        memory_region_size(&machine->device_memory->mr);
660     uint32_t node, nr_entries = 0;
661     sPAPRDRConnector *drc;
662     DrconfCellQueue *elem, *next;
663     MemoryDeviceInfoList *info;
664     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
665         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
666 
667     /* Entry to cover RAM and the gap area */
668     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
669                                  SPAPR_LMB_FLAGS_RESERVED |
670                                  SPAPR_LMB_FLAGS_DRC_INVALID);
671     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
672     nr_entries++;
673 
674     cur_addr = machine->device_memory->base;
675     for (info = dimms; info; info = info->next) {
676         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
677 
678         addr = di->addr;
679         size = di->size;
680         node = di->node;
681 
682         /* Entry for hot-pluggable area */
683         if (cur_addr < addr) {
684             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
685             g_assert(drc);
686             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
687                                          cur_addr, spapr_drc_index(drc), -1, 0);
688             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
689             nr_entries++;
690         }
691 
692         /* Entry for DIMM */
693         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
694         g_assert(drc);
695         elem = spapr_get_drconf_cell(size / lmb_size, addr,
696                                      spapr_drc_index(drc), node,
697                                      SPAPR_LMB_FLAGS_ASSIGNED);
698         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
699         nr_entries++;
700         cur_addr = addr + size;
701     }
702 
703     /* Entry for remaining hotpluggable area */
704     if (cur_addr < mem_end) {
705         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
706         g_assert(drc);
707         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
708                                      cur_addr, spapr_drc_index(drc), -1, 0);
709         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
710         nr_entries++;
711     }
712 
713     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
714     int_buf = cur_index = g_malloc0(buf_len);
715     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
716     cur_index += sizeof(nr_entries);
717 
718     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
719         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
720         cur_index += sizeof(elem->cell);
721         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
722         g_free(elem);
723     }
724 
725     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
726     g_free(int_buf);
727     if (ret < 0) {
728         return -1;
729     }
730     return 0;
731 }
732 
733 /* ibm,dynamic-memory */
734 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
735                                    int offset, MemoryDeviceInfoList *dimms)
736 {
737     MachineState *machine = MACHINE(spapr);
738     int i, ret;
739     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
740     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
741     uint32_t nr_lmbs = (machine->device_memory->base +
742                        memory_region_size(&machine->device_memory->mr)) /
743                        lmb_size;
744     uint32_t *int_buf, *cur_index, buf_len;
745 
746     /*
747      * Allocate enough buffer size to fit in ibm,dynamic-memory
748      */
749     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
750     cur_index = int_buf = g_malloc0(buf_len);
751     int_buf[0] = cpu_to_be32(nr_lmbs);
752     cur_index++;
753     for (i = 0; i < nr_lmbs; i++) {
754         uint64_t addr = i * lmb_size;
755         uint32_t *dynamic_memory = cur_index;
756 
757         if (i >= device_lmb_start) {
758             sPAPRDRConnector *drc;
759 
760             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
761             g_assert(drc);
762 
763             dynamic_memory[0] = cpu_to_be32(addr >> 32);
764             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
765             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
766             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
767             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
768             if (memory_region_present(get_system_memory(), addr)) {
769                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
770             } else {
771                 dynamic_memory[5] = cpu_to_be32(0);
772             }
773         } else {
774             /*
775              * LMB information for RMA, boot time RAM and gap b/n RAM and
776              * device memory region -- all these are marked as reserved
777              * and as having no valid DRC.
778              */
779             dynamic_memory[0] = cpu_to_be32(addr >> 32);
780             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
781             dynamic_memory[2] = cpu_to_be32(0);
782             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
783             dynamic_memory[4] = cpu_to_be32(-1);
784             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
785                                             SPAPR_LMB_FLAGS_DRC_INVALID);
786         }
787 
788         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
789     }
790     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
791     g_free(int_buf);
792     if (ret < 0) {
793         return -1;
794     }
795     return 0;
796 }
797 
798 /*
799  * Adds ibm,dynamic-reconfiguration-memory node.
800  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
801  * of this device tree node.
802  */
803 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
804 {
805     MachineState *machine = MACHINE(spapr);
806     int ret, i, offset;
807     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
808     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
809     uint32_t *int_buf, *cur_index, buf_len;
810     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
811     MemoryDeviceInfoList *dimms = NULL;
812 
813     /*
814      * Don't create the node if there is no device memory
815      */
816     if (machine->ram_size == machine->maxram_size) {
817         return 0;
818     }
819 
820     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
821 
822     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
823                     sizeof(prop_lmb_size));
824     if (ret < 0) {
825         return ret;
826     }
827 
828     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
829     if (ret < 0) {
830         return ret;
831     }
832 
833     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
834     if (ret < 0) {
835         return ret;
836     }
837 
838     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
839     dimms = qmp_memory_device_list();
840     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
841         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
842     } else {
843         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
844     }
845     qapi_free_MemoryDeviceInfoList(dimms);
846 
847     if (ret < 0) {
848         return ret;
849     }
850 
851     /* ibm,associativity-lookup-arrays */
852     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
853     cur_index = int_buf = g_malloc0(buf_len);
854 
855     cur_index = int_buf;
856     int_buf[0] = cpu_to_be32(nr_nodes);
857     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
858     cur_index += 2;
859     for (i = 0; i < nr_nodes; i++) {
860         uint32_t associativity[] = {
861             cpu_to_be32(0x0),
862             cpu_to_be32(0x0),
863             cpu_to_be32(0x0),
864             cpu_to_be32(i)
865         };
866         memcpy(cur_index, associativity, sizeof(associativity));
867         cur_index += 4;
868     }
869     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
870             (cur_index - int_buf) * sizeof(uint32_t));
871     g_free(int_buf);
872 
873     return ret;
874 }
875 
876 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
877                                 sPAPROptionVector *ov5_updates)
878 {
879     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
880     int ret = 0, offset;
881 
882     /* Generate ibm,dynamic-reconfiguration-memory node if required */
883     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
884         g_assert(smc->dr_lmb_enabled);
885         ret = spapr_populate_drconf_memory(spapr, fdt);
886         if (ret) {
887             goto out;
888         }
889     }
890 
891     offset = fdt_path_offset(fdt, "/chosen");
892     if (offset < 0) {
893         offset = fdt_add_subnode(fdt, 0, "chosen");
894         if (offset < 0) {
895             return offset;
896         }
897     }
898     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
899                                  "ibm,architecture-vec-5");
900 
901 out:
902     return ret;
903 }
904 
905 static bool spapr_hotplugged_dev_before_cas(void)
906 {
907     Object *drc_container, *obj;
908     ObjectProperty *prop;
909     ObjectPropertyIterator iter;
910 
911     drc_container = container_get(object_get_root(), "/dr-connector");
912     object_property_iter_init(&iter, drc_container);
913     while ((prop = object_property_iter_next(&iter))) {
914         if (!strstart(prop->type, "link<", NULL)) {
915             continue;
916         }
917         obj = object_property_get_link(drc_container, prop->name, NULL);
918         if (spapr_drc_needed(obj)) {
919             return true;
920         }
921     }
922     return false;
923 }
924 
925 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
926                                  target_ulong addr, target_ulong size,
927                                  sPAPROptionVector *ov5_updates)
928 {
929     void *fdt, *fdt_skel;
930     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
931 
932     if (spapr_hotplugged_dev_before_cas()) {
933         return 1;
934     }
935 
936     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
937         error_report("SLOF provided an unexpected CAS buffer size "
938                      TARGET_FMT_lu " (min: %zu, max: %u)",
939                      size, sizeof(hdr), FW_MAX_SIZE);
940         exit(EXIT_FAILURE);
941     }
942 
943     size -= sizeof(hdr);
944 
945     /* Create skeleton */
946     fdt_skel = g_malloc0(size);
947     _FDT((fdt_create(fdt_skel, size)));
948     _FDT((fdt_finish_reservemap(fdt_skel)));
949     _FDT((fdt_begin_node(fdt_skel, "")));
950     _FDT((fdt_end_node(fdt_skel)));
951     _FDT((fdt_finish(fdt_skel)));
952     fdt = g_malloc0(size);
953     _FDT((fdt_open_into(fdt_skel, fdt, size)));
954     g_free(fdt_skel);
955 
956     /* Fixup cpu nodes */
957     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
958 
959     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
960         return -1;
961     }
962 
963     /* Pack resulting tree */
964     _FDT((fdt_pack(fdt)));
965 
966     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
967         trace_spapr_cas_failed(size);
968         return -1;
969     }
970 
971     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
972     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
973     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
974     g_free(fdt);
975 
976     return 0;
977 }
978 
979 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
980 {
981     int rtas;
982     GString *hypertas = g_string_sized_new(256);
983     GString *qemu_hypertas = g_string_sized_new(256);
984     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
985     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
986         memory_region_size(&MACHINE(spapr)->device_memory->mr);
987     uint32_t lrdr_capacity[] = {
988         cpu_to_be32(max_device_addr >> 32),
989         cpu_to_be32(max_device_addr & 0xffffffff),
990         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
991         cpu_to_be32(max_cpus / smp_threads),
992     };
993     uint32_t maxdomains[] = {
994         cpu_to_be32(4),
995         cpu_to_be32(0),
996         cpu_to_be32(0),
997         cpu_to_be32(0),
998         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
999     };
1000 
1001     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1002 
1003     /* hypertas */
1004     add_str(hypertas, "hcall-pft");
1005     add_str(hypertas, "hcall-term");
1006     add_str(hypertas, "hcall-dabr");
1007     add_str(hypertas, "hcall-interrupt");
1008     add_str(hypertas, "hcall-tce");
1009     add_str(hypertas, "hcall-vio");
1010     add_str(hypertas, "hcall-splpar");
1011     add_str(hypertas, "hcall-bulk");
1012     add_str(hypertas, "hcall-set-mode");
1013     add_str(hypertas, "hcall-sprg0");
1014     add_str(hypertas, "hcall-copy");
1015     add_str(hypertas, "hcall-debug");
1016     add_str(qemu_hypertas, "hcall-memop1");
1017 
1018     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1019         add_str(hypertas, "hcall-multi-tce");
1020     }
1021 
1022     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1023         add_str(hypertas, "hcall-hpt-resize");
1024     }
1025 
1026     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1027                      hypertas->str, hypertas->len));
1028     g_string_free(hypertas, TRUE);
1029     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1030                      qemu_hypertas->str, qemu_hypertas->len));
1031     g_string_free(qemu_hypertas, TRUE);
1032 
1033     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1034                      refpoints, sizeof(refpoints)));
1035 
1036     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1037                      maxdomains, sizeof(maxdomains)));
1038 
1039     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1040                           RTAS_ERROR_LOG_MAX));
1041     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1042                           RTAS_EVENT_SCAN_RATE));
1043 
1044     g_assert(msi_nonbroken);
1045     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1046 
1047     /*
1048      * According to PAPR, rtas ibm,os-term does not guarantee a return
1049      * back to the guest cpu.
1050      *
1051      * While an additional ibm,extended-os-term property indicates
1052      * that rtas call return will always occur. Set this property.
1053      */
1054     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1055 
1056     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1057                      lrdr_capacity, sizeof(lrdr_capacity)));
1058 
1059     spapr_dt_rtas_tokens(fdt, rtas);
1060 }
1061 
1062 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1063  * that the guest may request and thus the valid values for bytes 24..26 of
1064  * option vector 5: */
1065 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1066 {
1067     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1068 
1069     char val[2 * 4] = {
1070         23, 0x00, /* Xive mode, filled in below. */
1071         24, 0x00, /* Hash/Radix, filled in below. */
1072         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1073         26, 0x40, /* Radix options: GTSE == yes. */
1074     };
1075 
1076     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1077                           first_ppc_cpu->compat_pvr)) {
1078         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1079         val[3] = 0x00; /* Hash */
1080     } else if (kvm_enabled()) {
1081         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1082             val[3] = 0x80; /* OV5_MMU_BOTH */
1083         } else if (kvmppc_has_cap_mmu_radix()) {
1084             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1085         } else {
1086             val[3] = 0x00; /* Hash */
1087         }
1088     } else {
1089         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1090         val[3] = 0xC0;
1091     }
1092     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1093                      val, sizeof(val)));
1094 }
1095 
1096 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1097 {
1098     MachineState *machine = MACHINE(spapr);
1099     int chosen;
1100     const char *boot_device = machine->boot_order;
1101     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1102     size_t cb = 0;
1103     char *bootlist = get_boot_devices_list(&cb);
1104 
1105     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1106 
1107     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1108     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1109                           spapr->initrd_base));
1110     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1111                           spapr->initrd_base + spapr->initrd_size));
1112 
1113     if (spapr->kernel_size) {
1114         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1115                               cpu_to_be64(spapr->kernel_size) };
1116 
1117         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1118                          &kprop, sizeof(kprop)));
1119         if (spapr->kernel_le) {
1120             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1121         }
1122     }
1123     if (boot_menu) {
1124         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1125     }
1126     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1127     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1128     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1129 
1130     if (cb && bootlist) {
1131         int i;
1132 
1133         for (i = 0; i < cb; i++) {
1134             if (bootlist[i] == '\n') {
1135                 bootlist[i] = ' ';
1136             }
1137         }
1138         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1139     }
1140 
1141     if (boot_device && strlen(boot_device)) {
1142         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1143     }
1144 
1145     if (!spapr->has_graphics && stdout_path) {
1146         /*
1147          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1148          * kernel. New platforms should only use the "stdout-path" property. Set
1149          * the new property and continue using older property to remain
1150          * compatible with the existing firmware.
1151          */
1152         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1153         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1154     }
1155 
1156     spapr_dt_ov5_platform_support(fdt, chosen);
1157 
1158     g_free(stdout_path);
1159     g_free(bootlist);
1160 }
1161 
1162 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1163 {
1164     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1165      * KVM to work under pHyp with some guest co-operation */
1166     int hypervisor;
1167     uint8_t hypercall[16];
1168 
1169     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1170     /* indicate KVM hypercall interface */
1171     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1172     if (kvmppc_has_cap_fixup_hcalls()) {
1173         /*
1174          * Older KVM versions with older guest kernels were broken
1175          * with the magic page, don't allow the guest to map it.
1176          */
1177         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1178                                   sizeof(hypercall))) {
1179             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1180                              hypercall, sizeof(hypercall)));
1181         }
1182     }
1183 }
1184 
1185 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1186                              hwaddr rtas_addr,
1187                              hwaddr rtas_size)
1188 {
1189     MachineState *machine = MACHINE(spapr);
1190     MachineClass *mc = MACHINE_GET_CLASS(machine);
1191     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1192     int ret;
1193     void *fdt;
1194     sPAPRPHBState *phb;
1195     char *buf;
1196 
1197     fdt = g_malloc0(FDT_MAX_SIZE);
1198     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1199 
1200     /* Root node */
1201     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1202     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1203     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1204 
1205     /*
1206      * Add info to guest to indentify which host is it being run on
1207      * and what is the uuid of the guest
1208      */
1209     if (kvmppc_get_host_model(&buf)) {
1210         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1211         g_free(buf);
1212     }
1213     if (kvmppc_get_host_serial(&buf)) {
1214         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1215         g_free(buf);
1216     }
1217 
1218     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1219 
1220     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1221     if (qemu_uuid_set) {
1222         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1223     }
1224     g_free(buf);
1225 
1226     if (qemu_get_vm_name()) {
1227         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1228                                 qemu_get_vm_name()));
1229     }
1230 
1231     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1232     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1233 
1234     /* /interrupt controller */
1235     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1236 
1237     ret = spapr_populate_memory(spapr, fdt);
1238     if (ret < 0) {
1239         error_report("couldn't setup memory nodes in fdt");
1240         exit(1);
1241     }
1242 
1243     /* /vdevice */
1244     spapr_dt_vdevice(spapr->vio_bus, fdt);
1245 
1246     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1247         ret = spapr_rng_populate_dt(fdt);
1248         if (ret < 0) {
1249             error_report("could not set up rng device in the fdt");
1250             exit(1);
1251         }
1252     }
1253 
1254     QLIST_FOREACH(phb, &spapr->phbs, list) {
1255         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1256         if (ret < 0) {
1257             error_report("couldn't setup PCI devices in fdt");
1258             exit(1);
1259         }
1260     }
1261 
1262     /* cpus */
1263     spapr_populate_cpus_dt_node(fdt, spapr);
1264 
1265     if (smc->dr_lmb_enabled) {
1266         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1267     }
1268 
1269     if (mc->has_hotpluggable_cpus) {
1270         int offset = fdt_path_offset(fdt, "/cpus");
1271         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1272                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1273         if (ret < 0) {
1274             error_report("Couldn't set up CPU DR device tree properties");
1275             exit(1);
1276         }
1277     }
1278 
1279     /* /event-sources */
1280     spapr_dt_events(spapr, fdt);
1281 
1282     /* /rtas */
1283     spapr_dt_rtas(spapr, fdt);
1284 
1285     /* /chosen */
1286     spapr_dt_chosen(spapr, fdt);
1287 
1288     /* /hypervisor */
1289     if (kvm_enabled()) {
1290         spapr_dt_hypervisor(spapr, fdt);
1291     }
1292 
1293     /* Build memory reserve map */
1294     if (spapr->kernel_size) {
1295         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1296     }
1297     if (spapr->initrd_size) {
1298         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1299     }
1300 
1301     /* ibm,client-architecture-support updates */
1302     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1303     if (ret < 0) {
1304         error_report("couldn't setup CAS properties fdt");
1305         exit(1);
1306     }
1307 
1308     return fdt;
1309 }
1310 
1311 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1312 {
1313     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1314 }
1315 
1316 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1317                                     PowerPCCPU *cpu)
1318 {
1319     CPUPPCState *env = &cpu->env;
1320 
1321     /* The TCG path should also be holding the BQL at this point */
1322     g_assert(qemu_mutex_iothread_locked());
1323 
1324     if (msr_pr) {
1325         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1326         env->gpr[3] = H_PRIVILEGE;
1327     } else {
1328         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1329     }
1330 }
1331 
1332 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1333 {
1334     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1335 
1336     return spapr->patb_entry;
1337 }
1338 
1339 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1340 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1341 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1342 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1343 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1344 
1345 /*
1346  * Get the fd to access the kernel htab, re-opening it if necessary
1347  */
1348 static int get_htab_fd(sPAPRMachineState *spapr)
1349 {
1350     Error *local_err = NULL;
1351 
1352     if (spapr->htab_fd >= 0) {
1353         return spapr->htab_fd;
1354     }
1355 
1356     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1357     if (spapr->htab_fd < 0) {
1358         error_report_err(local_err);
1359     }
1360 
1361     return spapr->htab_fd;
1362 }
1363 
1364 void close_htab_fd(sPAPRMachineState *spapr)
1365 {
1366     if (spapr->htab_fd >= 0) {
1367         close(spapr->htab_fd);
1368     }
1369     spapr->htab_fd = -1;
1370 }
1371 
1372 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1373 {
1374     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1375 
1376     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1377 }
1378 
1379 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1380 {
1381     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1382 
1383     assert(kvm_enabled());
1384 
1385     if (!spapr->htab) {
1386         return 0;
1387     }
1388 
1389     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1390 }
1391 
1392 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1393                                                 hwaddr ptex, int n)
1394 {
1395     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1396     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1397 
1398     if (!spapr->htab) {
1399         /*
1400          * HTAB is controlled by KVM. Fetch into temporary buffer
1401          */
1402         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1403         kvmppc_read_hptes(hptes, ptex, n);
1404         return hptes;
1405     }
1406 
1407     /*
1408      * HTAB is controlled by QEMU. Just point to the internally
1409      * accessible PTEG.
1410      */
1411     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1412 }
1413 
1414 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1415                               const ppc_hash_pte64_t *hptes,
1416                               hwaddr ptex, int n)
1417 {
1418     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1419 
1420     if (!spapr->htab) {
1421         g_free((void *)hptes);
1422     }
1423 
1424     /* Nothing to do for qemu managed HPT */
1425 }
1426 
1427 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1428                              uint64_t pte0, uint64_t pte1)
1429 {
1430     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1431     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1432 
1433     if (!spapr->htab) {
1434         kvmppc_write_hpte(ptex, pte0, pte1);
1435     } else {
1436         stq_p(spapr->htab + offset, pte0);
1437         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1438     }
1439 }
1440 
1441 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1442 {
1443     int shift;
1444 
1445     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1446      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1447      * that's much more than is needed for Linux guests */
1448     shift = ctz64(pow2ceil(ramsize)) - 7;
1449     shift = MAX(shift, 18); /* Minimum architected size */
1450     shift = MIN(shift, 46); /* Maximum architected size */
1451     return shift;
1452 }
1453 
1454 void spapr_free_hpt(sPAPRMachineState *spapr)
1455 {
1456     g_free(spapr->htab);
1457     spapr->htab = NULL;
1458     spapr->htab_shift = 0;
1459     close_htab_fd(spapr);
1460 }
1461 
1462 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1463                           Error **errp)
1464 {
1465     long rc;
1466 
1467     /* Clean up any HPT info from a previous boot */
1468     spapr_free_hpt(spapr);
1469 
1470     rc = kvmppc_reset_htab(shift);
1471     if (rc < 0) {
1472         /* kernel-side HPT needed, but couldn't allocate one */
1473         error_setg_errno(errp, errno,
1474                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1475                          shift);
1476         /* This is almost certainly fatal, but if the caller really
1477          * wants to carry on with shift == 0, it's welcome to try */
1478     } else if (rc > 0) {
1479         /* kernel-side HPT allocated */
1480         if (rc != shift) {
1481             error_setg(errp,
1482                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1483                        shift, rc);
1484         }
1485 
1486         spapr->htab_shift = shift;
1487         spapr->htab = NULL;
1488     } else {
1489         /* kernel-side HPT not needed, allocate in userspace instead */
1490         size_t size = 1ULL << shift;
1491         int i;
1492 
1493         spapr->htab = qemu_memalign(size, size);
1494         if (!spapr->htab) {
1495             error_setg_errno(errp, errno,
1496                              "Could not allocate HPT of order %d", shift);
1497             return;
1498         }
1499 
1500         memset(spapr->htab, 0, size);
1501         spapr->htab_shift = shift;
1502 
1503         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1504             DIRTY_HPTE(HPTE(spapr->htab, i));
1505         }
1506     }
1507     /* We're setting up a hash table, so that means we're not radix */
1508     spapr->patb_entry = 0;
1509 }
1510 
1511 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1512 {
1513     int hpt_shift;
1514 
1515     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1516         || (spapr->cas_reboot
1517             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1518         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1519     } else {
1520         uint64_t current_ram_size;
1521 
1522         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1523         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1524     }
1525     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1526 
1527     if (spapr->vrma_adjust) {
1528         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1529                                           spapr->htab_shift);
1530     }
1531 }
1532 
1533 static int spapr_reset_drcs(Object *child, void *opaque)
1534 {
1535     sPAPRDRConnector *drc =
1536         (sPAPRDRConnector *) object_dynamic_cast(child,
1537                                                  TYPE_SPAPR_DR_CONNECTOR);
1538 
1539     if (drc) {
1540         spapr_drc_reset(drc);
1541     }
1542 
1543     return 0;
1544 }
1545 
1546 static void spapr_machine_reset(void)
1547 {
1548     MachineState *machine = MACHINE(qdev_get_machine());
1549     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1550     PowerPCCPU *first_ppc_cpu;
1551     uint32_t rtas_limit;
1552     hwaddr rtas_addr, fdt_addr;
1553     void *fdt;
1554     int rc;
1555 
1556     spapr_caps_apply(spapr);
1557 
1558     first_ppc_cpu = POWERPC_CPU(first_cpu);
1559     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1560         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1561                               spapr->max_compat_pvr)) {
1562         /* If using KVM with radix mode available, VCPUs can be started
1563          * without a HPT because KVM will start them in radix mode.
1564          * Set the GR bit in PATB so that we know there is no HPT. */
1565         spapr->patb_entry = PATBE1_GR;
1566     } else {
1567         spapr_setup_hpt_and_vrma(spapr);
1568     }
1569 
1570     /* if this reset wasn't generated by CAS, we should reset our
1571      * negotiated options and start from scratch */
1572     if (!spapr->cas_reboot) {
1573         spapr_ovec_cleanup(spapr->ov5_cas);
1574         spapr->ov5_cas = spapr_ovec_new();
1575 
1576         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1577     }
1578 
1579     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1580         spapr_irq_msi_reset(spapr);
1581     }
1582 
1583     qemu_devices_reset();
1584 
1585     /* DRC reset may cause a device to be unplugged. This will cause troubles
1586      * if this device is used by another device (eg, a running vhost backend
1587      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1588      * situations, we reset DRCs after all devices have been reset.
1589      */
1590     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1591 
1592     spapr_clear_pending_events(spapr);
1593 
1594     /*
1595      * We place the device tree and RTAS just below either the top of the RMA,
1596      * or just below 2GB, whichever is lowere, so that it can be
1597      * processed with 32-bit real mode code if necessary
1598      */
1599     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1600     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1601     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1602 
1603     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1604 
1605     spapr_load_rtas(spapr, fdt, rtas_addr);
1606 
1607     rc = fdt_pack(fdt);
1608 
1609     /* Should only fail if we've built a corrupted tree */
1610     assert(rc == 0);
1611 
1612     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1613         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1614                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1615         exit(1);
1616     }
1617 
1618     /* Load the fdt */
1619     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1620     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1621     g_free(fdt);
1622 
1623     /* Set up the entry state */
1624     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1625     first_ppc_cpu->env.gpr[5] = 0;
1626 
1627     spapr->cas_reboot = false;
1628 }
1629 
1630 static void spapr_create_nvram(sPAPRMachineState *spapr)
1631 {
1632     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1633     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1634 
1635     if (dinfo) {
1636         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1637                             &error_fatal);
1638     }
1639 
1640     qdev_init_nofail(dev);
1641 
1642     spapr->nvram = (struct sPAPRNVRAM *)dev;
1643 }
1644 
1645 static void spapr_rtc_create(sPAPRMachineState *spapr)
1646 {
1647     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1648     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1649                               &error_fatal);
1650     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1651                               &error_fatal);
1652     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1653                               "date", &error_fatal);
1654 }
1655 
1656 /* Returns whether we want to use VGA or not */
1657 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1658 {
1659     switch (vga_interface_type) {
1660     case VGA_NONE:
1661         return false;
1662     case VGA_DEVICE:
1663         return true;
1664     case VGA_STD:
1665     case VGA_VIRTIO:
1666         return pci_vga_init(pci_bus) != NULL;
1667     default:
1668         error_setg(errp,
1669                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1670         return false;
1671     }
1672 }
1673 
1674 static int spapr_pre_load(void *opaque)
1675 {
1676     int rc;
1677 
1678     rc = spapr_caps_pre_load(opaque);
1679     if (rc) {
1680         return rc;
1681     }
1682 
1683     return 0;
1684 }
1685 
1686 static int spapr_post_load(void *opaque, int version_id)
1687 {
1688     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1689     int err = 0;
1690 
1691     err = spapr_caps_post_migration(spapr);
1692     if (err) {
1693         return err;
1694     }
1695 
1696     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1697         CPUState *cs;
1698         CPU_FOREACH(cs) {
1699             PowerPCCPU *cpu = POWERPC_CPU(cs);
1700             icp_resend(ICP(cpu->intc));
1701         }
1702     }
1703 
1704     /* In earlier versions, there was no separate qdev for the PAPR
1705      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1706      * So when migrating from those versions, poke the incoming offset
1707      * value into the RTC device */
1708     if (version_id < 3) {
1709         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1710     }
1711 
1712     if (kvm_enabled() && spapr->patb_entry) {
1713         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1714         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1715         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1716 
1717         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1718         if (err) {
1719             error_report("Process table config unsupported by the host");
1720             return -EINVAL;
1721         }
1722     }
1723 
1724     return err;
1725 }
1726 
1727 static int spapr_pre_save(void *opaque)
1728 {
1729     int rc;
1730 
1731     rc = spapr_caps_pre_save(opaque);
1732     if (rc) {
1733         return rc;
1734     }
1735 
1736     return 0;
1737 }
1738 
1739 static bool version_before_3(void *opaque, int version_id)
1740 {
1741     return version_id < 3;
1742 }
1743 
1744 static bool spapr_pending_events_needed(void *opaque)
1745 {
1746     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1747     return !QTAILQ_EMPTY(&spapr->pending_events);
1748 }
1749 
1750 static const VMStateDescription vmstate_spapr_event_entry = {
1751     .name = "spapr_event_log_entry",
1752     .version_id = 1,
1753     .minimum_version_id = 1,
1754     .fields = (VMStateField[]) {
1755         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1756         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1757         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1758                                      NULL, extended_length),
1759         VMSTATE_END_OF_LIST()
1760     },
1761 };
1762 
1763 static const VMStateDescription vmstate_spapr_pending_events = {
1764     .name = "spapr_pending_events",
1765     .version_id = 1,
1766     .minimum_version_id = 1,
1767     .needed = spapr_pending_events_needed,
1768     .fields = (VMStateField[]) {
1769         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1770                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1771         VMSTATE_END_OF_LIST()
1772     },
1773 };
1774 
1775 static bool spapr_ov5_cas_needed(void *opaque)
1776 {
1777     sPAPRMachineState *spapr = opaque;
1778     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1779     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1780     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1781     bool cas_needed;
1782 
1783     /* Prior to the introduction of sPAPROptionVector, we had two option
1784      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1785      * Both of these options encode machine topology into the device-tree
1786      * in such a way that the now-booted OS should still be able to interact
1787      * appropriately with QEMU regardless of what options were actually
1788      * negotiatied on the source side.
1789      *
1790      * As such, we can avoid migrating the CAS-negotiated options if these
1791      * are the only options available on the current machine/platform.
1792      * Since these are the only options available for pseries-2.7 and
1793      * earlier, this allows us to maintain old->new/new->old migration
1794      * compatibility.
1795      *
1796      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1797      * via default pseries-2.8 machines and explicit command-line parameters.
1798      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1799      * of the actual CAS-negotiated values to continue working properly. For
1800      * example, availability of memory unplug depends on knowing whether
1801      * OV5_HP_EVT was negotiated via CAS.
1802      *
1803      * Thus, for any cases where the set of available CAS-negotiatable
1804      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1805      * include the CAS-negotiated options in the migration stream, unless
1806      * if they affect boot time behaviour only.
1807      */
1808     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1809     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1810     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1811 
1812     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1813      * the mask itself since in the future it's possible "legacy" bits may be
1814      * removed via machine options, which could generate a false positive
1815      * that breaks migration.
1816      */
1817     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1818     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1819 
1820     spapr_ovec_cleanup(ov5_mask);
1821     spapr_ovec_cleanup(ov5_legacy);
1822     spapr_ovec_cleanup(ov5_removed);
1823 
1824     return cas_needed;
1825 }
1826 
1827 static const VMStateDescription vmstate_spapr_ov5_cas = {
1828     .name = "spapr_option_vector_ov5_cas",
1829     .version_id = 1,
1830     .minimum_version_id = 1,
1831     .needed = spapr_ov5_cas_needed,
1832     .fields = (VMStateField[]) {
1833         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1834                                  vmstate_spapr_ovec, sPAPROptionVector),
1835         VMSTATE_END_OF_LIST()
1836     },
1837 };
1838 
1839 static bool spapr_patb_entry_needed(void *opaque)
1840 {
1841     sPAPRMachineState *spapr = opaque;
1842 
1843     return !!spapr->patb_entry;
1844 }
1845 
1846 static const VMStateDescription vmstate_spapr_patb_entry = {
1847     .name = "spapr_patb_entry",
1848     .version_id = 1,
1849     .minimum_version_id = 1,
1850     .needed = spapr_patb_entry_needed,
1851     .fields = (VMStateField[]) {
1852         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1853         VMSTATE_END_OF_LIST()
1854     },
1855 };
1856 
1857 static bool spapr_irq_map_needed(void *opaque)
1858 {
1859     sPAPRMachineState *spapr = opaque;
1860 
1861     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1862 }
1863 
1864 static const VMStateDescription vmstate_spapr_irq_map = {
1865     .name = "spapr_irq_map",
1866     .version_id = 1,
1867     .minimum_version_id = 1,
1868     .needed = spapr_irq_map_needed,
1869     .fields = (VMStateField[]) {
1870         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1871         VMSTATE_END_OF_LIST()
1872     },
1873 };
1874 
1875 static const VMStateDescription vmstate_spapr = {
1876     .name = "spapr",
1877     .version_id = 3,
1878     .minimum_version_id = 1,
1879     .pre_load = spapr_pre_load,
1880     .post_load = spapr_post_load,
1881     .pre_save = spapr_pre_save,
1882     .fields = (VMStateField[]) {
1883         /* used to be @next_irq */
1884         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1885 
1886         /* RTC offset */
1887         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1888 
1889         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1890         VMSTATE_END_OF_LIST()
1891     },
1892     .subsections = (const VMStateDescription*[]) {
1893         &vmstate_spapr_ov5_cas,
1894         &vmstate_spapr_patb_entry,
1895         &vmstate_spapr_pending_events,
1896         &vmstate_spapr_cap_htm,
1897         &vmstate_spapr_cap_vsx,
1898         &vmstate_spapr_cap_dfp,
1899         &vmstate_spapr_cap_cfpc,
1900         &vmstate_spapr_cap_sbbc,
1901         &vmstate_spapr_cap_ibs,
1902         &vmstate_spapr_irq_map,
1903         NULL
1904     }
1905 };
1906 
1907 static int htab_save_setup(QEMUFile *f, void *opaque)
1908 {
1909     sPAPRMachineState *spapr = opaque;
1910 
1911     /* "Iteration" header */
1912     if (!spapr->htab_shift) {
1913         qemu_put_be32(f, -1);
1914     } else {
1915         qemu_put_be32(f, spapr->htab_shift);
1916     }
1917 
1918     if (spapr->htab) {
1919         spapr->htab_save_index = 0;
1920         spapr->htab_first_pass = true;
1921     } else {
1922         if (spapr->htab_shift) {
1923             assert(kvm_enabled());
1924         }
1925     }
1926 
1927 
1928     return 0;
1929 }
1930 
1931 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1932                             int chunkstart, int n_valid, int n_invalid)
1933 {
1934     qemu_put_be32(f, chunkstart);
1935     qemu_put_be16(f, n_valid);
1936     qemu_put_be16(f, n_invalid);
1937     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1938                     HASH_PTE_SIZE_64 * n_valid);
1939 }
1940 
1941 static void htab_save_end_marker(QEMUFile *f)
1942 {
1943     qemu_put_be32(f, 0);
1944     qemu_put_be16(f, 0);
1945     qemu_put_be16(f, 0);
1946 }
1947 
1948 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1949                                  int64_t max_ns)
1950 {
1951     bool has_timeout = max_ns != -1;
1952     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1953     int index = spapr->htab_save_index;
1954     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1955 
1956     assert(spapr->htab_first_pass);
1957 
1958     do {
1959         int chunkstart;
1960 
1961         /* Consume invalid HPTEs */
1962         while ((index < htabslots)
1963                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1964             CLEAN_HPTE(HPTE(spapr->htab, index));
1965             index++;
1966         }
1967 
1968         /* Consume valid HPTEs */
1969         chunkstart = index;
1970         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1971                && HPTE_VALID(HPTE(spapr->htab, index))) {
1972             CLEAN_HPTE(HPTE(spapr->htab, index));
1973             index++;
1974         }
1975 
1976         if (index > chunkstart) {
1977             int n_valid = index - chunkstart;
1978 
1979             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1980 
1981             if (has_timeout &&
1982                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1983                 break;
1984             }
1985         }
1986     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1987 
1988     if (index >= htabslots) {
1989         assert(index == htabslots);
1990         index = 0;
1991         spapr->htab_first_pass = false;
1992     }
1993     spapr->htab_save_index = index;
1994 }
1995 
1996 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1997                                 int64_t max_ns)
1998 {
1999     bool final = max_ns < 0;
2000     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2001     int examined = 0, sent = 0;
2002     int index = spapr->htab_save_index;
2003     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2004 
2005     assert(!spapr->htab_first_pass);
2006 
2007     do {
2008         int chunkstart, invalidstart;
2009 
2010         /* Consume non-dirty HPTEs */
2011         while ((index < htabslots)
2012                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2013             index++;
2014             examined++;
2015         }
2016 
2017         chunkstart = index;
2018         /* Consume valid dirty HPTEs */
2019         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2020                && HPTE_DIRTY(HPTE(spapr->htab, index))
2021                && HPTE_VALID(HPTE(spapr->htab, index))) {
2022             CLEAN_HPTE(HPTE(spapr->htab, index));
2023             index++;
2024             examined++;
2025         }
2026 
2027         invalidstart = index;
2028         /* Consume invalid dirty HPTEs */
2029         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2030                && HPTE_DIRTY(HPTE(spapr->htab, index))
2031                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2032             CLEAN_HPTE(HPTE(spapr->htab, index));
2033             index++;
2034             examined++;
2035         }
2036 
2037         if (index > chunkstart) {
2038             int n_valid = invalidstart - chunkstart;
2039             int n_invalid = index - invalidstart;
2040 
2041             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2042             sent += index - chunkstart;
2043 
2044             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2045                 break;
2046             }
2047         }
2048 
2049         if (examined >= htabslots) {
2050             break;
2051         }
2052 
2053         if (index >= htabslots) {
2054             assert(index == htabslots);
2055             index = 0;
2056         }
2057     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2058 
2059     if (index >= htabslots) {
2060         assert(index == htabslots);
2061         index = 0;
2062     }
2063 
2064     spapr->htab_save_index = index;
2065 
2066     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2067 }
2068 
2069 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2070 #define MAX_KVM_BUF_SIZE    2048
2071 
2072 static int htab_save_iterate(QEMUFile *f, void *opaque)
2073 {
2074     sPAPRMachineState *spapr = opaque;
2075     int fd;
2076     int rc = 0;
2077 
2078     /* Iteration header */
2079     if (!spapr->htab_shift) {
2080         qemu_put_be32(f, -1);
2081         return 1;
2082     } else {
2083         qemu_put_be32(f, 0);
2084     }
2085 
2086     if (!spapr->htab) {
2087         assert(kvm_enabled());
2088 
2089         fd = get_htab_fd(spapr);
2090         if (fd < 0) {
2091             return fd;
2092         }
2093 
2094         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2095         if (rc < 0) {
2096             return rc;
2097         }
2098     } else  if (spapr->htab_first_pass) {
2099         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2100     } else {
2101         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2102     }
2103 
2104     htab_save_end_marker(f);
2105 
2106     return rc;
2107 }
2108 
2109 static int htab_save_complete(QEMUFile *f, void *opaque)
2110 {
2111     sPAPRMachineState *spapr = opaque;
2112     int fd;
2113 
2114     /* Iteration header */
2115     if (!spapr->htab_shift) {
2116         qemu_put_be32(f, -1);
2117         return 0;
2118     } else {
2119         qemu_put_be32(f, 0);
2120     }
2121 
2122     if (!spapr->htab) {
2123         int rc;
2124 
2125         assert(kvm_enabled());
2126 
2127         fd = get_htab_fd(spapr);
2128         if (fd < 0) {
2129             return fd;
2130         }
2131 
2132         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2133         if (rc < 0) {
2134             return rc;
2135         }
2136     } else {
2137         if (spapr->htab_first_pass) {
2138             htab_save_first_pass(f, spapr, -1);
2139         }
2140         htab_save_later_pass(f, spapr, -1);
2141     }
2142 
2143     /* End marker */
2144     htab_save_end_marker(f);
2145 
2146     return 0;
2147 }
2148 
2149 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2150 {
2151     sPAPRMachineState *spapr = opaque;
2152     uint32_t section_hdr;
2153     int fd = -1;
2154     Error *local_err = NULL;
2155 
2156     if (version_id < 1 || version_id > 1) {
2157         error_report("htab_load() bad version");
2158         return -EINVAL;
2159     }
2160 
2161     section_hdr = qemu_get_be32(f);
2162 
2163     if (section_hdr == -1) {
2164         spapr_free_hpt(spapr);
2165         return 0;
2166     }
2167 
2168     if (section_hdr) {
2169         /* First section gives the htab size */
2170         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2171         if (local_err) {
2172             error_report_err(local_err);
2173             return -EINVAL;
2174         }
2175         return 0;
2176     }
2177 
2178     if (!spapr->htab) {
2179         assert(kvm_enabled());
2180 
2181         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2182         if (fd < 0) {
2183             error_report_err(local_err);
2184             return fd;
2185         }
2186     }
2187 
2188     while (true) {
2189         uint32_t index;
2190         uint16_t n_valid, n_invalid;
2191 
2192         index = qemu_get_be32(f);
2193         n_valid = qemu_get_be16(f);
2194         n_invalid = qemu_get_be16(f);
2195 
2196         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2197             /* End of Stream */
2198             break;
2199         }
2200 
2201         if ((index + n_valid + n_invalid) >
2202             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2203             /* Bad index in stream */
2204             error_report(
2205                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2206                 index, n_valid, n_invalid, spapr->htab_shift);
2207             return -EINVAL;
2208         }
2209 
2210         if (spapr->htab) {
2211             if (n_valid) {
2212                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2213                                 HASH_PTE_SIZE_64 * n_valid);
2214             }
2215             if (n_invalid) {
2216                 memset(HPTE(spapr->htab, index + n_valid), 0,
2217                        HASH_PTE_SIZE_64 * n_invalid);
2218             }
2219         } else {
2220             int rc;
2221 
2222             assert(fd >= 0);
2223 
2224             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2225             if (rc < 0) {
2226                 return rc;
2227             }
2228         }
2229     }
2230 
2231     if (!spapr->htab) {
2232         assert(fd >= 0);
2233         close(fd);
2234     }
2235 
2236     return 0;
2237 }
2238 
2239 static void htab_save_cleanup(void *opaque)
2240 {
2241     sPAPRMachineState *spapr = opaque;
2242 
2243     close_htab_fd(spapr);
2244 }
2245 
2246 static SaveVMHandlers savevm_htab_handlers = {
2247     .save_setup = htab_save_setup,
2248     .save_live_iterate = htab_save_iterate,
2249     .save_live_complete_precopy = htab_save_complete,
2250     .save_cleanup = htab_save_cleanup,
2251     .load_state = htab_load,
2252 };
2253 
2254 static void spapr_boot_set(void *opaque, const char *boot_device,
2255                            Error **errp)
2256 {
2257     MachineState *machine = MACHINE(opaque);
2258     machine->boot_order = g_strdup(boot_device);
2259 }
2260 
2261 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2262 {
2263     MachineState *machine = MACHINE(spapr);
2264     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2265     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2266     int i;
2267 
2268     for (i = 0; i < nr_lmbs; i++) {
2269         uint64_t addr;
2270 
2271         addr = i * lmb_size + machine->device_memory->base;
2272         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2273                                addr / lmb_size);
2274     }
2275 }
2276 
2277 /*
2278  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2279  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2280  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2281  */
2282 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2283 {
2284     int i;
2285 
2286     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2287         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2288                    " is not aligned to %" PRIu64 " MiB",
2289                    machine->ram_size,
2290                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2291         return;
2292     }
2293 
2294     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2295         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2296                    " is not aligned to %" PRIu64 " MiB",
2297                    machine->ram_size,
2298                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2299         return;
2300     }
2301 
2302     for (i = 0; i < nb_numa_nodes; i++) {
2303         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2304             error_setg(errp,
2305                        "Node %d memory size 0x%" PRIx64
2306                        " is not aligned to %" PRIu64 " MiB",
2307                        i, numa_info[i].node_mem,
2308                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2309             return;
2310         }
2311     }
2312 }
2313 
2314 /* find cpu slot in machine->possible_cpus by core_id */
2315 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2316 {
2317     int index = id / smp_threads;
2318 
2319     if (index >= ms->possible_cpus->len) {
2320         return NULL;
2321     }
2322     if (idx) {
2323         *idx = index;
2324     }
2325     return &ms->possible_cpus->cpus[index];
2326 }
2327 
2328 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2329 {
2330     Error *local_err = NULL;
2331     bool vsmt_user = !!spapr->vsmt;
2332     int kvm_smt = kvmppc_smt_threads();
2333     int ret;
2334 
2335     if (!kvm_enabled() && (smp_threads > 1)) {
2336         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2337                      "on a pseries machine");
2338         goto out;
2339     }
2340     if (!is_power_of_2(smp_threads)) {
2341         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2342                      "machine because it must be a power of 2", smp_threads);
2343         goto out;
2344     }
2345 
2346     /* Detemine the VSMT mode to use: */
2347     if (vsmt_user) {
2348         if (spapr->vsmt < smp_threads) {
2349             error_setg(&local_err, "Cannot support VSMT mode %d"
2350                          " because it must be >= threads/core (%d)",
2351                          spapr->vsmt, smp_threads);
2352             goto out;
2353         }
2354         /* In this case, spapr->vsmt has been set by the command line */
2355     } else {
2356         /*
2357          * Default VSMT value is tricky, because we need it to be as
2358          * consistent as possible (for migration), but this requires
2359          * changing it for at least some existing cases.  We pick 8 as
2360          * the value that we'd get with KVM on POWER8, the
2361          * overwhelmingly common case in production systems.
2362          */
2363         spapr->vsmt = MAX(8, smp_threads);
2364     }
2365 
2366     /* KVM: If necessary, set the SMT mode: */
2367     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2368         ret = kvmppc_set_smt_threads(spapr->vsmt);
2369         if (ret) {
2370             /* Looks like KVM isn't able to change VSMT mode */
2371             error_setg(&local_err,
2372                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2373                        spapr->vsmt, ret);
2374             /* We can live with that if the default one is big enough
2375              * for the number of threads, and a submultiple of the one
2376              * we want.  In this case we'll waste some vcpu ids, but
2377              * behaviour will be correct */
2378             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2379                 warn_report_err(local_err);
2380                 local_err = NULL;
2381                 goto out;
2382             } else {
2383                 if (!vsmt_user) {
2384                     error_append_hint(&local_err,
2385                                       "On PPC, a VM with %d threads/core"
2386                                       " on a host with %d threads/core"
2387                                       " requires the use of VSMT mode %d.\n",
2388                                       smp_threads, kvm_smt, spapr->vsmt);
2389                 }
2390                 kvmppc_hint_smt_possible(&local_err);
2391                 goto out;
2392             }
2393         }
2394     }
2395     /* else TCG: nothing to do currently */
2396 out:
2397     error_propagate(errp, local_err);
2398 }
2399 
2400 static void spapr_init_cpus(sPAPRMachineState *spapr)
2401 {
2402     MachineState *machine = MACHINE(spapr);
2403     MachineClass *mc = MACHINE_GET_CLASS(machine);
2404     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2405     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2406     const CPUArchIdList *possible_cpus;
2407     int boot_cores_nr = smp_cpus / smp_threads;
2408     int i;
2409 
2410     possible_cpus = mc->possible_cpu_arch_ids(machine);
2411     if (mc->has_hotpluggable_cpus) {
2412         if (smp_cpus % smp_threads) {
2413             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2414                          smp_cpus, smp_threads);
2415             exit(1);
2416         }
2417         if (max_cpus % smp_threads) {
2418             error_report("max_cpus (%u) must be multiple of threads (%u)",
2419                          max_cpus, smp_threads);
2420             exit(1);
2421         }
2422     } else {
2423         if (max_cpus != smp_cpus) {
2424             error_report("This machine version does not support CPU hotplug");
2425             exit(1);
2426         }
2427         boot_cores_nr = possible_cpus->len;
2428     }
2429 
2430     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2431      * call xics_max_server_number() or spapr_vcpu_id().
2432      */
2433     spapr_set_vsmt_mode(spapr, &error_fatal);
2434 
2435     if (smc->pre_2_10_has_unused_icps) {
2436         int i;
2437 
2438         for (i = 0; i < xics_max_server_number(spapr); i++) {
2439             /* Dummy entries get deregistered when real ICPState objects
2440              * are registered during CPU core hotplug.
2441              */
2442             pre_2_10_vmstate_register_dummy_icp(i);
2443         }
2444     }
2445 
2446     for (i = 0; i < possible_cpus->len; i++) {
2447         int core_id = i * smp_threads;
2448 
2449         if (mc->has_hotpluggable_cpus) {
2450             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2451                                    spapr_vcpu_id(spapr, core_id));
2452         }
2453 
2454         if (i < boot_cores_nr) {
2455             Object *core  = object_new(type);
2456             int nr_threads = smp_threads;
2457 
2458             /* Handle the partially filled core for older machine types */
2459             if ((i + 1) * smp_threads >= smp_cpus) {
2460                 nr_threads = smp_cpus - i * smp_threads;
2461             }
2462 
2463             object_property_set_int(core, nr_threads, "nr-threads",
2464                                     &error_fatal);
2465             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2466                                     &error_fatal);
2467             object_property_set_bool(core, true, "realized", &error_fatal);
2468         }
2469     }
2470 }
2471 
2472 /* pSeries LPAR / sPAPR hardware init */
2473 static void spapr_machine_init(MachineState *machine)
2474 {
2475     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2476     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2477     const char *kernel_filename = machine->kernel_filename;
2478     const char *initrd_filename = machine->initrd_filename;
2479     PCIHostState *phb;
2480     int i;
2481     MemoryRegion *sysmem = get_system_memory();
2482     MemoryRegion *ram = g_new(MemoryRegion, 1);
2483     hwaddr node0_size = spapr_node0_size(machine);
2484     long load_limit, fw_size;
2485     char *filename;
2486     Error *resize_hpt_err = NULL;
2487 
2488     msi_nonbroken = true;
2489 
2490     QLIST_INIT(&spapr->phbs);
2491     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2492 
2493     /* Determine capabilities to run with */
2494     spapr_caps_init(spapr);
2495 
2496     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2497     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2498         /*
2499          * If the user explicitly requested a mode we should either
2500          * supply it, or fail completely (which we do below).  But if
2501          * it's not set explicitly, we reset our mode to something
2502          * that works
2503          */
2504         if (resize_hpt_err) {
2505             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2506             error_free(resize_hpt_err);
2507             resize_hpt_err = NULL;
2508         } else {
2509             spapr->resize_hpt = smc->resize_hpt_default;
2510         }
2511     }
2512 
2513     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2514 
2515     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2516         /*
2517          * User requested HPT resize, but this host can't supply it.  Bail out
2518          */
2519         error_report_err(resize_hpt_err);
2520         exit(1);
2521     }
2522 
2523     spapr->rma_size = node0_size;
2524 
2525     /* With KVM, we don't actually know whether KVM supports an
2526      * unbounded RMA (PR KVM) or is limited by the hash table size
2527      * (HV KVM using VRMA), so we always assume the latter
2528      *
2529      * In that case, we also limit the initial allocations for RTAS
2530      * etc... to 256M since we have no way to know what the VRMA size
2531      * is going to be as it depends on the size of the hash table
2532      * which isn't determined yet.
2533      */
2534     if (kvm_enabled()) {
2535         spapr->vrma_adjust = 1;
2536         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2537     }
2538 
2539     /* Actually we don't support unbounded RMA anymore since we added
2540      * proper emulation of HV mode. The max we can get is 16G which
2541      * also happens to be what we configure for PAPR mode so make sure
2542      * we don't do anything bigger than that
2543      */
2544     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2545 
2546     if (spapr->rma_size > node0_size) {
2547         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2548                      spapr->rma_size);
2549         exit(1);
2550     }
2551 
2552     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2553     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2554 
2555     /* Set up Interrupt Controller before we create the VCPUs */
2556     smc->irq->init(spapr, &error_fatal);
2557 
2558     /* Set up containers for ibm,client-architecture-support negotiated options
2559      */
2560     spapr->ov5 = spapr_ovec_new();
2561     spapr->ov5_cas = spapr_ovec_new();
2562 
2563     if (smc->dr_lmb_enabled) {
2564         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2565         spapr_validate_node_memory(machine, &error_fatal);
2566     }
2567 
2568     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2569 
2570     /* advertise support for dedicated HP event source to guests */
2571     if (spapr->use_hotplug_event_source) {
2572         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2573     }
2574 
2575     /* advertise support for HPT resizing */
2576     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2577         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2578     }
2579 
2580     /* advertise support for ibm,dyamic-memory-v2 */
2581     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2582 
2583     /* init CPUs */
2584     spapr_init_cpus(spapr);
2585 
2586     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2587         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2588                               spapr->max_compat_pvr)) {
2589         /* KVM and TCG always allow GTSE with radix... */
2590         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2591     }
2592     /* ... but not with hash (currently). */
2593 
2594     if (kvm_enabled()) {
2595         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2596         kvmppc_enable_logical_ci_hcalls();
2597         kvmppc_enable_set_mode_hcall();
2598 
2599         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2600         kvmppc_enable_clear_ref_mod_hcalls();
2601     }
2602 
2603     /* allocate RAM */
2604     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2605                                          machine->ram_size);
2606     memory_region_add_subregion(sysmem, 0, ram);
2607 
2608     /* always allocate the device memory information */
2609     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2610 
2611     /* initialize hotplug memory address space */
2612     if (machine->ram_size < machine->maxram_size) {
2613         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2614         /*
2615          * Limit the number of hotpluggable memory slots to half the number
2616          * slots that KVM supports, leaving the other half for PCI and other
2617          * devices. However ensure that number of slots doesn't drop below 32.
2618          */
2619         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2620                            SPAPR_MAX_RAM_SLOTS;
2621 
2622         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2623             max_memslots = SPAPR_MAX_RAM_SLOTS;
2624         }
2625         if (machine->ram_slots > max_memslots) {
2626             error_report("Specified number of memory slots %"
2627                          PRIu64" exceeds max supported %d",
2628                          machine->ram_slots, max_memslots);
2629             exit(1);
2630         }
2631 
2632         machine->device_memory->base = ROUND_UP(machine->ram_size,
2633                                                 SPAPR_DEVICE_MEM_ALIGN);
2634         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2635                            "device-memory", device_mem_size);
2636         memory_region_add_subregion(sysmem, machine->device_memory->base,
2637                                     &machine->device_memory->mr);
2638     }
2639 
2640     if (smc->dr_lmb_enabled) {
2641         spapr_create_lmb_dr_connectors(spapr);
2642     }
2643 
2644     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2645     if (!filename) {
2646         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2647         exit(1);
2648     }
2649     spapr->rtas_size = get_image_size(filename);
2650     if (spapr->rtas_size < 0) {
2651         error_report("Could not get size of LPAR rtas '%s'", filename);
2652         exit(1);
2653     }
2654     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2655     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2656         error_report("Could not load LPAR rtas '%s'", filename);
2657         exit(1);
2658     }
2659     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2660         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2661                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2662         exit(1);
2663     }
2664     g_free(filename);
2665 
2666     /* Set up RTAS event infrastructure */
2667     spapr_events_init(spapr);
2668 
2669     /* Set up the RTC RTAS interfaces */
2670     spapr_rtc_create(spapr);
2671 
2672     /* Set up VIO bus */
2673     spapr->vio_bus = spapr_vio_bus_init();
2674 
2675     for (i = 0; i < serial_max_hds(); i++) {
2676         if (serial_hd(i)) {
2677             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2678         }
2679     }
2680 
2681     /* We always have at least the nvram device on VIO */
2682     spapr_create_nvram(spapr);
2683 
2684     /* Set up PCI */
2685     spapr_pci_rtas_init();
2686 
2687     phb = spapr_create_phb(spapr, 0);
2688 
2689     for (i = 0; i < nb_nics; i++) {
2690         NICInfo *nd = &nd_table[i];
2691 
2692         if (!nd->model) {
2693             nd->model = g_strdup("spapr-vlan");
2694         }
2695 
2696         if (g_str_equal(nd->model, "spapr-vlan") ||
2697             g_str_equal(nd->model, "ibmveth")) {
2698             spapr_vlan_create(spapr->vio_bus, nd);
2699         } else {
2700             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2701         }
2702     }
2703 
2704     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2705         spapr_vscsi_create(spapr->vio_bus);
2706     }
2707 
2708     /* Graphics */
2709     if (spapr_vga_init(phb->bus, &error_fatal)) {
2710         spapr->has_graphics = true;
2711         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2712     }
2713 
2714     if (machine->usb) {
2715         if (smc->use_ohci_by_default) {
2716             pci_create_simple(phb->bus, -1, "pci-ohci");
2717         } else {
2718             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2719         }
2720 
2721         if (spapr->has_graphics) {
2722             USBBus *usb_bus = usb_bus_find(-1);
2723 
2724             usb_create_simple(usb_bus, "usb-kbd");
2725             usb_create_simple(usb_bus, "usb-mouse");
2726         }
2727     }
2728 
2729     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2730         error_report(
2731             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2732             MIN_RMA_SLOF);
2733         exit(1);
2734     }
2735 
2736     if (kernel_filename) {
2737         uint64_t lowaddr = 0;
2738 
2739         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2740                                       NULL, NULL, &lowaddr, NULL, 1,
2741                                       PPC_ELF_MACHINE, 0, 0);
2742         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2743             spapr->kernel_size = load_elf(kernel_filename,
2744                                           translate_kernel_address, NULL, NULL,
2745                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2746                                           0, 0);
2747             spapr->kernel_le = spapr->kernel_size > 0;
2748         }
2749         if (spapr->kernel_size < 0) {
2750             error_report("error loading %s: %s", kernel_filename,
2751                          load_elf_strerror(spapr->kernel_size));
2752             exit(1);
2753         }
2754 
2755         /* load initrd */
2756         if (initrd_filename) {
2757             /* Try to locate the initrd in the gap between the kernel
2758              * and the firmware. Add a bit of space just in case
2759              */
2760             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2761                                   + 0x1ffff) & ~0xffff;
2762             spapr->initrd_size = load_image_targphys(initrd_filename,
2763                                                      spapr->initrd_base,
2764                                                      load_limit
2765                                                      - spapr->initrd_base);
2766             if (spapr->initrd_size < 0) {
2767                 error_report("could not load initial ram disk '%s'",
2768                              initrd_filename);
2769                 exit(1);
2770             }
2771         }
2772     }
2773 
2774     if (bios_name == NULL) {
2775         bios_name = FW_FILE_NAME;
2776     }
2777     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2778     if (!filename) {
2779         error_report("Could not find LPAR firmware '%s'", bios_name);
2780         exit(1);
2781     }
2782     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2783     if (fw_size <= 0) {
2784         error_report("Could not load LPAR firmware '%s'", filename);
2785         exit(1);
2786     }
2787     g_free(filename);
2788 
2789     /* FIXME: Should register things through the MachineState's qdev
2790      * interface, this is a legacy from the sPAPREnvironment structure
2791      * which predated MachineState but had a similar function */
2792     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2793     register_savevm_live(NULL, "spapr/htab", -1, 1,
2794                          &savevm_htab_handlers, spapr);
2795 
2796     qemu_register_boot_set(spapr_boot_set, spapr);
2797 
2798     if (kvm_enabled()) {
2799         /* to stop and start vmclock */
2800         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2801                                          &spapr->tb);
2802 
2803         kvmppc_spapr_enable_inkernel_multitce();
2804     }
2805 }
2806 
2807 static int spapr_kvm_type(const char *vm_type)
2808 {
2809     if (!vm_type) {
2810         return 0;
2811     }
2812 
2813     if (!strcmp(vm_type, "HV")) {
2814         return 1;
2815     }
2816 
2817     if (!strcmp(vm_type, "PR")) {
2818         return 2;
2819     }
2820 
2821     error_report("Unknown kvm-type specified '%s'", vm_type);
2822     exit(1);
2823 }
2824 
2825 /*
2826  * Implementation of an interface to adjust firmware path
2827  * for the bootindex property handling.
2828  */
2829 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2830                                    DeviceState *dev)
2831 {
2832 #define CAST(type, obj, name) \
2833     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2834     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2835     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2836     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2837 
2838     if (d) {
2839         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2840         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2841         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2842 
2843         if (spapr) {
2844             /*
2845              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2846              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2847              * in the top 16 bits of the 64-bit LUN
2848              */
2849             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2850             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2851                                    (uint64_t)id << 48);
2852         } else if (virtio) {
2853             /*
2854              * We use SRP luns of the form 01000000 | (target << 8) | lun
2855              * in the top 32 bits of the 64-bit LUN
2856              * Note: the quote above is from SLOF and it is wrong,
2857              * the actual binding is:
2858              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2859              */
2860             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2861             if (d->lun >= 256) {
2862                 /* Use the LUN "flat space addressing method" */
2863                 id |= 0x4000;
2864             }
2865             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2866                                    (uint64_t)id << 32);
2867         } else if (usb) {
2868             /*
2869              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2870              * in the top 32 bits of the 64-bit LUN
2871              */
2872             unsigned usb_port = atoi(usb->port->path);
2873             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2874             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2875                                    (uint64_t)id << 32);
2876         }
2877     }
2878 
2879     /*
2880      * SLOF probes the USB devices, and if it recognizes that the device is a
2881      * storage device, it changes its name to "storage" instead of "usb-host",
2882      * and additionally adds a child node for the SCSI LUN, so the correct
2883      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2884      */
2885     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2886         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2887         if (usb_host_dev_is_scsi_storage(usbdev)) {
2888             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2889         }
2890     }
2891 
2892     if (phb) {
2893         /* Replace "pci" with "pci@800000020000000" */
2894         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2895     }
2896 
2897     if (vsc) {
2898         /* Same logic as virtio above */
2899         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2900         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2901     }
2902 
2903     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2904         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2905         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2906         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2907     }
2908 
2909     return NULL;
2910 }
2911 
2912 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2913 {
2914     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2915 
2916     return g_strdup(spapr->kvm_type);
2917 }
2918 
2919 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2920 {
2921     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2922 
2923     g_free(spapr->kvm_type);
2924     spapr->kvm_type = g_strdup(value);
2925 }
2926 
2927 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2928 {
2929     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2930 
2931     return spapr->use_hotplug_event_source;
2932 }
2933 
2934 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2935                                             Error **errp)
2936 {
2937     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2938 
2939     spapr->use_hotplug_event_source = value;
2940 }
2941 
2942 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2943 {
2944     return true;
2945 }
2946 
2947 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2948 {
2949     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2950 
2951     switch (spapr->resize_hpt) {
2952     case SPAPR_RESIZE_HPT_DEFAULT:
2953         return g_strdup("default");
2954     case SPAPR_RESIZE_HPT_DISABLED:
2955         return g_strdup("disabled");
2956     case SPAPR_RESIZE_HPT_ENABLED:
2957         return g_strdup("enabled");
2958     case SPAPR_RESIZE_HPT_REQUIRED:
2959         return g_strdup("required");
2960     }
2961     g_assert_not_reached();
2962 }
2963 
2964 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2965 {
2966     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2967 
2968     if (strcmp(value, "default") == 0) {
2969         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2970     } else if (strcmp(value, "disabled") == 0) {
2971         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2972     } else if (strcmp(value, "enabled") == 0) {
2973         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2974     } else if (strcmp(value, "required") == 0) {
2975         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2976     } else {
2977         error_setg(errp, "Bad value for \"resize-hpt\" property");
2978     }
2979 }
2980 
2981 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2982                                    void *opaque, Error **errp)
2983 {
2984     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2985 }
2986 
2987 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
2988                                    void *opaque, Error **errp)
2989 {
2990     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2991 }
2992 
2993 static void spapr_instance_init(Object *obj)
2994 {
2995     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2996 
2997     spapr->htab_fd = -1;
2998     spapr->use_hotplug_event_source = true;
2999     object_property_add_str(obj, "kvm-type",
3000                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3001     object_property_set_description(obj, "kvm-type",
3002                                     "Specifies the KVM virtualization mode (HV, PR)",
3003                                     NULL);
3004     object_property_add_bool(obj, "modern-hotplug-events",
3005                             spapr_get_modern_hotplug_events,
3006                             spapr_set_modern_hotplug_events,
3007                             NULL);
3008     object_property_set_description(obj, "modern-hotplug-events",
3009                                     "Use dedicated hotplug event mechanism in"
3010                                     " place of standard EPOW events when possible"
3011                                     " (required for memory hot-unplug support)",
3012                                     NULL);
3013     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3014                             "Maximum permitted CPU compatibility mode",
3015                             &error_fatal);
3016 
3017     object_property_add_str(obj, "resize-hpt",
3018                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3019     object_property_set_description(obj, "resize-hpt",
3020                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3021                                     NULL);
3022     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3023                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3024     object_property_set_description(obj, "vsmt",
3025                                     "Virtual SMT: KVM behaves as if this were"
3026                                     " the host's SMT mode", &error_abort);
3027     object_property_add_bool(obj, "vfio-no-msix-emulation",
3028                              spapr_get_msix_emulation, NULL, NULL);
3029 }
3030 
3031 static void spapr_machine_finalizefn(Object *obj)
3032 {
3033     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3034 
3035     g_free(spapr->kvm_type);
3036 }
3037 
3038 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3039 {
3040     cpu_synchronize_state(cs);
3041     ppc_cpu_do_system_reset(cs);
3042 }
3043 
3044 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3045 {
3046     CPUState *cs;
3047 
3048     CPU_FOREACH(cs) {
3049         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3050     }
3051 }
3052 
3053 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3054                            uint32_t node, bool dedicated_hp_event_source,
3055                            Error **errp)
3056 {
3057     sPAPRDRConnector *drc;
3058     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3059     int i, fdt_offset, fdt_size;
3060     void *fdt;
3061     uint64_t addr = addr_start;
3062     bool hotplugged = spapr_drc_hotplugged(dev);
3063     Error *local_err = NULL;
3064 
3065     for (i = 0; i < nr_lmbs; i++) {
3066         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3067                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3068         g_assert(drc);
3069 
3070         fdt = create_device_tree(&fdt_size);
3071         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3072                                                 SPAPR_MEMORY_BLOCK_SIZE);
3073 
3074         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3075         if (local_err) {
3076             while (addr > addr_start) {
3077                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3078                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3079                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3080                 spapr_drc_detach(drc);
3081             }
3082             g_free(fdt);
3083             error_propagate(errp, local_err);
3084             return;
3085         }
3086         if (!hotplugged) {
3087             spapr_drc_reset(drc);
3088         }
3089         addr += SPAPR_MEMORY_BLOCK_SIZE;
3090     }
3091     /* send hotplug notification to the
3092      * guest only in case of hotplugged memory
3093      */
3094     if (hotplugged) {
3095         if (dedicated_hp_event_source) {
3096             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3097                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3098             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3099                                                    nr_lmbs,
3100                                                    spapr_drc_index(drc));
3101         } else {
3102             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3103                                            nr_lmbs);
3104         }
3105     }
3106 }
3107 
3108 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3109                               Error **errp)
3110 {
3111     Error *local_err = NULL;
3112     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3113     PCDIMMDevice *dimm = PC_DIMM(dev);
3114     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3115     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3116     uint64_t align, size, addr;
3117     uint32_t node;
3118 
3119     align = memory_region_get_alignment(mr);
3120     size = memory_region_size(mr);
3121 
3122     pc_dimm_plug(dev, MACHINE(ms), align, &local_err);
3123     if (local_err) {
3124         goto out;
3125     }
3126 
3127     addr = object_property_get_uint(OBJECT(dimm),
3128                                     PC_DIMM_ADDR_PROP, &local_err);
3129     if (local_err) {
3130         goto out_unplug;
3131     }
3132 
3133     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3134                                     &error_abort);
3135     spapr_add_lmbs(dev, addr, size, node,
3136                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3137                    &local_err);
3138     if (local_err) {
3139         goto out_unplug;
3140     }
3141 
3142     return;
3143 
3144 out_unplug:
3145     pc_dimm_unplug(dev, MACHINE(ms));
3146 out:
3147     error_propagate(errp, local_err);
3148 }
3149 
3150 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3151                                   Error **errp)
3152 {
3153     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3154     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3155     PCDIMMDevice *dimm = PC_DIMM(dev);
3156     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3157     MemoryRegion *mr;
3158     uint64_t size;
3159     Object *memdev;
3160     hwaddr pagesize;
3161 
3162     if (!smc->dr_lmb_enabled) {
3163         error_setg(errp, "Memory hotplug not supported for this machine");
3164         return;
3165     }
3166 
3167     mr = ddc->get_memory_region(dimm, errp);
3168     if (!mr) {
3169         return;
3170     }
3171     size = memory_region_size(mr);
3172 
3173     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3174         error_setg(errp, "Hotplugged memory size must be a multiple of "
3175                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3176         return;
3177     }
3178 
3179     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3180                                       &error_abort);
3181     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3182     spapr_check_pagesize(spapr, pagesize, errp);
3183 }
3184 
3185 struct sPAPRDIMMState {
3186     PCDIMMDevice *dimm;
3187     uint32_t nr_lmbs;
3188     QTAILQ_ENTRY(sPAPRDIMMState) next;
3189 };
3190 
3191 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3192                                                        PCDIMMDevice *dimm)
3193 {
3194     sPAPRDIMMState *dimm_state = NULL;
3195 
3196     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3197         if (dimm_state->dimm == dimm) {
3198             break;
3199         }
3200     }
3201     return dimm_state;
3202 }
3203 
3204 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3205                                                       uint32_t nr_lmbs,
3206                                                       PCDIMMDevice *dimm)
3207 {
3208     sPAPRDIMMState *ds = NULL;
3209 
3210     /*
3211      * If this request is for a DIMM whose removal had failed earlier
3212      * (due to guest's refusal to remove the LMBs), we would have this
3213      * dimm already in the pending_dimm_unplugs list. In that
3214      * case don't add again.
3215      */
3216     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3217     if (!ds) {
3218         ds = g_malloc0(sizeof(sPAPRDIMMState));
3219         ds->nr_lmbs = nr_lmbs;
3220         ds->dimm = dimm;
3221         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3222     }
3223     return ds;
3224 }
3225 
3226 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3227                                               sPAPRDIMMState *dimm_state)
3228 {
3229     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3230     g_free(dimm_state);
3231 }
3232 
3233 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3234                                                         PCDIMMDevice *dimm)
3235 {
3236     sPAPRDRConnector *drc;
3237     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3238     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3239     uint64_t size = memory_region_size(mr);
3240     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3241     uint32_t avail_lmbs = 0;
3242     uint64_t addr_start, addr;
3243     int i;
3244 
3245     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3246                                          &error_abort);
3247 
3248     addr = addr_start;
3249     for (i = 0; i < nr_lmbs; i++) {
3250         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3251                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3252         g_assert(drc);
3253         if (drc->dev) {
3254             avail_lmbs++;
3255         }
3256         addr += SPAPR_MEMORY_BLOCK_SIZE;
3257     }
3258 
3259     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3260 }
3261 
3262 /* Callback to be called during DRC release. */
3263 void spapr_lmb_release(DeviceState *dev)
3264 {
3265     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3266     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3267     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3268 
3269     /* This information will get lost if a migration occurs
3270      * during the unplug process. In this case recover it. */
3271     if (ds == NULL) {
3272         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3273         g_assert(ds);
3274         /* The DRC being examined by the caller at least must be counted */
3275         g_assert(ds->nr_lmbs);
3276     }
3277 
3278     if (--ds->nr_lmbs) {
3279         return;
3280     }
3281 
3282     /*
3283      * Now that all the LMBs have been removed by the guest, call the
3284      * unplug handler chain. This can never fail.
3285      */
3286     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3287 }
3288 
3289 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3290 {
3291     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3292     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3293 
3294     pc_dimm_unplug(dev, MACHINE(hotplug_dev));
3295     object_unparent(OBJECT(dev));
3296     spapr_pending_dimm_unplugs_remove(spapr, ds);
3297 }
3298 
3299 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3300                                         DeviceState *dev, Error **errp)
3301 {
3302     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3303     Error *local_err = NULL;
3304     PCDIMMDevice *dimm = PC_DIMM(dev);
3305     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3306     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3307     uint32_t nr_lmbs;
3308     uint64_t size, addr_start, addr;
3309     int i;
3310     sPAPRDRConnector *drc;
3311 
3312     size = memory_region_size(mr);
3313     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3314 
3315     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3316                                          &local_err);
3317     if (local_err) {
3318         goto out;
3319     }
3320 
3321     /*
3322      * An existing pending dimm state for this DIMM means that there is an
3323      * unplug operation in progress, waiting for the spapr_lmb_release
3324      * callback to complete the job (BQL can't cover that far). In this case,
3325      * bail out to avoid detaching DRCs that were already released.
3326      */
3327     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3328         error_setg(&local_err,
3329                    "Memory unplug already in progress for device %s",
3330                    dev->id);
3331         goto out;
3332     }
3333 
3334     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3335 
3336     addr = addr_start;
3337     for (i = 0; i < nr_lmbs; i++) {
3338         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3339                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3340         g_assert(drc);
3341 
3342         spapr_drc_detach(drc);
3343         addr += SPAPR_MEMORY_BLOCK_SIZE;
3344     }
3345 
3346     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3347                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3348     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3349                                               nr_lmbs, spapr_drc_index(drc));
3350 out:
3351     error_propagate(errp, local_err);
3352 }
3353 
3354 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3355                                            sPAPRMachineState *spapr)
3356 {
3357     PowerPCCPU *cpu = POWERPC_CPU(cs);
3358     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3359     int id = spapr_get_vcpu_id(cpu);
3360     void *fdt;
3361     int offset, fdt_size;
3362     char *nodename;
3363 
3364     fdt = create_device_tree(&fdt_size);
3365     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3366     offset = fdt_add_subnode(fdt, 0, nodename);
3367 
3368     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3369     g_free(nodename);
3370 
3371     *fdt_offset = offset;
3372     return fdt;
3373 }
3374 
3375 /* Callback to be called during DRC release. */
3376 void spapr_core_release(DeviceState *dev)
3377 {
3378     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3379 
3380     /* Call the unplug handler chain. This can never fail. */
3381     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3382 }
3383 
3384 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3385 {
3386     MachineState *ms = MACHINE(hotplug_dev);
3387     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3388     CPUCore *cc = CPU_CORE(dev);
3389     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3390 
3391     if (smc->pre_2_10_has_unused_icps) {
3392         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3393         int i;
3394 
3395         for (i = 0; i < cc->nr_threads; i++) {
3396             CPUState *cs = CPU(sc->threads[i]);
3397 
3398             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3399         }
3400     }
3401 
3402     assert(core_slot);
3403     core_slot->cpu = NULL;
3404     object_unparent(OBJECT(dev));
3405 }
3406 
3407 static
3408 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3409                                Error **errp)
3410 {
3411     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3412     int index;
3413     sPAPRDRConnector *drc;
3414     CPUCore *cc = CPU_CORE(dev);
3415 
3416     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3417         error_setg(errp, "Unable to find CPU core with core-id: %d",
3418                    cc->core_id);
3419         return;
3420     }
3421     if (index == 0) {
3422         error_setg(errp, "Boot CPU core may not be unplugged");
3423         return;
3424     }
3425 
3426     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3427                           spapr_vcpu_id(spapr, cc->core_id));
3428     g_assert(drc);
3429 
3430     spapr_drc_detach(drc);
3431 
3432     spapr_hotplug_req_remove_by_index(drc);
3433 }
3434 
3435 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3436                             Error **errp)
3437 {
3438     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3439     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3440     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3441     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3442     CPUCore *cc = CPU_CORE(dev);
3443     CPUState *cs = CPU(core->threads[0]);
3444     sPAPRDRConnector *drc;
3445     Error *local_err = NULL;
3446     CPUArchId *core_slot;
3447     int index;
3448     bool hotplugged = spapr_drc_hotplugged(dev);
3449 
3450     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3451     if (!core_slot) {
3452         error_setg(errp, "Unable to find CPU core with core-id: %d",
3453                    cc->core_id);
3454         return;
3455     }
3456     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3457                           spapr_vcpu_id(spapr, cc->core_id));
3458 
3459     g_assert(drc || !mc->has_hotpluggable_cpus);
3460 
3461     if (drc) {
3462         void *fdt;
3463         int fdt_offset;
3464 
3465         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3466 
3467         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3468         if (local_err) {
3469             g_free(fdt);
3470             error_propagate(errp, local_err);
3471             return;
3472         }
3473 
3474         if (hotplugged) {
3475             /*
3476              * Send hotplug notification interrupt to the guest only
3477              * in case of hotplugged CPUs.
3478              */
3479             spapr_hotplug_req_add_by_index(drc);
3480         } else {
3481             spapr_drc_reset(drc);
3482         }
3483     }
3484 
3485     core_slot->cpu = OBJECT(dev);
3486 
3487     if (smc->pre_2_10_has_unused_icps) {
3488         int i;
3489 
3490         for (i = 0; i < cc->nr_threads; i++) {
3491             cs = CPU(core->threads[i]);
3492             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3493         }
3494     }
3495 }
3496 
3497 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3498                                 Error **errp)
3499 {
3500     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3501     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3502     Error *local_err = NULL;
3503     CPUCore *cc = CPU_CORE(dev);
3504     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3505     const char *type = object_get_typename(OBJECT(dev));
3506     CPUArchId *core_slot;
3507     int index;
3508 
3509     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3510         error_setg(&local_err, "CPU hotplug not supported for this machine");
3511         goto out;
3512     }
3513 
3514     if (strcmp(base_core_type, type)) {
3515         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3516         goto out;
3517     }
3518 
3519     if (cc->core_id % smp_threads) {
3520         error_setg(&local_err, "invalid core id %d", cc->core_id);
3521         goto out;
3522     }
3523 
3524     /*
3525      * In general we should have homogeneous threads-per-core, but old
3526      * (pre hotplug support) machine types allow the last core to have
3527      * reduced threads as a compatibility hack for when we allowed
3528      * total vcpus not a multiple of threads-per-core.
3529      */
3530     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3531         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3532                    cc->nr_threads, smp_threads);
3533         goto out;
3534     }
3535 
3536     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3537     if (!core_slot) {
3538         error_setg(&local_err, "core id %d out of range", cc->core_id);
3539         goto out;
3540     }
3541 
3542     if (core_slot->cpu) {
3543         error_setg(&local_err, "core %d already populated", cc->core_id);
3544         goto out;
3545     }
3546 
3547     numa_cpu_pre_plug(core_slot, dev, &local_err);
3548 
3549 out:
3550     error_propagate(errp, local_err);
3551 }
3552 
3553 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3554                                       DeviceState *dev, Error **errp)
3555 {
3556     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3557         spapr_memory_plug(hotplug_dev, dev, errp);
3558     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3559         spapr_core_plug(hotplug_dev, dev, errp);
3560     }
3561 }
3562 
3563 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3564                                         DeviceState *dev, Error **errp)
3565 {
3566     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3567         spapr_memory_unplug(hotplug_dev, dev);
3568     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3569         spapr_core_unplug(hotplug_dev, dev);
3570     }
3571 }
3572 
3573 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3574                                                 DeviceState *dev, Error **errp)
3575 {
3576     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3577     MachineClass *mc = MACHINE_GET_CLASS(sms);
3578 
3579     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3580         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3581             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3582         } else {
3583             /* NOTE: this means there is a window after guest reset, prior to
3584              * CAS negotiation, where unplug requests will fail due to the
3585              * capability not being detected yet. This is a bit different than
3586              * the case with PCI unplug, where the events will be queued and
3587              * eventually handled by the guest after boot
3588              */
3589             error_setg(errp, "Memory hot unplug not supported for this guest");
3590         }
3591     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3592         if (!mc->has_hotpluggable_cpus) {
3593             error_setg(errp, "CPU hot unplug not supported on this machine");
3594             return;
3595         }
3596         spapr_core_unplug_request(hotplug_dev, dev, errp);
3597     }
3598 }
3599 
3600 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3601                                           DeviceState *dev, Error **errp)
3602 {
3603     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3604         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3605     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3606         spapr_core_pre_plug(hotplug_dev, dev, errp);
3607     }
3608 }
3609 
3610 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3611                                                  DeviceState *dev)
3612 {
3613     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3614         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3615         return HOTPLUG_HANDLER(machine);
3616     }
3617     return NULL;
3618 }
3619 
3620 static CpuInstanceProperties
3621 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3622 {
3623     CPUArchId *core_slot;
3624     MachineClass *mc = MACHINE_GET_CLASS(machine);
3625 
3626     /* make sure possible_cpu are intialized */
3627     mc->possible_cpu_arch_ids(machine);
3628     /* get CPU core slot containing thread that matches cpu_index */
3629     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3630     assert(core_slot);
3631     return core_slot->props;
3632 }
3633 
3634 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3635 {
3636     return idx / smp_cores % nb_numa_nodes;
3637 }
3638 
3639 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3640 {
3641     int i;
3642     const char *core_type;
3643     int spapr_max_cores = max_cpus / smp_threads;
3644     MachineClass *mc = MACHINE_GET_CLASS(machine);
3645 
3646     if (!mc->has_hotpluggable_cpus) {
3647         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3648     }
3649     if (machine->possible_cpus) {
3650         assert(machine->possible_cpus->len == spapr_max_cores);
3651         return machine->possible_cpus;
3652     }
3653 
3654     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3655     if (!core_type) {
3656         error_report("Unable to find sPAPR CPU Core definition");
3657         exit(1);
3658     }
3659 
3660     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3661                              sizeof(CPUArchId) * spapr_max_cores);
3662     machine->possible_cpus->len = spapr_max_cores;
3663     for (i = 0; i < machine->possible_cpus->len; i++) {
3664         int core_id = i * smp_threads;
3665 
3666         machine->possible_cpus->cpus[i].type = core_type;
3667         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3668         machine->possible_cpus->cpus[i].arch_id = core_id;
3669         machine->possible_cpus->cpus[i].props.has_core_id = true;
3670         machine->possible_cpus->cpus[i].props.core_id = core_id;
3671     }
3672     return machine->possible_cpus;
3673 }
3674 
3675 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3676                                 uint64_t *buid, hwaddr *pio,
3677                                 hwaddr *mmio32, hwaddr *mmio64,
3678                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3679 {
3680     /*
3681      * New-style PHB window placement.
3682      *
3683      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3684      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3685      * windows.
3686      *
3687      * Some guest kernels can't work with MMIO windows above 1<<46
3688      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3689      *
3690      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3691      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3692      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3693      * 1TiB 64-bit MMIO windows for each PHB.
3694      */
3695     const uint64_t base_buid = 0x800000020000000ULL;
3696 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3697                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3698     int i;
3699 
3700     /* Sanity check natural alignments */
3701     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3702     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3703     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3704     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3705     /* Sanity check bounds */
3706     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3707                       SPAPR_PCI_MEM32_WIN_SIZE);
3708     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3709                       SPAPR_PCI_MEM64_WIN_SIZE);
3710 
3711     if (index >= SPAPR_MAX_PHBS) {
3712         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3713                    SPAPR_MAX_PHBS - 1);
3714         return;
3715     }
3716 
3717     *buid = base_buid + index;
3718     for (i = 0; i < n_dma; ++i) {
3719         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3720     }
3721 
3722     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3723     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3724     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3725 }
3726 
3727 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3728 {
3729     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3730 
3731     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3732 }
3733 
3734 static void spapr_ics_resend(XICSFabric *dev)
3735 {
3736     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3737 
3738     ics_resend(spapr->ics);
3739 }
3740 
3741 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3742 {
3743     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3744 
3745     return cpu ? ICP(cpu->intc) : NULL;
3746 }
3747 
3748 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3749                                  Monitor *mon)
3750 {
3751     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3752     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3753 
3754     smc->irq->print_info(spapr, mon);
3755 }
3756 
3757 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3758 {
3759     return cpu->vcpu_id;
3760 }
3761 
3762 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3763 {
3764     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3765     int vcpu_id;
3766 
3767     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3768 
3769     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3770         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3771         error_append_hint(errp, "Adjust the number of cpus to %d "
3772                           "or try to raise the number of threads per core\n",
3773                           vcpu_id * smp_threads / spapr->vsmt);
3774         return;
3775     }
3776 
3777     cpu->vcpu_id = vcpu_id;
3778 }
3779 
3780 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3781 {
3782     CPUState *cs;
3783 
3784     CPU_FOREACH(cs) {
3785         PowerPCCPU *cpu = POWERPC_CPU(cs);
3786 
3787         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3788             return cpu;
3789         }
3790     }
3791 
3792     return NULL;
3793 }
3794 
3795 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3796 {
3797     MachineClass *mc = MACHINE_CLASS(oc);
3798     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3799     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3800     NMIClass *nc = NMI_CLASS(oc);
3801     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3802     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3803     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3804     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3805 
3806     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3807     mc->ignore_boot_device_suffixes = true;
3808 
3809     /*
3810      * We set up the default / latest behaviour here.  The class_init
3811      * functions for the specific versioned machine types can override
3812      * these details for backwards compatibility
3813      */
3814     mc->init = spapr_machine_init;
3815     mc->reset = spapr_machine_reset;
3816     mc->block_default_type = IF_SCSI;
3817     mc->max_cpus = 1024;
3818     mc->no_parallel = 1;
3819     mc->default_boot_order = "";
3820     mc->default_ram_size = 512 * MiB;
3821     mc->default_display = "std";
3822     mc->kvm_type = spapr_kvm_type;
3823     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3824     mc->pci_allow_0_address = true;
3825     assert(!mc->get_hotplug_handler);
3826     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3827     hc->pre_plug = spapr_machine_device_pre_plug;
3828     hc->plug = spapr_machine_device_plug;
3829     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3830     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3831     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3832     hc->unplug_request = spapr_machine_device_unplug_request;
3833     hc->unplug = spapr_machine_device_unplug;
3834 
3835     smc->dr_lmb_enabled = true;
3836     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3837     mc->has_hotpluggable_cpus = true;
3838     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3839     fwc->get_dev_path = spapr_get_fw_dev_path;
3840     nc->nmi_monitor_handler = spapr_nmi;
3841     smc->phb_placement = spapr_phb_placement;
3842     vhc->hypercall = emulate_spapr_hypercall;
3843     vhc->hpt_mask = spapr_hpt_mask;
3844     vhc->map_hptes = spapr_map_hptes;
3845     vhc->unmap_hptes = spapr_unmap_hptes;
3846     vhc->store_hpte = spapr_store_hpte;
3847     vhc->get_patbe = spapr_get_patbe;
3848     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3849     xic->ics_get = spapr_ics_get;
3850     xic->ics_resend = spapr_ics_resend;
3851     xic->icp_get = spapr_icp_get;
3852     ispc->print_info = spapr_pic_print_info;
3853     /* Force NUMA node memory size to be a multiple of
3854      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3855      * in which LMBs are represented and hot-added
3856      */
3857     mc->numa_mem_align_shift = 28;
3858 
3859     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3860     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3861     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3862     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3863     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3864     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3865     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
3866     spapr_caps_add_properties(smc, &error_abort);
3867     smc->irq = &spapr_irq_xics;
3868 }
3869 
3870 static const TypeInfo spapr_machine_info = {
3871     .name          = TYPE_SPAPR_MACHINE,
3872     .parent        = TYPE_MACHINE,
3873     .abstract      = true,
3874     .instance_size = sizeof(sPAPRMachineState),
3875     .instance_init = spapr_instance_init,
3876     .instance_finalize = spapr_machine_finalizefn,
3877     .class_size    = sizeof(sPAPRMachineClass),
3878     .class_init    = spapr_machine_class_init,
3879     .interfaces = (InterfaceInfo[]) {
3880         { TYPE_FW_PATH_PROVIDER },
3881         { TYPE_NMI },
3882         { TYPE_HOTPLUG_HANDLER },
3883         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3884         { TYPE_XICS_FABRIC },
3885         { TYPE_INTERRUPT_STATS_PROVIDER },
3886         { }
3887     },
3888 };
3889 
3890 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3891     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3892                                                     void *data)      \
3893     {                                                                \
3894         MachineClass *mc = MACHINE_CLASS(oc);                        \
3895         spapr_machine_##suffix##_class_options(mc);                  \
3896         if (latest) {                                                \
3897             mc->alias = "pseries";                                   \
3898             mc->is_default = 1;                                      \
3899         }                                                            \
3900     }                                                                \
3901     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3902     {                                                                \
3903         MachineState *machine = MACHINE(obj);                        \
3904         spapr_machine_##suffix##_instance_options(machine);          \
3905     }                                                                \
3906     static const TypeInfo spapr_machine_##suffix##_info = {          \
3907         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3908         .parent = TYPE_SPAPR_MACHINE,                                \
3909         .class_init = spapr_machine_##suffix##_class_init,           \
3910         .instance_init = spapr_machine_##suffix##_instance_init,     \
3911     };                                                               \
3912     static void spapr_machine_register_##suffix(void)                \
3913     {                                                                \
3914         type_register(&spapr_machine_##suffix##_info);               \
3915     }                                                                \
3916     type_init(spapr_machine_register_##suffix)
3917 
3918  /*
3919  * pseries-3.1
3920  */
3921 static void spapr_machine_3_1_instance_options(MachineState *machine)
3922 {
3923 }
3924 
3925 static void spapr_machine_3_1_class_options(MachineClass *mc)
3926 {
3927     /* Defaults for the latest behaviour inherited from the base class */
3928 }
3929 
3930 DEFINE_SPAPR_MACHINE(3_1, "3.1", true);
3931 
3932 /*
3933  * pseries-3.0
3934  */
3935 #define SPAPR_COMPAT_3_0                                              \
3936     HW_COMPAT_3_0
3937 
3938 static void spapr_machine_3_0_instance_options(MachineState *machine)
3939 {
3940     spapr_machine_3_1_instance_options(machine);
3941 }
3942 
3943 static void spapr_machine_3_0_class_options(MachineClass *mc)
3944 {
3945     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3946 
3947     spapr_machine_3_1_class_options(mc);
3948     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_3_0);
3949 
3950     smc->legacy_irq_allocation = true;
3951 }
3952 
3953 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
3954 
3955 /*
3956  * pseries-2.12
3957  */
3958 #define SPAPR_COMPAT_2_12                                              \
3959     HW_COMPAT_2_12                                                     \
3960     {                                                                  \
3961         .driver = TYPE_POWERPC_CPU,                                    \
3962         .property = "pre-3.0-migration",                               \
3963         .value    = "on",                                              \
3964     },                                                                 \
3965     {                                                                  \
3966         .driver = TYPE_SPAPR_CPU_CORE,                                 \
3967         .property = "pre-3.0-migration",                               \
3968         .value    = "on",                                              \
3969     },
3970 
3971 static void spapr_machine_2_12_instance_options(MachineState *machine)
3972 {
3973     spapr_machine_3_0_instance_options(machine);
3974 }
3975 
3976 static void spapr_machine_2_12_class_options(MachineClass *mc)
3977 {
3978     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3979 
3980     spapr_machine_3_0_class_options(mc);
3981     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
3982 
3983     /* We depend on kvm_enabled() to choose a default value for the
3984      * hpt-max-page-size capability. Of course we can't do it here
3985      * because this is too early and the HW accelerator isn't initialzed
3986      * yet. Postpone this to machine init (see default_caps_with_cpu()).
3987      */
3988     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
3989 }
3990 
3991 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
3992 
3993 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
3994 {
3995     spapr_machine_2_12_instance_options(machine);
3996 }
3997 
3998 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
3999 {
4000     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4001 
4002     spapr_machine_2_12_class_options(mc);
4003     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4004     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4005     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4006 }
4007 
4008 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4009 
4010 /*
4011  * pseries-2.11
4012  */
4013 #define SPAPR_COMPAT_2_11                                              \
4014     HW_COMPAT_2_11
4015 
4016 static void spapr_machine_2_11_instance_options(MachineState *machine)
4017 {
4018     spapr_machine_2_12_instance_options(machine);
4019 }
4020 
4021 static void spapr_machine_2_11_class_options(MachineClass *mc)
4022 {
4023     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4024 
4025     spapr_machine_2_12_class_options(mc);
4026     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4027     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4028 }
4029 
4030 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4031 
4032 /*
4033  * pseries-2.10
4034  */
4035 #define SPAPR_COMPAT_2_10                                              \
4036     HW_COMPAT_2_10
4037 
4038 static void spapr_machine_2_10_instance_options(MachineState *machine)
4039 {
4040     spapr_machine_2_11_instance_options(machine);
4041 }
4042 
4043 static void spapr_machine_2_10_class_options(MachineClass *mc)
4044 {
4045     spapr_machine_2_11_class_options(mc);
4046     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4047 }
4048 
4049 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4050 
4051 /*
4052  * pseries-2.9
4053  */
4054 #define SPAPR_COMPAT_2_9                                               \
4055     HW_COMPAT_2_9                                                      \
4056     {                                                                  \
4057         .driver = TYPE_POWERPC_CPU,                                    \
4058         .property = "pre-2.10-migration",                              \
4059         .value    = "on",                                              \
4060     },                                                                 \
4061 
4062 static void spapr_machine_2_9_instance_options(MachineState *machine)
4063 {
4064     spapr_machine_2_10_instance_options(machine);
4065 }
4066 
4067 static void spapr_machine_2_9_class_options(MachineClass *mc)
4068 {
4069     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4070 
4071     spapr_machine_2_10_class_options(mc);
4072     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4073     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4074     smc->pre_2_10_has_unused_icps = true;
4075     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4076 }
4077 
4078 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4079 
4080 /*
4081  * pseries-2.8
4082  */
4083 #define SPAPR_COMPAT_2_8                                        \
4084     HW_COMPAT_2_8                                               \
4085     {                                                           \
4086         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4087         .property = "pcie-extended-configuration-space",        \
4088         .value    = "off",                                      \
4089     },
4090 
4091 static void spapr_machine_2_8_instance_options(MachineState *machine)
4092 {
4093     spapr_machine_2_9_instance_options(machine);
4094 }
4095 
4096 static void spapr_machine_2_8_class_options(MachineClass *mc)
4097 {
4098     spapr_machine_2_9_class_options(mc);
4099     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4100     mc->numa_mem_align_shift = 23;
4101 }
4102 
4103 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4104 
4105 /*
4106  * pseries-2.7
4107  */
4108 #define SPAPR_COMPAT_2_7                            \
4109     HW_COMPAT_2_7                                   \
4110     {                                               \
4111         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4112         .property = "mem_win_size",                 \
4113         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4114     },                                              \
4115     {                                               \
4116         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4117         .property = "mem64_win_size",               \
4118         .value    = "0",                            \
4119     },                                              \
4120     {                                               \
4121         .driver = TYPE_POWERPC_CPU,                 \
4122         .property = "pre-2.8-migration",            \
4123         .value    = "on",                           \
4124     },                                              \
4125     {                                               \
4126         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4127         .property = "pre-2.8-migration",            \
4128         .value    = "on",                           \
4129     },
4130 
4131 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4132                               uint64_t *buid, hwaddr *pio,
4133                               hwaddr *mmio32, hwaddr *mmio64,
4134                               unsigned n_dma, uint32_t *liobns, Error **errp)
4135 {
4136     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4137     const uint64_t base_buid = 0x800000020000000ULL;
4138     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4139     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4140     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4141     const uint32_t max_index = 255;
4142     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4143 
4144     uint64_t ram_top = MACHINE(spapr)->ram_size;
4145     hwaddr phb0_base, phb_base;
4146     int i;
4147 
4148     /* Do we have device memory? */
4149     if (MACHINE(spapr)->maxram_size > ram_top) {
4150         /* Can't just use maxram_size, because there may be an
4151          * alignment gap between normal and device memory regions
4152          */
4153         ram_top = MACHINE(spapr)->device_memory->base +
4154             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4155     }
4156 
4157     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4158 
4159     if (index > max_index) {
4160         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4161                    max_index);
4162         return;
4163     }
4164 
4165     *buid = base_buid + index;
4166     for (i = 0; i < n_dma; ++i) {
4167         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4168     }
4169 
4170     phb_base = phb0_base + index * phb_spacing;
4171     *pio = phb_base + pio_offset;
4172     *mmio32 = phb_base + mmio_offset;
4173     /*
4174      * We don't set the 64-bit MMIO window, relying on the PHB's
4175      * fallback behaviour of automatically splitting a large "32-bit"
4176      * window into contiguous 32-bit and 64-bit windows
4177      */
4178 }
4179 
4180 static void spapr_machine_2_7_instance_options(MachineState *machine)
4181 {
4182     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4183 
4184     spapr_machine_2_8_instance_options(machine);
4185     spapr->use_hotplug_event_source = false;
4186 }
4187 
4188 static void spapr_machine_2_7_class_options(MachineClass *mc)
4189 {
4190     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4191 
4192     spapr_machine_2_8_class_options(mc);
4193     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4194     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4195     smc->phb_placement = phb_placement_2_7;
4196 }
4197 
4198 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4199 
4200 /*
4201  * pseries-2.6
4202  */
4203 #define SPAPR_COMPAT_2_6 \
4204     HW_COMPAT_2_6 \
4205     { \
4206         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4207         .property = "ddw",\
4208         .value    = stringify(off),\
4209     },
4210 
4211 static void spapr_machine_2_6_instance_options(MachineState *machine)
4212 {
4213     spapr_machine_2_7_instance_options(machine);
4214 }
4215 
4216 static void spapr_machine_2_6_class_options(MachineClass *mc)
4217 {
4218     spapr_machine_2_7_class_options(mc);
4219     mc->has_hotpluggable_cpus = false;
4220     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4221 }
4222 
4223 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4224 
4225 /*
4226  * pseries-2.5
4227  */
4228 #define SPAPR_COMPAT_2_5 \
4229     HW_COMPAT_2_5 \
4230     { \
4231         .driver   = "spapr-vlan", \
4232         .property = "use-rx-buffer-pools", \
4233         .value    = "off", \
4234     },
4235 
4236 static void spapr_machine_2_5_instance_options(MachineState *machine)
4237 {
4238     spapr_machine_2_6_instance_options(machine);
4239 }
4240 
4241 static void spapr_machine_2_5_class_options(MachineClass *mc)
4242 {
4243     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4244 
4245     spapr_machine_2_6_class_options(mc);
4246     smc->use_ohci_by_default = true;
4247     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4248 }
4249 
4250 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4251 
4252 /*
4253  * pseries-2.4
4254  */
4255 #define SPAPR_COMPAT_2_4 \
4256         HW_COMPAT_2_4
4257 
4258 static void spapr_machine_2_4_instance_options(MachineState *machine)
4259 {
4260     spapr_machine_2_5_instance_options(machine);
4261 }
4262 
4263 static void spapr_machine_2_4_class_options(MachineClass *mc)
4264 {
4265     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4266 
4267     spapr_machine_2_5_class_options(mc);
4268     smc->dr_lmb_enabled = false;
4269     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4270 }
4271 
4272 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4273 
4274 /*
4275  * pseries-2.3
4276  */
4277 #define SPAPR_COMPAT_2_3 \
4278         HW_COMPAT_2_3 \
4279         {\
4280             .driver   = "spapr-pci-host-bridge",\
4281             .property = "dynamic-reconfiguration",\
4282             .value    = "off",\
4283         },
4284 
4285 static void spapr_machine_2_3_instance_options(MachineState *machine)
4286 {
4287     spapr_machine_2_4_instance_options(machine);
4288 }
4289 
4290 static void spapr_machine_2_3_class_options(MachineClass *mc)
4291 {
4292     spapr_machine_2_4_class_options(mc);
4293     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4294 }
4295 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4296 
4297 /*
4298  * pseries-2.2
4299  */
4300 
4301 #define SPAPR_COMPAT_2_2 \
4302         HW_COMPAT_2_2 \
4303         {\
4304             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4305             .property = "mem_win_size",\
4306             .value    = "0x20000000",\
4307         },
4308 
4309 static void spapr_machine_2_2_instance_options(MachineState *machine)
4310 {
4311     spapr_machine_2_3_instance_options(machine);
4312     machine->suppress_vmdesc = true;
4313 }
4314 
4315 static void spapr_machine_2_2_class_options(MachineClass *mc)
4316 {
4317     spapr_machine_2_3_class_options(mc);
4318     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4319 }
4320 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4321 
4322 /*
4323  * pseries-2.1
4324  */
4325 #define SPAPR_COMPAT_2_1 \
4326         HW_COMPAT_2_1
4327 
4328 static void spapr_machine_2_1_instance_options(MachineState *machine)
4329 {
4330     spapr_machine_2_2_instance_options(machine);
4331 }
4332 
4333 static void spapr_machine_2_1_class_options(MachineClass *mc)
4334 {
4335     spapr_machine_2_2_class_options(mc);
4336     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4337 }
4338 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4339 
4340 static void spapr_machine_register_types(void)
4341 {
4342     type_register_static(&spapr_machine_info);
4343 }
4344 
4345 type_init(spapr_machine_register_types)
4346