xref: /openbmc/qemu/hw/ppc/spapr.c (revision 4f4c6206ca6eda478ec0377545ce26eb41090672)
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 **rev;
566     CPUState *cs;
567     int n_cpus;
568     int cpus_offset;
569     char *nodename;
570     int i;
571 
572     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
573     _FDT(cpus_offset);
574     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
575     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
576 
577     /*
578      * We walk the CPUs in reverse order to ensure that CPU DT nodes
579      * created by fdt_add_subnode() end up in the right order in FDT
580      * for the guest kernel the enumerate the CPUs correctly.
581      *
582      * The CPU list cannot be traversed in reverse order, so we need
583      * to do extra work.
584      */
585     n_cpus = 0;
586     rev = NULL;
587     CPU_FOREACH(cs) {
588         rev = g_renew(CPUState *, rev, n_cpus + 1);
589         rev[n_cpus++] = cs;
590     }
591 
592     for (i = n_cpus - 1; i >= 0; i--) {
593         CPUState *cs = rev[i];
594         PowerPCCPU *cpu = POWERPC_CPU(cs);
595         int index = spapr_get_vcpu_id(cpu);
596         DeviceClass *dc = DEVICE_GET_CLASS(cs);
597         int offset;
598 
599         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
600             continue;
601         }
602 
603         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
604         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
605         g_free(nodename);
606         _FDT(offset);
607         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
608     }
609 
610 }
611 
612 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
613 {
614     MemoryDeviceInfoList *info;
615 
616     for (info = list; info; info = info->next) {
617         MemoryDeviceInfo *value = info->value;
618 
619         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
620             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
621 
622             if (addr >= pcdimm_info->addr &&
623                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
624                 return pcdimm_info->node;
625             }
626         }
627     }
628 
629     return -1;
630 }
631 
632 struct sPAPRDrconfCellV2 {
633      uint32_t seq_lmbs;
634      uint64_t base_addr;
635      uint32_t drc_index;
636      uint32_t aa_index;
637      uint32_t flags;
638 } QEMU_PACKED;
639 
640 typedef struct DrconfCellQueue {
641     struct sPAPRDrconfCellV2 cell;
642     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
643 } DrconfCellQueue;
644 
645 static DrconfCellQueue *
646 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
647                       uint32_t drc_index, uint32_t aa_index,
648                       uint32_t flags)
649 {
650     DrconfCellQueue *elem;
651 
652     elem = g_malloc0(sizeof(*elem));
653     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
654     elem->cell.base_addr = cpu_to_be64(base_addr);
655     elem->cell.drc_index = cpu_to_be32(drc_index);
656     elem->cell.aa_index = cpu_to_be32(aa_index);
657     elem->cell.flags = cpu_to_be32(flags);
658 
659     return elem;
660 }
661 
662 /* ibm,dynamic-memory-v2 */
663 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
664                                    int offset, MemoryDeviceInfoList *dimms)
665 {
666     MachineState *machine = MACHINE(spapr);
667     uint8_t *int_buf, *cur_index, buf_len;
668     int ret;
669     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
670     uint64_t addr, cur_addr, size;
671     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
672     uint64_t mem_end = machine->device_memory->base +
673                        memory_region_size(&machine->device_memory->mr);
674     uint32_t node, nr_entries = 0;
675     sPAPRDRConnector *drc;
676     DrconfCellQueue *elem, *next;
677     MemoryDeviceInfoList *info;
678     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
679         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
680 
681     /* Entry to cover RAM and the gap area */
682     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
683                                  SPAPR_LMB_FLAGS_RESERVED |
684                                  SPAPR_LMB_FLAGS_DRC_INVALID);
685     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
686     nr_entries++;
687 
688     cur_addr = machine->device_memory->base;
689     for (info = dimms; info; info = info->next) {
690         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
691 
692         addr = di->addr;
693         size = di->size;
694         node = di->node;
695 
696         /* Entry for hot-pluggable area */
697         if (cur_addr < addr) {
698             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
699             g_assert(drc);
700             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
701                                          cur_addr, spapr_drc_index(drc), -1, 0);
702             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
703             nr_entries++;
704         }
705 
706         /* Entry for DIMM */
707         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
708         g_assert(drc);
709         elem = spapr_get_drconf_cell(size / lmb_size, addr,
710                                      spapr_drc_index(drc), node,
711                                      SPAPR_LMB_FLAGS_ASSIGNED);
712         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
713         nr_entries++;
714         cur_addr = addr + size;
715     }
716 
717     /* Entry for remaining hotpluggable area */
718     if (cur_addr < mem_end) {
719         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
720         g_assert(drc);
721         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
722                                      cur_addr, spapr_drc_index(drc), -1, 0);
723         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
724         nr_entries++;
725     }
726 
727     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
728     int_buf = cur_index = g_malloc0(buf_len);
729     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
730     cur_index += sizeof(nr_entries);
731 
732     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
733         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
734         cur_index += sizeof(elem->cell);
735         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
736         g_free(elem);
737     }
738 
739     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
740     g_free(int_buf);
741     if (ret < 0) {
742         return -1;
743     }
744     return 0;
745 }
746 
747 /* ibm,dynamic-memory */
748 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
749                                    int offset, MemoryDeviceInfoList *dimms)
750 {
751     MachineState *machine = MACHINE(spapr);
752     int i, ret;
753     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
754     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
755     uint32_t nr_lmbs = (machine->device_memory->base +
756                        memory_region_size(&machine->device_memory->mr)) /
757                        lmb_size;
758     uint32_t *int_buf, *cur_index, buf_len;
759 
760     /*
761      * Allocate enough buffer size to fit in ibm,dynamic-memory
762      */
763     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
764     cur_index = int_buf = g_malloc0(buf_len);
765     int_buf[0] = cpu_to_be32(nr_lmbs);
766     cur_index++;
767     for (i = 0; i < nr_lmbs; i++) {
768         uint64_t addr = i * lmb_size;
769         uint32_t *dynamic_memory = cur_index;
770 
771         if (i >= device_lmb_start) {
772             sPAPRDRConnector *drc;
773 
774             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
775             g_assert(drc);
776 
777             dynamic_memory[0] = cpu_to_be32(addr >> 32);
778             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
779             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
780             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
781             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
782             if (memory_region_present(get_system_memory(), addr)) {
783                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
784             } else {
785                 dynamic_memory[5] = cpu_to_be32(0);
786             }
787         } else {
788             /*
789              * LMB information for RMA, boot time RAM and gap b/n RAM and
790              * device memory region -- all these are marked as reserved
791              * and as having no valid DRC.
792              */
793             dynamic_memory[0] = cpu_to_be32(addr >> 32);
794             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
795             dynamic_memory[2] = cpu_to_be32(0);
796             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
797             dynamic_memory[4] = cpu_to_be32(-1);
798             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
799                                             SPAPR_LMB_FLAGS_DRC_INVALID);
800         }
801 
802         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
803     }
804     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
805     g_free(int_buf);
806     if (ret < 0) {
807         return -1;
808     }
809     return 0;
810 }
811 
812 /*
813  * Adds ibm,dynamic-reconfiguration-memory node.
814  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
815  * of this device tree node.
816  */
817 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
818 {
819     MachineState *machine = MACHINE(spapr);
820     int ret, i, offset;
821     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
822     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
823     uint32_t *int_buf, *cur_index, buf_len;
824     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
825     MemoryDeviceInfoList *dimms = NULL;
826 
827     /*
828      * Don't create the node if there is no device memory
829      */
830     if (machine->ram_size == machine->maxram_size) {
831         return 0;
832     }
833 
834     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
835 
836     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
837                     sizeof(prop_lmb_size));
838     if (ret < 0) {
839         return ret;
840     }
841 
842     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
843     if (ret < 0) {
844         return ret;
845     }
846 
847     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
848     if (ret < 0) {
849         return ret;
850     }
851 
852     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
853     dimms = qmp_memory_device_list();
854     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
855         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
856     } else {
857         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
858     }
859     qapi_free_MemoryDeviceInfoList(dimms);
860 
861     if (ret < 0) {
862         return ret;
863     }
864 
865     /* ibm,associativity-lookup-arrays */
866     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
867     cur_index = int_buf = g_malloc0(buf_len);
868 
869     cur_index = int_buf;
870     int_buf[0] = cpu_to_be32(nr_nodes);
871     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
872     cur_index += 2;
873     for (i = 0; i < nr_nodes; i++) {
874         uint32_t associativity[] = {
875             cpu_to_be32(0x0),
876             cpu_to_be32(0x0),
877             cpu_to_be32(0x0),
878             cpu_to_be32(i)
879         };
880         memcpy(cur_index, associativity, sizeof(associativity));
881         cur_index += 4;
882     }
883     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
884             (cur_index - int_buf) * sizeof(uint32_t));
885     g_free(int_buf);
886 
887     return ret;
888 }
889 
890 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
891                                 sPAPROptionVector *ov5_updates)
892 {
893     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
894     int ret = 0, offset;
895 
896     /* Generate ibm,dynamic-reconfiguration-memory node if required */
897     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
898         g_assert(smc->dr_lmb_enabled);
899         ret = spapr_populate_drconf_memory(spapr, fdt);
900         if (ret) {
901             goto out;
902         }
903     }
904 
905     offset = fdt_path_offset(fdt, "/chosen");
906     if (offset < 0) {
907         offset = fdt_add_subnode(fdt, 0, "chosen");
908         if (offset < 0) {
909             return offset;
910         }
911     }
912     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
913                                  "ibm,architecture-vec-5");
914 
915 out:
916     return ret;
917 }
918 
919 static bool spapr_hotplugged_dev_before_cas(void)
920 {
921     Object *drc_container, *obj;
922     ObjectProperty *prop;
923     ObjectPropertyIterator iter;
924 
925     drc_container = container_get(object_get_root(), "/dr-connector");
926     object_property_iter_init(&iter, drc_container);
927     while ((prop = object_property_iter_next(&iter))) {
928         if (!strstart(prop->type, "link<", NULL)) {
929             continue;
930         }
931         obj = object_property_get_link(drc_container, prop->name, NULL);
932         if (spapr_drc_needed(obj)) {
933             return true;
934         }
935     }
936     return false;
937 }
938 
939 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
940                                  target_ulong addr, target_ulong size,
941                                  sPAPROptionVector *ov5_updates)
942 {
943     void *fdt, *fdt_skel;
944     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
945 
946     if (spapr_hotplugged_dev_before_cas()) {
947         return 1;
948     }
949 
950     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
951         error_report("SLOF provided an unexpected CAS buffer size "
952                      TARGET_FMT_lu " (min: %zu, max: %u)",
953                      size, sizeof(hdr), FW_MAX_SIZE);
954         exit(EXIT_FAILURE);
955     }
956 
957     size -= sizeof(hdr);
958 
959     /* Create skeleton */
960     fdt_skel = g_malloc0(size);
961     _FDT((fdt_create(fdt_skel, size)));
962     _FDT((fdt_finish_reservemap(fdt_skel)));
963     _FDT((fdt_begin_node(fdt_skel, "")));
964     _FDT((fdt_end_node(fdt_skel)));
965     _FDT((fdt_finish(fdt_skel)));
966     fdt = g_malloc0(size);
967     _FDT((fdt_open_into(fdt_skel, fdt, size)));
968     g_free(fdt_skel);
969 
970     /* Fixup cpu nodes */
971     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
972 
973     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
974         return -1;
975     }
976 
977     /* Pack resulting tree */
978     _FDT((fdt_pack(fdt)));
979 
980     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
981         trace_spapr_cas_failed(size);
982         return -1;
983     }
984 
985     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
986     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
987     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
988     g_free(fdt);
989 
990     return 0;
991 }
992 
993 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
994 {
995     int rtas;
996     GString *hypertas = g_string_sized_new(256);
997     GString *qemu_hypertas = g_string_sized_new(256);
998     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
999     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1000         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1001     uint32_t lrdr_capacity[] = {
1002         cpu_to_be32(max_device_addr >> 32),
1003         cpu_to_be32(max_device_addr & 0xffffffff),
1004         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1005         cpu_to_be32(max_cpus / smp_threads),
1006     };
1007     uint32_t maxdomains[] = {
1008         cpu_to_be32(4),
1009         cpu_to_be32(0),
1010         cpu_to_be32(0),
1011         cpu_to_be32(0),
1012         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1013     };
1014 
1015     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1016 
1017     /* hypertas */
1018     add_str(hypertas, "hcall-pft");
1019     add_str(hypertas, "hcall-term");
1020     add_str(hypertas, "hcall-dabr");
1021     add_str(hypertas, "hcall-interrupt");
1022     add_str(hypertas, "hcall-tce");
1023     add_str(hypertas, "hcall-vio");
1024     add_str(hypertas, "hcall-splpar");
1025     add_str(hypertas, "hcall-bulk");
1026     add_str(hypertas, "hcall-set-mode");
1027     add_str(hypertas, "hcall-sprg0");
1028     add_str(hypertas, "hcall-copy");
1029     add_str(hypertas, "hcall-debug");
1030     add_str(qemu_hypertas, "hcall-memop1");
1031 
1032     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1033         add_str(hypertas, "hcall-multi-tce");
1034     }
1035 
1036     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1037         add_str(hypertas, "hcall-hpt-resize");
1038     }
1039 
1040     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1041                      hypertas->str, hypertas->len));
1042     g_string_free(hypertas, TRUE);
1043     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1044                      qemu_hypertas->str, qemu_hypertas->len));
1045     g_string_free(qemu_hypertas, TRUE);
1046 
1047     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1048                      refpoints, sizeof(refpoints)));
1049 
1050     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1051                      maxdomains, sizeof(maxdomains)));
1052 
1053     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1054                           RTAS_ERROR_LOG_MAX));
1055     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1056                           RTAS_EVENT_SCAN_RATE));
1057 
1058     g_assert(msi_nonbroken);
1059     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1060 
1061     /*
1062      * According to PAPR, rtas ibm,os-term does not guarantee a return
1063      * back to the guest cpu.
1064      *
1065      * While an additional ibm,extended-os-term property indicates
1066      * that rtas call return will always occur. Set this property.
1067      */
1068     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1069 
1070     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1071                      lrdr_capacity, sizeof(lrdr_capacity)));
1072 
1073     spapr_dt_rtas_tokens(fdt, rtas);
1074 }
1075 
1076 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1077  * that the guest may request and thus the valid values for bytes 24..26 of
1078  * option vector 5: */
1079 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1080 {
1081     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1082 
1083     char val[2 * 4] = {
1084         23, 0x00, /* Xive mode, filled in below. */
1085         24, 0x00, /* Hash/Radix, filled in below. */
1086         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1087         26, 0x40, /* Radix options: GTSE == yes. */
1088     };
1089 
1090     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1091                           first_ppc_cpu->compat_pvr)) {
1092         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1093         val[3] = 0x00; /* Hash */
1094     } else if (kvm_enabled()) {
1095         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1096             val[3] = 0x80; /* OV5_MMU_BOTH */
1097         } else if (kvmppc_has_cap_mmu_radix()) {
1098             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1099         } else {
1100             val[3] = 0x00; /* Hash */
1101         }
1102     } else {
1103         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1104         val[3] = 0xC0;
1105     }
1106     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1107                      val, sizeof(val)));
1108 }
1109 
1110 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1111 {
1112     MachineState *machine = MACHINE(spapr);
1113     int chosen;
1114     const char *boot_device = machine->boot_order;
1115     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1116     size_t cb = 0;
1117     char *bootlist = get_boot_devices_list(&cb);
1118 
1119     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1120 
1121     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1122     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1123                           spapr->initrd_base));
1124     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1125                           spapr->initrd_base + spapr->initrd_size));
1126 
1127     if (spapr->kernel_size) {
1128         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1129                               cpu_to_be64(spapr->kernel_size) };
1130 
1131         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1132                          &kprop, sizeof(kprop)));
1133         if (spapr->kernel_le) {
1134             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1135         }
1136     }
1137     if (boot_menu) {
1138         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1139     }
1140     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1141     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1142     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1143 
1144     if (cb && bootlist) {
1145         int i;
1146 
1147         for (i = 0; i < cb; i++) {
1148             if (bootlist[i] == '\n') {
1149                 bootlist[i] = ' ';
1150             }
1151         }
1152         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1153     }
1154 
1155     if (boot_device && strlen(boot_device)) {
1156         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1157     }
1158 
1159     if (!spapr->has_graphics && stdout_path) {
1160         /*
1161          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1162          * kernel. New platforms should only use the "stdout-path" property. Set
1163          * the new property and continue using older property to remain
1164          * compatible with the existing firmware.
1165          */
1166         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1167         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1168     }
1169 
1170     spapr_dt_ov5_platform_support(fdt, chosen);
1171 
1172     g_free(stdout_path);
1173     g_free(bootlist);
1174 }
1175 
1176 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1177 {
1178     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1179      * KVM to work under pHyp with some guest co-operation */
1180     int hypervisor;
1181     uint8_t hypercall[16];
1182 
1183     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1184     /* indicate KVM hypercall interface */
1185     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1186     if (kvmppc_has_cap_fixup_hcalls()) {
1187         /*
1188          * Older KVM versions with older guest kernels were broken
1189          * with the magic page, don't allow the guest to map it.
1190          */
1191         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1192                                   sizeof(hypercall))) {
1193             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1194                              hypercall, sizeof(hypercall)));
1195         }
1196     }
1197 }
1198 
1199 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1200                              hwaddr rtas_addr,
1201                              hwaddr rtas_size)
1202 {
1203     MachineState *machine = MACHINE(spapr);
1204     MachineClass *mc = MACHINE_GET_CLASS(machine);
1205     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1206     int ret;
1207     void *fdt;
1208     sPAPRPHBState *phb;
1209     char *buf;
1210 
1211     fdt = g_malloc0(FDT_MAX_SIZE);
1212     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1213 
1214     /* Root node */
1215     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1216     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1217     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1218 
1219     /*
1220      * Add info to guest to indentify which host is it being run on
1221      * and what is the uuid of the guest
1222      */
1223     if (kvmppc_get_host_model(&buf)) {
1224         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1225         g_free(buf);
1226     }
1227     if (kvmppc_get_host_serial(&buf)) {
1228         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1229         g_free(buf);
1230     }
1231 
1232     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1233 
1234     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1235     if (qemu_uuid_set) {
1236         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1237     }
1238     g_free(buf);
1239 
1240     if (qemu_get_vm_name()) {
1241         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1242                                 qemu_get_vm_name()));
1243     }
1244 
1245     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1246     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1247 
1248     /* /interrupt controller */
1249     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1250 
1251     ret = spapr_populate_memory(spapr, fdt);
1252     if (ret < 0) {
1253         error_report("couldn't setup memory nodes in fdt");
1254         exit(1);
1255     }
1256 
1257     /* /vdevice */
1258     spapr_dt_vdevice(spapr->vio_bus, fdt);
1259 
1260     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1261         ret = spapr_rng_populate_dt(fdt);
1262         if (ret < 0) {
1263             error_report("could not set up rng device in the fdt");
1264             exit(1);
1265         }
1266     }
1267 
1268     QLIST_FOREACH(phb, &spapr->phbs, list) {
1269         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1270         if (ret < 0) {
1271             error_report("couldn't setup PCI devices in fdt");
1272             exit(1);
1273         }
1274     }
1275 
1276     /* cpus */
1277     spapr_populate_cpus_dt_node(fdt, spapr);
1278 
1279     if (smc->dr_lmb_enabled) {
1280         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1281     }
1282 
1283     if (mc->has_hotpluggable_cpus) {
1284         int offset = fdt_path_offset(fdt, "/cpus");
1285         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1286                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1287         if (ret < 0) {
1288             error_report("Couldn't set up CPU DR device tree properties");
1289             exit(1);
1290         }
1291     }
1292 
1293     /* /event-sources */
1294     spapr_dt_events(spapr, fdt);
1295 
1296     /* /rtas */
1297     spapr_dt_rtas(spapr, fdt);
1298 
1299     /* /chosen */
1300     spapr_dt_chosen(spapr, fdt);
1301 
1302     /* /hypervisor */
1303     if (kvm_enabled()) {
1304         spapr_dt_hypervisor(spapr, fdt);
1305     }
1306 
1307     /* Build memory reserve map */
1308     if (spapr->kernel_size) {
1309         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1310     }
1311     if (spapr->initrd_size) {
1312         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1313     }
1314 
1315     /* ibm,client-architecture-support updates */
1316     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1317     if (ret < 0) {
1318         error_report("couldn't setup CAS properties fdt");
1319         exit(1);
1320     }
1321 
1322     return fdt;
1323 }
1324 
1325 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1326 {
1327     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1328 }
1329 
1330 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1331                                     PowerPCCPU *cpu)
1332 {
1333     CPUPPCState *env = &cpu->env;
1334 
1335     /* The TCG path should also be holding the BQL at this point */
1336     g_assert(qemu_mutex_iothread_locked());
1337 
1338     if (msr_pr) {
1339         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1340         env->gpr[3] = H_PRIVILEGE;
1341     } else {
1342         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1343     }
1344 }
1345 
1346 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1347 {
1348     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1349 
1350     return spapr->patb_entry;
1351 }
1352 
1353 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1354 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1355 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1356 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1357 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1358 
1359 /*
1360  * Get the fd to access the kernel htab, re-opening it if necessary
1361  */
1362 static int get_htab_fd(sPAPRMachineState *spapr)
1363 {
1364     Error *local_err = NULL;
1365 
1366     if (spapr->htab_fd >= 0) {
1367         return spapr->htab_fd;
1368     }
1369 
1370     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1371     if (spapr->htab_fd < 0) {
1372         error_report_err(local_err);
1373     }
1374 
1375     return spapr->htab_fd;
1376 }
1377 
1378 void close_htab_fd(sPAPRMachineState *spapr)
1379 {
1380     if (spapr->htab_fd >= 0) {
1381         close(spapr->htab_fd);
1382     }
1383     spapr->htab_fd = -1;
1384 }
1385 
1386 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1387 {
1388     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1389 
1390     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1391 }
1392 
1393 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1394 {
1395     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1396 
1397     assert(kvm_enabled());
1398 
1399     if (!spapr->htab) {
1400         return 0;
1401     }
1402 
1403     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1404 }
1405 
1406 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1407                                                 hwaddr ptex, int n)
1408 {
1409     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1410     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1411 
1412     if (!spapr->htab) {
1413         /*
1414          * HTAB is controlled by KVM. Fetch into temporary buffer
1415          */
1416         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1417         kvmppc_read_hptes(hptes, ptex, n);
1418         return hptes;
1419     }
1420 
1421     /*
1422      * HTAB is controlled by QEMU. Just point to the internally
1423      * accessible PTEG.
1424      */
1425     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1426 }
1427 
1428 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1429                               const ppc_hash_pte64_t *hptes,
1430                               hwaddr ptex, int n)
1431 {
1432     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1433 
1434     if (!spapr->htab) {
1435         g_free((void *)hptes);
1436     }
1437 
1438     /* Nothing to do for qemu managed HPT */
1439 }
1440 
1441 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1442                              uint64_t pte0, uint64_t pte1)
1443 {
1444     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1445     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1446 
1447     if (!spapr->htab) {
1448         kvmppc_write_hpte(ptex, pte0, pte1);
1449     } else {
1450         stq_p(spapr->htab + offset, pte0);
1451         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1452     }
1453 }
1454 
1455 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1456 {
1457     int shift;
1458 
1459     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1460      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1461      * that's much more than is needed for Linux guests */
1462     shift = ctz64(pow2ceil(ramsize)) - 7;
1463     shift = MAX(shift, 18); /* Minimum architected size */
1464     shift = MIN(shift, 46); /* Maximum architected size */
1465     return shift;
1466 }
1467 
1468 void spapr_free_hpt(sPAPRMachineState *spapr)
1469 {
1470     g_free(spapr->htab);
1471     spapr->htab = NULL;
1472     spapr->htab_shift = 0;
1473     close_htab_fd(spapr);
1474 }
1475 
1476 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1477                           Error **errp)
1478 {
1479     long rc;
1480 
1481     /* Clean up any HPT info from a previous boot */
1482     spapr_free_hpt(spapr);
1483 
1484     rc = kvmppc_reset_htab(shift);
1485     if (rc < 0) {
1486         /* kernel-side HPT needed, but couldn't allocate one */
1487         error_setg_errno(errp, errno,
1488                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1489                          shift);
1490         /* This is almost certainly fatal, but if the caller really
1491          * wants to carry on with shift == 0, it's welcome to try */
1492     } else if (rc > 0) {
1493         /* kernel-side HPT allocated */
1494         if (rc != shift) {
1495             error_setg(errp,
1496                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1497                        shift, rc);
1498         }
1499 
1500         spapr->htab_shift = shift;
1501         spapr->htab = NULL;
1502     } else {
1503         /* kernel-side HPT not needed, allocate in userspace instead */
1504         size_t size = 1ULL << shift;
1505         int i;
1506 
1507         spapr->htab = qemu_memalign(size, size);
1508         if (!spapr->htab) {
1509             error_setg_errno(errp, errno,
1510                              "Could not allocate HPT of order %d", shift);
1511             return;
1512         }
1513 
1514         memset(spapr->htab, 0, size);
1515         spapr->htab_shift = shift;
1516 
1517         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1518             DIRTY_HPTE(HPTE(spapr->htab, i));
1519         }
1520     }
1521     /* We're setting up a hash table, so that means we're not radix */
1522     spapr->patb_entry = 0;
1523 }
1524 
1525 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1526 {
1527     int hpt_shift;
1528 
1529     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1530         || (spapr->cas_reboot
1531             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1532         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1533     } else {
1534         uint64_t current_ram_size;
1535 
1536         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1537         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1538     }
1539     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1540 
1541     if (spapr->vrma_adjust) {
1542         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1543                                           spapr->htab_shift);
1544     }
1545 }
1546 
1547 static int spapr_reset_drcs(Object *child, void *opaque)
1548 {
1549     sPAPRDRConnector *drc =
1550         (sPAPRDRConnector *) object_dynamic_cast(child,
1551                                                  TYPE_SPAPR_DR_CONNECTOR);
1552 
1553     if (drc) {
1554         spapr_drc_reset(drc);
1555     }
1556 
1557     return 0;
1558 }
1559 
1560 static void spapr_machine_reset(void)
1561 {
1562     MachineState *machine = MACHINE(qdev_get_machine());
1563     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1564     PowerPCCPU *first_ppc_cpu;
1565     uint32_t rtas_limit;
1566     hwaddr rtas_addr, fdt_addr;
1567     void *fdt;
1568     int rc;
1569 
1570     spapr_caps_apply(spapr);
1571 
1572     first_ppc_cpu = POWERPC_CPU(first_cpu);
1573     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1574         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1575                               spapr->max_compat_pvr)) {
1576         /* If using KVM with radix mode available, VCPUs can be started
1577          * without a HPT because KVM will start them in radix mode.
1578          * Set the GR bit in PATB so that we know there is no HPT. */
1579         spapr->patb_entry = PATBE1_GR;
1580     } else {
1581         spapr_setup_hpt_and_vrma(spapr);
1582     }
1583 
1584     /* if this reset wasn't generated by CAS, we should reset our
1585      * negotiated options and start from scratch */
1586     if (!spapr->cas_reboot) {
1587         spapr_ovec_cleanup(spapr->ov5_cas);
1588         spapr->ov5_cas = spapr_ovec_new();
1589 
1590         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1591     }
1592 
1593     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1594         spapr_irq_msi_reset(spapr);
1595     }
1596 
1597     qemu_devices_reset();
1598 
1599     /* DRC reset may cause a device to be unplugged. This will cause troubles
1600      * if this device is used by another device (eg, a running vhost backend
1601      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1602      * situations, we reset DRCs after all devices have been reset.
1603      */
1604     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1605 
1606     spapr_clear_pending_events(spapr);
1607 
1608     /*
1609      * We place the device tree and RTAS just below either the top of the RMA,
1610      * or just below 2GB, whichever is lowere, so that it can be
1611      * processed with 32-bit real mode code if necessary
1612      */
1613     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1614     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1615     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1616 
1617     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1618 
1619     spapr_load_rtas(spapr, fdt, rtas_addr);
1620 
1621     rc = fdt_pack(fdt);
1622 
1623     /* Should only fail if we've built a corrupted tree */
1624     assert(rc == 0);
1625 
1626     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1627         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1628                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1629         exit(1);
1630     }
1631 
1632     /* Load the fdt */
1633     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1634     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1635     g_free(fdt);
1636 
1637     /* Set up the entry state */
1638     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1639     first_ppc_cpu->env.gpr[5] = 0;
1640 
1641     spapr->cas_reboot = false;
1642 }
1643 
1644 static void spapr_create_nvram(sPAPRMachineState *spapr)
1645 {
1646     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1647     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1648 
1649     if (dinfo) {
1650         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1651                             &error_fatal);
1652     }
1653 
1654     qdev_init_nofail(dev);
1655 
1656     spapr->nvram = (struct sPAPRNVRAM *)dev;
1657 }
1658 
1659 static void spapr_rtc_create(sPAPRMachineState *spapr)
1660 {
1661     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1662     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1663                               &error_fatal);
1664     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1665                               &error_fatal);
1666     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1667                               "date", &error_fatal);
1668 }
1669 
1670 /* Returns whether we want to use VGA or not */
1671 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1672 {
1673     switch (vga_interface_type) {
1674     case VGA_NONE:
1675         return false;
1676     case VGA_DEVICE:
1677         return true;
1678     case VGA_STD:
1679     case VGA_VIRTIO:
1680         return pci_vga_init(pci_bus) != NULL;
1681     default:
1682         error_setg(errp,
1683                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1684         return false;
1685     }
1686 }
1687 
1688 static int spapr_pre_load(void *opaque)
1689 {
1690     int rc;
1691 
1692     rc = spapr_caps_pre_load(opaque);
1693     if (rc) {
1694         return rc;
1695     }
1696 
1697     return 0;
1698 }
1699 
1700 static int spapr_post_load(void *opaque, int version_id)
1701 {
1702     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1703     int err = 0;
1704 
1705     err = spapr_caps_post_migration(spapr);
1706     if (err) {
1707         return err;
1708     }
1709 
1710     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1711         CPUState *cs;
1712         CPU_FOREACH(cs) {
1713             PowerPCCPU *cpu = POWERPC_CPU(cs);
1714             icp_resend(ICP(cpu->intc));
1715         }
1716     }
1717 
1718     /* In earlier versions, there was no separate qdev for the PAPR
1719      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1720      * So when migrating from those versions, poke the incoming offset
1721      * value into the RTC device */
1722     if (version_id < 3) {
1723         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1724     }
1725 
1726     if (kvm_enabled() && spapr->patb_entry) {
1727         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1728         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1729         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1730 
1731         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1732         if (err) {
1733             error_report("Process table config unsupported by the host");
1734             return -EINVAL;
1735         }
1736     }
1737 
1738     return err;
1739 }
1740 
1741 static int spapr_pre_save(void *opaque)
1742 {
1743     int rc;
1744 
1745     rc = spapr_caps_pre_save(opaque);
1746     if (rc) {
1747         return rc;
1748     }
1749 
1750     return 0;
1751 }
1752 
1753 static bool version_before_3(void *opaque, int version_id)
1754 {
1755     return version_id < 3;
1756 }
1757 
1758 static bool spapr_pending_events_needed(void *opaque)
1759 {
1760     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1761     return !QTAILQ_EMPTY(&spapr->pending_events);
1762 }
1763 
1764 static const VMStateDescription vmstate_spapr_event_entry = {
1765     .name = "spapr_event_log_entry",
1766     .version_id = 1,
1767     .minimum_version_id = 1,
1768     .fields = (VMStateField[]) {
1769         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1770         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1771         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1772                                      NULL, extended_length),
1773         VMSTATE_END_OF_LIST()
1774     },
1775 };
1776 
1777 static const VMStateDescription vmstate_spapr_pending_events = {
1778     .name = "spapr_pending_events",
1779     .version_id = 1,
1780     .minimum_version_id = 1,
1781     .needed = spapr_pending_events_needed,
1782     .fields = (VMStateField[]) {
1783         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1784                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1785         VMSTATE_END_OF_LIST()
1786     },
1787 };
1788 
1789 static bool spapr_ov5_cas_needed(void *opaque)
1790 {
1791     sPAPRMachineState *spapr = opaque;
1792     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1793     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1794     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1795     bool cas_needed;
1796 
1797     /* Prior to the introduction of sPAPROptionVector, we had two option
1798      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1799      * Both of these options encode machine topology into the device-tree
1800      * in such a way that the now-booted OS should still be able to interact
1801      * appropriately with QEMU regardless of what options were actually
1802      * negotiatied on the source side.
1803      *
1804      * As such, we can avoid migrating the CAS-negotiated options if these
1805      * are the only options available on the current machine/platform.
1806      * Since these are the only options available for pseries-2.7 and
1807      * earlier, this allows us to maintain old->new/new->old migration
1808      * compatibility.
1809      *
1810      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1811      * via default pseries-2.8 machines and explicit command-line parameters.
1812      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1813      * of the actual CAS-negotiated values to continue working properly. For
1814      * example, availability of memory unplug depends on knowing whether
1815      * OV5_HP_EVT was negotiated via CAS.
1816      *
1817      * Thus, for any cases where the set of available CAS-negotiatable
1818      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1819      * include the CAS-negotiated options in the migration stream, unless
1820      * if they affect boot time behaviour only.
1821      */
1822     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1823     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1824     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1825 
1826     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1827      * the mask itself since in the future it's possible "legacy" bits may be
1828      * removed via machine options, which could generate a false positive
1829      * that breaks migration.
1830      */
1831     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1832     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1833 
1834     spapr_ovec_cleanup(ov5_mask);
1835     spapr_ovec_cleanup(ov5_legacy);
1836     spapr_ovec_cleanup(ov5_removed);
1837 
1838     return cas_needed;
1839 }
1840 
1841 static const VMStateDescription vmstate_spapr_ov5_cas = {
1842     .name = "spapr_option_vector_ov5_cas",
1843     .version_id = 1,
1844     .minimum_version_id = 1,
1845     .needed = spapr_ov5_cas_needed,
1846     .fields = (VMStateField[]) {
1847         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1848                                  vmstate_spapr_ovec, sPAPROptionVector),
1849         VMSTATE_END_OF_LIST()
1850     },
1851 };
1852 
1853 static bool spapr_patb_entry_needed(void *opaque)
1854 {
1855     sPAPRMachineState *spapr = opaque;
1856 
1857     return !!spapr->patb_entry;
1858 }
1859 
1860 static const VMStateDescription vmstate_spapr_patb_entry = {
1861     .name = "spapr_patb_entry",
1862     .version_id = 1,
1863     .minimum_version_id = 1,
1864     .needed = spapr_patb_entry_needed,
1865     .fields = (VMStateField[]) {
1866         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1867         VMSTATE_END_OF_LIST()
1868     },
1869 };
1870 
1871 static bool spapr_irq_map_needed(void *opaque)
1872 {
1873     sPAPRMachineState *spapr = opaque;
1874 
1875     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1876 }
1877 
1878 static const VMStateDescription vmstate_spapr_irq_map = {
1879     .name = "spapr_irq_map",
1880     .version_id = 1,
1881     .minimum_version_id = 1,
1882     .needed = spapr_irq_map_needed,
1883     .fields = (VMStateField[]) {
1884         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1885         VMSTATE_END_OF_LIST()
1886     },
1887 };
1888 
1889 static const VMStateDescription vmstate_spapr = {
1890     .name = "spapr",
1891     .version_id = 3,
1892     .minimum_version_id = 1,
1893     .pre_load = spapr_pre_load,
1894     .post_load = spapr_post_load,
1895     .pre_save = spapr_pre_save,
1896     .fields = (VMStateField[]) {
1897         /* used to be @next_irq */
1898         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1899 
1900         /* RTC offset */
1901         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1902 
1903         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1904         VMSTATE_END_OF_LIST()
1905     },
1906     .subsections = (const VMStateDescription*[]) {
1907         &vmstate_spapr_ov5_cas,
1908         &vmstate_spapr_patb_entry,
1909         &vmstate_spapr_pending_events,
1910         &vmstate_spapr_cap_htm,
1911         &vmstate_spapr_cap_vsx,
1912         &vmstate_spapr_cap_dfp,
1913         &vmstate_spapr_cap_cfpc,
1914         &vmstate_spapr_cap_sbbc,
1915         &vmstate_spapr_cap_ibs,
1916         &vmstate_spapr_irq_map,
1917         NULL
1918     }
1919 };
1920 
1921 static int htab_save_setup(QEMUFile *f, void *opaque)
1922 {
1923     sPAPRMachineState *spapr = opaque;
1924 
1925     /* "Iteration" header */
1926     if (!spapr->htab_shift) {
1927         qemu_put_be32(f, -1);
1928     } else {
1929         qemu_put_be32(f, spapr->htab_shift);
1930     }
1931 
1932     if (spapr->htab) {
1933         spapr->htab_save_index = 0;
1934         spapr->htab_first_pass = true;
1935     } else {
1936         if (spapr->htab_shift) {
1937             assert(kvm_enabled());
1938         }
1939     }
1940 
1941 
1942     return 0;
1943 }
1944 
1945 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1946                             int chunkstart, int n_valid, int n_invalid)
1947 {
1948     qemu_put_be32(f, chunkstart);
1949     qemu_put_be16(f, n_valid);
1950     qemu_put_be16(f, n_invalid);
1951     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1952                     HASH_PTE_SIZE_64 * n_valid);
1953 }
1954 
1955 static void htab_save_end_marker(QEMUFile *f)
1956 {
1957     qemu_put_be32(f, 0);
1958     qemu_put_be16(f, 0);
1959     qemu_put_be16(f, 0);
1960 }
1961 
1962 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1963                                  int64_t max_ns)
1964 {
1965     bool has_timeout = max_ns != -1;
1966     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1967     int index = spapr->htab_save_index;
1968     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1969 
1970     assert(spapr->htab_first_pass);
1971 
1972     do {
1973         int chunkstart;
1974 
1975         /* Consume invalid HPTEs */
1976         while ((index < htabslots)
1977                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1978             CLEAN_HPTE(HPTE(spapr->htab, index));
1979             index++;
1980         }
1981 
1982         /* Consume valid HPTEs */
1983         chunkstart = index;
1984         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1985                && HPTE_VALID(HPTE(spapr->htab, index))) {
1986             CLEAN_HPTE(HPTE(spapr->htab, index));
1987             index++;
1988         }
1989 
1990         if (index > chunkstart) {
1991             int n_valid = index - chunkstart;
1992 
1993             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1994 
1995             if (has_timeout &&
1996                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1997                 break;
1998             }
1999         }
2000     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2001 
2002     if (index >= htabslots) {
2003         assert(index == htabslots);
2004         index = 0;
2005         spapr->htab_first_pass = false;
2006     }
2007     spapr->htab_save_index = index;
2008 }
2009 
2010 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2011                                 int64_t max_ns)
2012 {
2013     bool final = max_ns < 0;
2014     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2015     int examined = 0, sent = 0;
2016     int index = spapr->htab_save_index;
2017     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2018 
2019     assert(!spapr->htab_first_pass);
2020 
2021     do {
2022         int chunkstart, invalidstart;
2023 
2024         /* Consume non-dirty HPTEs */
2025         while ((index < htabslots)
2026                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2027             index++;
2028             examined++;
2029         }
2030 
2031         chunkstart = index;
2032         /* Consume valid dirty HPTEs */
2033         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2034                && HPTE_DIRTY(HPTE(spapr->htab, index))
2035                && HPTE_VALID(HPTE(spapr->htab, index))) {
2036             CLEAN_HPTE(HPTE(spapr->htab, index));
2037             index++;
2038             examined++;
2039         }
2040 
2041         invalidstart = index;
2042         /* Consume invalid dirty HPTEs */
2043         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2044                && HPTE_DIRTY(HPTE(spapr->htab, index))
2045                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2046             CLEAN_HPTE(HPTE(spapr->htab, index));
2047             index++;
2048             examined++;
2049         }
2050 
2051         if (index > chunkstart) {
2052             int n_valid = invalidstart - chunkstart;
2053             int n_invalid = index - invalidstart;
2054 
2055             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2056             sent += index - chunkstart;
2057 
2058             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2059                 break;
2060             }
2061         }
2062 
2063         if (examined >= htabslots) {
2064             break;
2065         }
2066 
2067         if (index >= htabslots) {
2068             assert(index == htabslots);
2069             index = 0;
2070         }
2071     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2072 
2073     if (index >= htabslots) {
2074         assert(index == htabslots);
2075         index = 0;
2076     }
2077 
2078     spapr->htab_save_index = index;
2079 
2080     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2081 }
2082 
2083 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2084 #define MAX_KVM_BUF_SIZE    2048
2085 
2086 static int htab_save_iterate(QEMUFile *f, void *opaque)
2087 {
2088     sPAPRMachineState *spapr = opaque;
2089     int fd;
2090     int rc = 0;
2091 
2092     /* Iteration header */
2093     if (!spapr->htab_shift) {
2094         qemu_put_be32(f, -1);
2095         return 1;
2096     } else {
2097         qemu_put_be32(f, 0);
2098     }
2099 
2100     if (!spapr->htab) {
2101         assert(kvm_enabled());
2102 
2103         fd = get_htab_fd(spapr);
2104         if (fd < 0) {
2105             return fd;
2106         }
2107 
2108         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2109         if (rc < 0) {
2110             return rc;
2111         }
2112     } else  if (spapr->htab_first_pass) {
2113         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2114     } else {
2115         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2116     }
2117 
2118     htab_save_end_marker(f);
2119 
2120     return rc;
2121 }
2122 
2123 static int htab_save_complete(QEMUFile *f, void *opaque)
2124 {
2125     sPAPRMachineState *spapr = opaque;
2126     int fd;
2127 
2128     /* Iteration header */
2129     if (!spapr->htab_shift) {
2130         qemu_put_be32(f, -1);
2131         return 0;
2132     } else {
2133         qemu_put_be32(f, 0);
2134     }
2135 
2136     if (!spapr->htab) {
2137         int rc;
2138 
2139         assert(kvm_enabled());
2140 
2141         fd = get_htab_fd(spapr);
2142         if (fd < 0) {
2143             return fd;
2144         }
2145 
2146         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2147         if (rc < 0) {
2148             return rc;
2149         }
2150     } else {
2151         if (spapr->htab_first_pass) {
2152             htab_save_first_pass(f, spapr, -1);
2153         }
2154         htab_save_later_pass(f, spapr, -1);
2155     }
2156 
2157     /* End marker */
2158     htab_save_end_marker(f);
2159 
2160     return 0;
2161 }
2162 
2163 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2164 {
2165     sPAPRMachineState *spapr = opaque;
2166     uint32_t section_hdr;
2167     int fd = -1;
2168     Error *local_err = NULL;
2169 
2170     if (version_id < 1 || version_id > 1) {
2171         error_report("htab_load() bad version");
2172         return -EINVAL;
2173     }
2174 
2175     section_hdr = qemu_get_be32(f);
2176 
2177     if (section_hdr == -1) {
2178         spapr_free_hpt(spapr);
2179         return 0;
2180     }
2181 
2182     if (section_hdr) {
2183         /* First section gives the htab size */
2184         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2185         if (local_err) {
2186             error_report_err(local_err);
2187             return -EINVAL;
2188         }
2189         return 0;
2190     }
2191 
2192     if (!spapr->htab) {
2193         assert(kvm_enabled());
2194 
2195         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2196         if (fd < 0) {
2197             error_report_err(local_err);
2198             return fd;
2199         }
2200     }
2201 
2202     while (true) {
2203         uint32_t index;
2204         uint16_t n_valid, n_invalid;
2205 
2206         index = qemu_get_be32(f);
2207         n_valid = qemu_get_be16(f);
2208         n_invalid = qemu_get_be16(f);
2209 
2210         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2211             /* End of Stream */
2212             break;
2213         }
2214 
2215         if ((index + n_valid + n_invalid) >
2216             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2217             /* Bad index in stream */
2218             error_report(
2219                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2220                 index, n_valid, n_invalid, spapr->htab_shift);
2221             return -EINVAL;
2222         }
2223 
2224         if (spapr->htab) {
2225             if (n_valid) {
2226                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2227                                 HASH_PTE_SIZE_64 * n_valid);
2228             }
2229             if (n_invalid) {
2230                 memset(HPTE(spapr->htab, index + n_valid), 0,
2231                        HASH_PTE_SIZE_64 * n_invalid);
2232             }
2233         } else {
2234             int rc;
2235 
2236             assert(fd >= 0);
2237 
2238             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2239             if (rc < 0) {
2240                 return rc;
2241             }
2242         }
2243     }
2244 
2245     if (!spapr->htab) {
2246         assert(fd >= 0);
2247         close(fd);
2248     }
2249 
2250     return 0;
2251 }
2252 
2253 static void htab_save_cleanup(void *opaque)
2254 {
2255     sPAPRMachineState *spapr = opaque;
2256 
2257     close_htab_fd(spapr);
2258 }
2259 
2260 static SaveVMHandlers savevm_htab_handlers = {
2261     .save_setup = htab_save_setup,
2262     .save_live_iterate = htab_save_iterate,
2263     .save_live_complete_precopy = htab_save_complete,
2264     .save_cleanup = htab_save_cleanup,
2265     .load_state = htab_load,
2266 };
2267 
2268 static void spapr_boot_set(void *opaque, const char *boot_device,
2269                            Error **errp)
2270 {
2271     MachineState *machine = MACHINE(opaque);
2272     machine->boot_order = g_strdup(boot_device);
2273 }
2274 
2275 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2276 {
2277     MachineState *machine = MACHINE(spapr);
2278     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2279     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2280     int i;
2281 
2282     for (i = 0; i < nr_lmbs; i++) {
2283         uint64_t addr;
2284 
2285         addr = i * lmb_size + machine->device_memory->base;
2286         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2287                                addr / lmb_size);
2288     }
2289 }
2290 
2291 /*
2292  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2293  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2294  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2295  */
2296 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2297 {
2298     int i;
2299 
2300     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2301         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2302                    " is not aligned to %" PRIu64 " MiB",
2303                    machine->ram_size,
2304                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2305         return;
2306     }
2307 
2308     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2309         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2310                    " is not aligned to %" PRIu64 " MiB",
2311                    machine->ram_size,
2312                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2313         return;
2314     }
2315 
2316     for (i = 0; i < nb_numa_nodes; i++) {
2317         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2318             error_setg(errp,
2319                        "Node %d memory size 0x%" PRIx64
2320                        " is not aligned to %" PRIu64 " MiB",
2321                        i, numa_info[i].node_mem,
2322                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2323             return;
2324         }
2325     }
2326 }
2327 
2328 /* find cpu slot in machine->possible_cpus by core_id */
2329 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2330 {
2331     int index = id / smp_threads;
2332 
2333     if (index >= ms->possible_cpus->len) {
2334         return NULL;
2335     }
2336     if (idx) {
2337         *idx = index;
2338     }
2339     return &ms->possible_cpus->cpus[index];
2340 }
2341 
2342 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2343 {
2344     Error *local_err = NULL;
2345     bool vsmt_user = !!spapr->vsmt;
2346     int kvm_smt = kvmppc_smt_threads();
2347     int ret;
2348 
2349     if (!kvm_enabled() && (smp_threads > 1)) {
2350         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2351                      "on a pseries machine");
2352         goto out;
2353     }
2354     if (!is_power_of_2(smp_threads)) {
2355         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2356                      "machine because it must be a power of 2", smp_threads);
2357         goto out;
2358     }
2359 
2360     /* Detemine the VSMT mode to use: */
2361     if (vsmt_user) {
2362         if (spapr->vsmt < smp_threads) {
2363             error_setg(&local_err, "Cannot support VSMT mode %d"
2364                          " because it must be >= threads/core (%d)",
2365                          spapr->vsmt, smp_threads);
2366             goto out;
2367         }
2368         /* In this case, spapr->vsmt has been set by the command line */
2369     } else {
2370         /*
2371          * Default VSMT value is tricky, because we need it to be as
2372          * consistent as possible (for migration), but this requires
2373          * changing it for at least some existing cases.  We pick 8 as
2374          * the value that we'd get with KVM on POWER8, the
2375          * overwhelmingly common case in production systems.
2376          */
2377         spapr->vsmt = MAX(8, smp_threads);
2378     }
2379 
2380     /* KVM: If necessary, set the SMT mode: */
2381     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2382         ret = kvmppc_set_smt_threads(spapr->vsmt);
2383         if (ret) {
2384             /* Looks like KVM isn't able to change VSMT mode */
2385             error_setg(&local_err,
2386                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2387                        spapr->vsmt, ret);
2388             /* We can live with that if the default one is big enough
2389              * for the number of threads, and a submultiple of the one
2390              * we want.  In this case we'll waste some vcpu ids, but
2391              * behaviour will be correct */
2392             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2393                 warn_report_err(local_err);
2394                 local_err = NULL;
2395                 goto out;
2396             } else {
2397                 if (!vsmt_user) {
2398                     error_append_hint(&local_err,
2399                                       "On PPC, a VM with %d threads/core"
2400                                       " on a host with %d threads/core"
2401                                       " requires the use of VSMT mode %d.\n",
2402                                       smp_threads, kvm_smt, spapr->vsmt);
2403                 }
2404                 kvmppc_hint_smt_possible(&local_err);
2405                 goto out;
2406             }
2407         }
2408     }
2409     /* else TCG: nothing to do currently */
2410 out:
2411     error_propagate(errp, local_err);
2412 }
2413 
2414 static void spapr_init_cpus(sPAPRMachineState *spapr)
2415 {
2416     MachineState *machine = MACHINE(spapr);
2417     MachineClass *mc = MACHINE_GET_CLASS(machine);
2418     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2419     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2420     const CPUArchIdList *possible_cpus;
2421     int boot_cores_nr = smp_cpus / smp_threads;
2422     int i;
2423 
2424     possible_cpus = mc->possible_cpu_arch_ids(machine);
2425     if (mc->has_hotpluggable_cpus) {
2426         if (smp_cpus % smp_threads) {
2427             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2428                          smp_cpus, smp_threads);
2429             exit(1);
2430         }
2431         if (max_cpus % smp_threads) {
2432             error_report("max_cpus (%u) must be multiple of threads (%u)",
2433                          max_cpus, smp_threads);
2434             exit(1);
2435         }
2436     } else {
2437         if (max_cpus != smp_cpus) {
2438             error_report("This machine version does not support CPU hotplug");
2439             exit(1);
2440         }
2441         boot_cores_nr = possible_cpus->len;
2442     }
2443 
2444     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2445      * call xics_max_server_number() or spapr_vcpu_id().
2446      */
2447     spapr_set_vsmt_mode(spapr, &error_fatal);
2448 
2449     if (smc->pre_2_10_has_unused_icps) {
2450         int i;
2451 
2452         for (i = 0; i < xics_max_server_number(spapr); i++) {
2453             /* Dummy entries get deregistered when real ICPState objects
2454              * are registered during CPU core hotplug.
2455              */
2456             pre_2_10_vmstate_register_dummy_icp(i);
2457         }
2458     }
2459 
2460     for (i = 0; i < possible_cpus->len; i++) {
2461         int core_id = i * smp_threads;
2462 
2463         if (mc->has_hotpluggable_cpus) {
2464             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2465                                    spapr_vcpu_id(spapr, core_id));
2466         }
2467 
2468         if (i < boot_cores_nr) {
2469             Object *core  = object_new(type);
2470             int nr_threads = smp_threads;
2471 
2472             /* Handle the partially filled core for older machine types */
2473             if ((i + 1) * smp_threads >= smp_cpus) {
2474                 nr_threads = smp_cpus - i * smp_threads;
2475             }
2476 
2477             object_property_set_int(core, nr_threads, "nr-threads",
2478                                     &error_fatal);
2479             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2480                                     &error_fatal);
2481             object_property_set_bool(core, true, "realized", &error_fatal);
2482         }
2483     }
2484 }
2485 
2486 /* pSeries LPAR / sPAPR hardware init */
2487 static void spapr_machine_init(MachineState *machine)
2488 {
2489     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2490     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2491     const char *kernel_filename = machine->kernel_filename;
2492     const char *initrd_filename = machine->initrd_filename;
2493     PCIHostState *phb;
2494     int i;
2495     MemoryRegion *sysmem = get_system_memory();
2496     MemoryRegion *ram = g_new(MemoryRegion, 1);
2497     hwaddr node0_size = spapr_node0_size(machine);
2498     long load_limit, fw_size;
2499     char *filename;
2500     Error *resize_hpt_err = NULL;
2501 
2502     msi_nonbroken = true;
2503 
2504     QLIST_INIT(&spapr->phbs);
2505     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2506 
2507     /* Determine capabilities to run with */
2508     spapr_caps_init(spapr);
2509 
2510     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2511     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2512         /*
2513          * If the user explicitly requested a mode we should either
2514          * supply it, or fail completely (which we do below).  But if
2515          * it's not set explicitly, we reset our mode to something
2516          * that works
2517          */
2518         if (resize_hpt_err) {
2519             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2520             error_free(resize_hpt_err);
2521             resize_hpt_err = NULL;
2522         } else {
2523             spapr->resize_hpt = smc->resize_hpt_default;
2524         }
2525     }
2526 
2527     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2528 
2529     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2530         /*
2531          * User requested HPT resize, but this host can't supply it.  Bail out
2532          */
2533         error_report_err(resize_hpt_err);
2534         exit(1);
2535     }
2536 
2537     spapr->rma_size = node0_size;
2538 
2539     /* With KVM, we don't actually know whether KVM supports an
2540      * unbounded RMA (PR KVM) or is limited by the hash table size
2541      * (HV KVM using VRMA), so we always assume the latter
2542      *
2543      * In that case, we also limit the initial allocations for RTAS
2544      * etc... to 256M since we have no way to know what the VRMA size
2545      * is going to be as it depends on the size of the hash table
2546      * which isn't determined yet.
2547      */
2548     if (kvm_enabled()) {
2549         spapr->vrma_adjust = 1;
2550         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2551     }
2552 
2553     /* Actually we don't support unbounded RMA anymore since we added
2554      * proper emulation of HV mode. The max we can get is 16G which
2555      * also happens to be what we configure for PAPR mode so make sure
2556      * we don't do anything bigger than that
2557      */
2558     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2559 
2560     if (spapr->rma_size > node0_size) {
2561         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2562                      spapr->rma_size);
2563         exit(1);
2564     }
2565 
2566     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2567     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2568 
2569     /* Set up Interrupt Controller before we create the VCPUs */
2570     smc->irq->init(spapr, &error_fatal);
2571 
2572     /* Set up containers for ibm,client-architecture-support negotiated options
2573      */
2574     spapr->ov5 = spapr_ovec_new();
2575     spapr->ov5_cas = spapr_ovec_new();
2576 
2577     if (smc->dr_lmb_enabled) {
2578         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2579         spapr_validate_node_memory(machine, &error_fatal);
2580     }
2581 
2582     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2583 
2584     /* advertise support for dedicated HP event source to guests */
2585     if (spapr->use_hotplug_event_source) {
2586         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2587     }
2588 
2589     /* advertise support for HPT resizing */
2590     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2591         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2592     }
2593 
2594     /* advertise support for ibm,dyamic-memory-v2 */
2595     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2596 
2597     /* init CPUs */
2598     spapr_init_cpus(spapr);
2599 
2600     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2601         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2602                               spapr->max_compat_pvr)) {
2603         /* KVM and TCG always allow GTSE with radix... */
2604         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2605     }
2606     /* ... but not with hash (currently). */
2607 
2608     if (kvm_enabled()) {
2609         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2610         kvmppc_enable_logical_ci_hcalls();
2611         kvmppc_enable_set_mode_hcall();
2612 
2613         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2614         kvmppc_enable_clear_ref_mod_hcalls();
2615     }
2616 
2617     /* allocate RAM */
2618     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2619                                          machine->ram_size);
2620     memory_region_add_subregion(sysmem, 0, ram);
2621 
2622     /* always allocate the device memory information */
2623     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2624 
2625     /* initialize hotplug memory address space */
2626     if (machine->ram_size < machine->maxram_size) {
2627         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2628         /*
2629          * Limit the number of hotpluggable memory slots to half the number
2630          * slots that KVM supports, leaving the other half for PCI and other
2631          * devices. However ensure that number of slots doesn't drop below 32.
2632          */
2633         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2634                            SPAPR_MAX_RAM_SLOTS;
2635 
2636         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2637             max_memslots = SPAPR_MAX_RAM_SLOTS;
2638         }
2639         if (machine->ram_slots > max_memslots) {
2640             error_report("Specified number of memory slots %"
2641                          PRIu64" exceeds max supported %d",
2642                          machine->ram_slots, max_memslots);
2643             exit(1);
2644         }
2645 
2646         machine->device_memory->base = ROUND_UP(machine->ram_size,
2647                                                 SPAPR_DEVICE_MEM_ALIGN);
2648         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2649                            "device-memory", device_mem_size);
2650         memory_region_add_subregion(sysmem, machine->device_memory->base,
2651                                     &machine->device_memory->mr);
2652     }
2653 
2654     if (smc->dr_lmb_enabled) {
2655         spapr_create_lmb_dr_connectors(spapr);
2656     }
2657 
2658     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2659     if (!filename) {
2660         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2661         exit(1);
2662     }
2663     spapr->rtas_size = get_image_size(filename);
2664     if (spapr->rtas_size < 0) {
2665         error_report("Could not get size of LPAR rtas '%s'", filename);
2666         exit(1);
2667     }
2668     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2669     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2670         error_report("Could not load LPAR rtas '%s'", filename);
2671         exit(1);
2672     }
2673     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2674         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2675                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2676         exit(1);
2677     }
2678     g_free(filename);
2679 
2680     /* Set up RTAS event infrastructure */
2681     spapr_events_init(spapr);
2682 
2683     /* Set up the RTC RTAS interfaces */
2684     spapr_rtc_create(spapr);
2685 
2686     /* Set up VIO bus */
2687     spapr->vio_bus = spapr_vio_bus_init();
2688 
2689     for (i = 0; i < serial_max_hds(); i++) {
2690         if (serial_hd(i)) {
2691             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2692         }
2693     }
2694 
2695     /* We always have at least the nvram device on VIO */
2696     spapr_create_nvram(spapr);
2697 
2698     /* Set up PCI */
2699     spapr_pci_rtas_init();
2700 
2701     phb = spapr_create_phb(spapr, 0);
2702 
2703     for (i = 0; i < nb_nics; i++) {
2704         NICInfo *nd = &nd_table[i];
2705 
2706         if (!nd->model) {
2707             nd->model = g_strdup("spapr-vlan");
2708         }
2709 
2710         if (g_str_equal(nd->model, "spapr-vlan") ||
2711             g_str_equal(nd->model, "ibmveth")) {
2712             spapr_vlan_create(spapr->vio_bus, nd);
2713         } else {
2714             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2715         }
2716     }
2717 
2718     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2719         spapr_vscsi_create(spapr->vio_bus);
2720     }
2721 
2722     /* Graphics */
2723     if (spapr_vga_init(phb->bus, &error_fatal)) {
2724         spapr->has_graphics = true;
2725         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2726     }
2727 
2728     if (machine->usb) {
2729         if (smc->use_ohci_by_default) {
2730             pci_create_simple(phb->bus, -1, "pci-ohci");
2731         } else {
2732             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2733         }
2734 
2735         if (spapr->has_graphics) {
2736             USBBus *usb_bus = usb_bus_find(-1);
2737 
2738             usb_create_simple(usb_bus, "usb-kbd");
2739             usb_create_simple(usb_bus, "usb-mouse");
2740         }
2741     }
2742 
2743     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2744         error_report(
2745             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2746             MIN_RMA_SLOF);
2747         exit(1);
2748     }
2749 
2750     if (kernel_filename) {
2751         uint64_t lowaddr = 0;
2752 
2753         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2754                                       NULL, NULL, &lowaddr, NULL, 1,
2755                                       PPC_ELF_MACHINE, 0, 0);
2756         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2757             spapr->kernel_size = load_elf(kernel_filename,
2758                                           translate_kernel_address, NULL, NULL,
2759                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2760                                           0, 0);
2761             spapr->kernel_le = spapr->kernel_size > 0;
2762         }
2763         if (spapr->kernel_size < 0) {
2764             error_report("error loading %s: %s", kernel_filename,
2765                          load_elf_strerror(spapr->kernel_size));
2766             exit(1);
2767         }
2768 
2769         /* load initrd */
2770         if (initrd_filename) {
2771             /* Try to locate the initrd in the gap between the kernel
2772              * and the firmware. Add a bit of space just in case
2773              */
2774             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2775                                   + 0x1ffff) & ~0xffff;
2776             spapr->initrd_size = load_image_targphys(initrd_filename,
2777                                                      spapr->initrd_base,
2778                                                      load_limit
2779                                                      - spapr->initrd_base);
2780             if (spapr->initrd_size < 0) {
2781                 error_report("could not load initial ram disk '%s'",
2782                              initrd_filename);
2783                 exit(1);
2784             }
2785         }
2786     }
2787 
2788     if (bios_name == NULL) {
2789         bios_name = FW_FILE_NAME;
2790     }
2791     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2792     if (!filename) {
2793         error_report("Could not find LPAR firmware '%s'", bios_name);
2794         exit(1);
2795     }
2796     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2797     if (fw_size <= 0) {
2798         error_report("Could not load LPAR firmware '%s'", filename);
2799         exit(1);
2800     }
2801     g_free(filename);
2802 
2803     /* FIXME: Should register things through the MachineState's qdev
2804      * interface, this is a legacy from the sPAPREnvironment structure
2805      * which predated MachineState but had a similar function */
2806     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2807     register_savevm_live(NULL, "spapr/htab", -1, 1,
2808                          &savevm_htab_handlers, spapr);
2809 
2810     qemu_register_boot_set(spapr_boot_set, spapr);
2811 
2812     if (kvm_enabled()) {
2813         /* to stop and start vmclock */
2814         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2815                                          &spapr->tb);
2816 
2817         kvmppc_spapr_enable_inkernel_multitce();
2818     }
2819 }
2820 
2821 static int spapr_kvm_type(const char *vm_type)
2822 {
2823     if (!vm_type) {
2824         return 0;
2825     }
2826 
2827     if (!strcmp(vm_type, "HV")) {
2828         return 1;
2829     }
2830 
2831     if (!strcmp(vm_type, "PR")) {
2832         return 2;
2833     }
2834 
2835     error_report("Unknown kvm-type specified '%s'", vm_type);
2836     exit(1);
2837 }
2838 
2839 /*
2840  * Implementation of an interface to adjust firmware path
2841  * for the bootindex property handling.
2842  */
2843 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2844                                    DeviceState *dev)
2845 {
2846 #define CAST(type, obj, name) \
2847     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2848     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2849     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2850     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2851 
2852     if (d) {
2853         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2854         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2855         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2856 
2857         if (spapr) {
2858             /*
2859              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2860              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2861              * in the top 16 bits of the 64-bit LUN
2862              */
2863             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2864             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2865                                    (uint64_t)id << 48);
2866         } else if (virtio) {
2867             /*
2868              * We use SRP luns of the form 01000000 | (target << 8) | lun
2869              * in the top 32 bits of the 64-bit LUN
2870              * Note: the quote above is from SLOF and it is wrong,
2871              * the actual binding is:
2872              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2873              */
2874             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2875             if (d->lun >= 256) {
2876                 /* Use the LUN "flat space addressing method" */
2877                 id |= 0x4000;
2878             }
2879             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2880                                    (uint64_t)id << 32);
2881         } else if (usb) {
2882             /*
2883              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2884              * in the top 32 bits of the 64-bit LUN
2885              */
2886             unsigned usb_port = atoi(usb->port->path);
2887             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2888             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2889                                    (uint64_t)id << 32);
2890         }
2891     }
2892 
2893     /*
2894      * SLOF probes the USB devices, and if it recognizes that the device is a
2895      * storage device, it changes its name to "storage" instead of "usb-host",
2896      * and additionally adds a child node for the SCSI LUN, so the correct
2897      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2898      */
2899     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2900         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2901         if (usb_host_dev_is_scsi_storage(usbdev)) {
2902             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2903         }
2904     }
2905 
2906     if (phb) {
2907         /* Replace "pci" with "pci@800000020000000" */
2908         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2909     }
2910 
2911     if (vsc) {
2912         /* Same logic as virtio above */
2913         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2914         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2915     }
2916 
2917     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2918         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2919         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2920         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2921     }
2922 
2923     return NULL;
2924 }
2925 
2926 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2927 {
2928     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2929 
2930     return g_strdup(spapr->kvm_type);
2931 }
2932 
2933 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2934 {
2935     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2936 
2937     g_free(spapr->kvm_type);
2938     spapr->kvm_type = g_strdup(value);
2939 }
2940 
2941 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2942 {
2943     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2944 
2945     return spapr->use_hotplug_event_source;
2946 }
2947 
2948 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2949                                             Error **errp)
2950 {
2951     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2952 
2953     spapr->use_hotplug_event_source = value;
2954 }
2955 
2956 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2957 {
2958     return true;
2959 }
2960 
2961 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2962 {
2963     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2964 
2965     switch (spapr->resize_hpt) {
2966     case SPAPR_RESIZE_HPT_DEFAULT:
2967         return g_strdup("default");
2968     case SPAPR_RESIZE_HPT_DISABLED:
2969         return g_strdup("disabled");
2970     case SPAPR_RESIZE_HPT_ENABLED:
2971         return g_strdup("enabled");
2972     case SPAPR_RESIZE_HPT_REQUIRED:
2973         return g_strdup("required");
2974     }
2975     g_assert_not_reached();
2976 }
2977 
2978 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2979 {
2980     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2981 
2982     if (strcmp(value, "default") == 0) {
2983         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2984     } else if (strcmp(value, "disabled") == 0) {
2985         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2986     } else if (strcmp(value, "enabled") == 0) {
2987         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2988     } else if (strcmp(value, "required") == 0) {
2989         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2990     } else {
2991         error_setg(errp, "Bad value for \"resize-hpt\" property");
2992     }
2993 }
2994 
2995 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2996                                    void *opaque, Error **errp)
2997 {
2998     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2999 }
3000 
3001 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3002                                    void *opaque, Error **errp)
3003 {
3004     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3005 }
3006 
3007 static void spapr_instance_init(Object *obj)
3008 {
3009     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3010 
3011     spapr->htab_fd = -1;
3012     spapr->use_hotplug_event_source = true;
3013     object_property_add_str(obj, "kvm-type",
3014                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3015     object_property_set_description(obj, "kvm-type",
3016                                     "Specifies the KVM virtualization mode (HV, PR)",
3017                                     NULL);
3018     object_property_add_bool(obj, "modern-hotplug-events",
3019                             spapr_get_modern_hotplug_events,
3020                             spapr_set_modern_hotplug_events,
3021                             NULL);
3022     object_property_set_description(obj, "modern-hotplug-events",
3023                                     "Use dedicated hotplug event mechanism in"
3024                                     " place of standard EPOW events when possible"
3025                                     " (required for memory hot-unplug support)",
3026                                     NULL);
3027     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3028                             "Maximum permitted CPU compatibility mode",
3029                             &error_fatal);
3030 
3031     object_property_add_str(obj, "resize-hpt",
3032                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3033     object_property_set_description(obj, "resize-hpt",
3034                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3035                                     NULL);
3036     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3037                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3038     object_property_set_description(obj, "vsmt",
3039                                     "Virtual SMT: KVM behaves as if this were"
3040                                     " the host's SMT mode", &error_abort);
3041     object_property_add_bool(obj, "vfio-no-msix-emulation",
3042                              spapr_get_msix_emulation, NULL, NULL);
3043 }
3044 
3045 static void spapr_machine_finalizefn(Object *obj)
3046 {
3047     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3048 
3049     g_free(spapr->kvm_type);
3050 }
3051 
3052 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3053 {
3054     cpu_synchronize_state(cs);
3055     ppc_cpu_do_system_reset(cs);
3056 }
3057 
3058 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3059 {
3060     CPUState *cs;
3061 
3062     CPU_FOREACH(cs) {
3063         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3064     }
3065 }
3066 
3067 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3068                            uint32_t node, bool dedicated_hp_event_source,
3069                            Error **errp)
3070 {
3071     sPAPRDRConnector *drc;
3072     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3073     int i, fdt_offset, fdt_size;
3074     void *fdt;
3075     uint64_t addr = addr_start;
3076     bool hotplugged = spapr_drc_hotplugged(dev);
3077     Error *local_err = NULL;
3078 
3079     for (i = 0; i < nr_lmbs; i++) {
3080         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3081                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3082         g_assert(drc);
3083 
3084         fdt = create_device_tree(&fdt_size);
3085         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3086                                                 SPAPR_MEMORY_BLOCK_SIZE);
3087 
3088         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3089         if (local_err) {
3090             while (addr > addr_start) {
3091                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3092                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3093                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3094                 spapr_drc_detach(drc);
3095             }
3096             g_free(fdt);
3097             error_propagate(errp, local_err);
3098             return;
3099         }
3100         if (!hotplugged) {
3101             spapr_drc_reset(drc);
3102         }
3103         addr += SPAPR_MEMORY_BLOCK_SIZE;
3104     }
3105     /* send hotplug notification to the
3106      * guest only in case of hotplugged memory
3107      */
3108     if (hotplugged) {
3109         if (dedicated_hp_event_source) {
3110             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3111                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3112             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3113                                                    nr_lmbs,
3114                                                    spapr_drc_index(drc));
3115         } else {
3116             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3117                                            nr_lmbs);
3118         }
3119     }
3120 }
3121 
3122 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3123                               Error **errp)
3124 {
3125     Error *local_err = NULL;
3126     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3127     PCDIMMDevice *dimm = PC_DIMM(dev);
3128     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3129     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3130     uint64_t size, addr;
3131     uint32_t node;
3132 
3133     size = memory_region_size(mr);
3134 
3135     pc_dimm_plug(dev, MACHINE(ms), &local_err);
3136     if (local_err) {
3137         goto out;
3138     }
3139 
3140     addr = object_property_get_uint(OBJECT(dimm),
3141                                     PC_DIMM_ADDR_PROP, &local_err);
3142     if (local_err) {
3143         goto out_unplug;
3144     }
3145 
3146     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3147                                     &error_abort);
3148     spapr_add_lmbs(dev, addr, size, node,
3149                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3150                    &local_err);
3151     if (local_err) {
3152         goto out_unplug;
3153     }
3154 
3155     return;
3156 
3157 out_unplug:
3158     pc_dimm_unplug(dev, MACHINE(ms));
3159 out:
3160     error_propagate(errp, local_err);
3161 }
3162 
3163 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3164                                   Error **errp)
3165 {
3166     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3167     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3168     PCDIMMDevice *dimm = PC_DIMM(dev);
3169     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3170     Error *local_err = NULL;
3171     MemoryRegion *mr;
3172     uint64_t size;
3173     Object *memdev;
3174     hwaddr pagesize;
3175 
3176     if (!smc->dr_lmb_enabled) {
3177         error_setg(errp, "Memory hotplug not supported for this machine");
3178         return;
3179     }
3180 
3181     mr = ddc->get_memory_region(dimm, errp);
3182     if (!mr) {
3183         return;
3184     }
3185     size = memory_region_size(mr);
3186 
3187     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3188         error_setg(errp, "Hotplugged memory size must be a multiple of "
3189                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3190         return;
3191     }
3192 
3193     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3194                                       &error_abort);
3195     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3196     spapr_check_pagesize(spapr, pagesize, &local_err);
3197     if (local_err) {
3198         error_propagate(errp, local_err);
3199         return;
3200     }
3201 
3202     pc_dimm_pre_plug(dev, MACHINE(hotplug_dev), NULL, errp);
3203 }
3204 
3205 struct sPAPRDIMMState {
3206     PCDIMMDevice *dimm;
3207     uint32_t nr_lmbs;
3208     QTAILQ_ENTRY(sPAPRDIMMState) next;
3209 };
3210 
3211 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3212                                                        PCDIMMDevice *dimm)
3213 {
3214     sPAPRDIMMState *dimm_state = NULL;
3215 
3216     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3217         if (dimm_state->dimm == dimm) {
3218             break;
3219         }
3220     }
3221     return dimm_state;
3222 }
3223 
3224 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3225                                                       uint32_t nr_lmbs,
3226                                                       PCDIMMDevice *dimm)
3227 {
3228     sPAPRDIMMState *ds = NULL;
3229 
3230     /*
3231      * If this request is for a DIMM whose removal had failed earlier
3232      * (due to guest's refusal to remove the LMBs), we would have this
3233      * dimm already in the pending_dimm_unplugs list. In that
3234      * case don't add again.
3235      */
3236     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3237     if (!ds) {
3238         ds = g_malloc0(sizeof(sPAPRDIMMState));
3239         ds->nr_lmbs = nr_lmbs;
3240         ds->dimm = dimm;
3241         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3242     }
3243     return ds;
3244 }
3245 
3246 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3247                                               sPAPRDIMMState *dimm_state)
3248 {
3249     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3250     g_free(dimm_state);
3251 }
3252 
3253 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3254                                                         PCDIMMDevice *dimm)
3255 {
3256     sPAPRDRConnector *drc;
3257     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3258     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3259     uint64_t size = memory_region_size(mr);
3260     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3261     uint32_t avail_lmbs = 0;
3262     uint64_t addr_start, addr;
3263     int i;
3264 
3265     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3266                                          &error_abort);
3267 
3268     addr = addr_start;
3269     for (i = 0; i < nr_lmbs; i++) {
3270         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3271                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3272         g_assert(drc);
3273         if (drc->dev) {
3274             avail_lmbs++;
3275         }
3276         addr += SPAPR_MEMORY_BLOCK_SIZE;
3277     }
3278 
3279     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3280 }
3281 
3282 /* Callback to be called during DRC release. */
3283 void spapr_lmb_release(DeviceState *dev)
3284 {
3285     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3286     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3287     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3288 
3289     /* This information will get lost if a migration occurs
3290      * during the unplug process. In this case recover it. */
3291     if (ds == NULL) {
3292         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3293         g_assert(ds);
3294         /* The DRC being examined by the caller at least must be counted */
3295         g_assert(ds->nr_lmbs);
3296     }
3297 
3298     if (--ds->nr_lmbs) {
3299         return;
3300     }
3301 
3302     /*
3303      * Now that all the LMBs have been removed by the guest, call the
3304      * unplug handler chain. This can never fail.
3305      */
3306     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3307 }
3308 
3309 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3310 {
3311     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3312     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3313 
3314     pc_dimm_unplug(dev, MACHINE(hotplug_dev));
3315     object_unparent(OBJECT(dev));
3316     spapr_pending_dimm_unplugs_remove(spapr, ds);
3317 }
3318 
3319 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3320                                         DeviceState *dev, Error **errp)
3321 {
3322     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3323     Error *local_err = NULL;
3324     PCDIMMDevice *dimm = PC_DIMM(dev);
3325     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3326     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3327     uint32_t nr_lmbs;
3328     uint64_t size, addr_start, addr;
3329     int i;
3330     sPAPRDRConnector *drc;
3331 
3332     size = memory_region_size(mr);
3333     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3334 
3335     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3336                                          &local_err);
3337     if (local_err) {
3338         goto out;
3339     }
3340 
3341     /*
3342      * An existing pending dimm state for this DIMM means that there is an
3343      * unplug operation in progress, waiting for the spapr_lmb_release
3344      * callback to complete the job (BQL can't cover that far). In this case,
3345      * bail out to avoid detaching DRCs that were already released.
3346      */
3347     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3348         error_setg(&local_err,
3349                    "Memory unplug already in progress for device %s",
3350                    dev->id);
3351         goto out;
3352     }
3353 
3354     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3355 
3356     addr = addr_start;
3357     for (i = 0; i < nr_lmbs; i++) {
3358         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3359                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3360         g_assert(drc);
3361 
3362         spapr_drc_detach(drc);
3363         addr += SPAPR_MEMORY_BLOCK_SIZE;
3364     }
3365 
3366     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3367                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3368     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3369                                               nr_lmbs, spapr_drc_index(drc));
3370 out:
3371     error_propagate(errp, local_err);
3372 }
3373 
3374 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3375                                            sPAPRMachineState *spapr)
3376 {
3377     PowerPCCPU *cpu = POWERPC_CPU(cs);
3378     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3379     int id = spapr_get_vcpu_id(cpu);
3380     void *fdt;
3381     int offset, fdt_size;
3382     char *nodename;
3383 
3384     fdt = create_device_tree(&fdt_size);
3385     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3386     offset = fdt_add_subnode(fdt, 0, nodename);
3387 
3388     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3389     g_free(nodename);
3390 
3391     *fdt_offset = offset;
3392     return fdt;
3393 }
3394 
3395 /* Callback to be called during DRC release. */
3396 void spapr_core_release(DeviceState *dev)
3397 {
3398     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3399 
3400     /* Call the unplug handler chain. This can never fail. */
3401     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3402 }
3403 
3404 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3405 {
3406     MachineState *ms = MACHINE(hotplug_dev);
3407     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3408     CPUCore *cc = CPU_CORE(dev);
3409     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3410 
3411     if (smc->pre_2_10_has_unused_icps) {
3412         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3413         int i;
3414 
3415         for (i = 0; i < cc->nr_threads; i++) {
3416             CPUState *cs = CPU(sc->threads[i]);
3417 
3418             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3419         }
3420     }
3421 
3422     assert(core_slot);
3423     core_slot->cpu = NULL;
3424     object_unparent(OBJECT(dev));
3425 }
3426 
3427 static
3428 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3429                                Error **errp)
3430 {
3431     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3432     int index;
3433     sPAPRDRConnector *drc;
3434     CPUCore *cc = CPU_CORE(dev);
3435 
3436     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3437         error_setg(errp, "Unable to find CPU core with core-id: %d",
3438                    cc->core_id);
3439         return;
3440     }
3441     if (index == 0) {
3442         error_setg(errp, "Boot CPU core may not be unplugged");
3443         return;
3444     }
3445 
3446     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3447                           spapr_vcpu_id(spapr, cc->core_id));
3448     g_assert(drc);
3449 
3450     spapr_drc_detach(drc);
3451 
3452     spapr_hotplug_req_remove_by_index(drc);
3453 }
3454 
3455 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3456                             Error **errp)
3457 {
3458     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3459     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3460     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3461     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3462     CPUCore *cc = CPU_CORE(dev);
3463     CPUState *cs = CPU(core->threads[0]);
3464     sPAPRDRConnector *drc;
3465     Error *local_err = NULL;
3466     CPUArchId *core_slot;
3467     int index;
3468     bool hotplugged = spapr_drc_hotplugged(dev);
3469 
3470     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3471     if (!core_slot) {
3472         error_setg(errp, "Unable to find CPU core with core-id: %d",
3473                    cc->core_id);
3474         return;
3475     }
3476     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3477                           spapr_vcpu_id(spapr, cc->core_id));
3478 
3479     g_assert(drc || !mc->has_hotpluggable_cpus);
3480 
3481     if (drc) {
3482         void *fdt;
3483         int fdt_offset;
3484 
3485         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3486 
3487         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3488         if (local_err) {
3489             g_free(fdt);
3490             error_propagate(errp, local_err);
3491             return;
3492         }
3493 
3494         if (hotplugged) {
3495             /*
3496              * Send hotplug notification interrupt to the guest only
3497              * in case of hotplugged CPUs.
3498              */
3499             spapr_hotplug_req_add_by_index(drc);
3500         } else {
3501             spapr_drc_reset(drc);
3502         }
3503     }
3504 
3505     core_slot->cpu = OBJECT(dev);
3506 
3507     if (smc->pre_2_10_has_unused_icps) {
3508         int i;
3509 
3510         for (i = 0; i < cc->nr_threads; i++) {
3511             cs = CPU(core->threads[i]);
3512             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3513         }
3514     }
3515 }
3516 
3517 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3518                                 Error **errp)
3519 {
3520     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3521     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3522     Error *local_err = NULL;
3523     CPUCore *cc = CPU_CORE(dev);
3524     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3525     const char *type = object_get_typename(OBJECT(dev));
3526     CPUArchId *core_slot;
3527     int index;
3528 
3529     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3530         error_setg(&local_err, "CPU hotplug not supported for this machine");
3531         goto out;
3532     }
3533 
3534     if (strcmp(base_core_type, type)) {
3535         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3536         goto out;
3537     }
3538 
3539     if (cc->core_id % smp_threads) {
3540         error_setg(&local_err, "invalid core id %d", cc->core_id);
3541         goto out;
3542     }
3543 
3544     /*
3545      * In general we should have homogeneous threads-per-core, but old
3546      * (pre hotplug support) machine types allow the last core to have
3547      * reduced threads as a compatibility hack for when we allowed
3548      * total vcpus not a multiple of threads-per-core.
3549      */
3550     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3551         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3552                    cc->nr_threads, smp_threads);
3553         goto out;
3554     }
3555 
3556     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3557     if (!core_slot) {
3558         error_setg(&local_err, "core id %d out of range", cc->core_id);
3559         goto out;
3560     }
3561 
3562     if (core_slot->cpu) {
3563         error_setg(&local_err, "core %d already populated", cc->core_id);
3564         goto out;
3565     }
3566 
3567     numa_cpu_pre_plug(core_slot, dev, &local_err);
3568 
3569 out:
3570     error_propagate(errp, local_err);
3571 }
3572 
3573 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3574                                       DeviceState *dev, Error **errp)
3575 {
3576     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3577         spapr_memory_plug(hotplug_dev, dev, errp);
3578     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3579         spapr_core_plug(hotplug_dev, dev, errp);
3580     }
3581 }
3582 
3583 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3584                                         DeviceState *dev, Error **errp)
3585 {
3586     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3587         spapr_memory_unplug(hotplug_dev, dev);
3588     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3589         spapr_core_unplug(hotplug_dev, dev);
3590     }
3591 }
3592 
3593 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3594                                                 DeviceState *dev, Error **errp)
3595 {
3596     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3597     MachineClass *mc = MACHINE_GET_CLASS(sms);
3598 
3599     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3600         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3601             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3602         } else {
3603             /* NOTE: this means there is a window after guest reset, prior to
3604              * CAS negotiation, where unplug requests will fail due to the
3605              * capability not being detected yet. This is a bit different than
3606              * the case with PCI unplug, where the events will be queued and
3607              * eventually handled by the guest after boot
3608              */
3609             error_setg(errp, "Memory hot unplug not supported for this guest");
3610         }
3611     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3612         if (!mc->has_hotpluggable_cpus) {
3613             error_setg(errp, "CPU hot unplug not supported on this machine");
3614             return;
3615         }
3616         spapr_core_unplug_request(hotplug_dev, dev, errp);
3617     }
3618 }
3619 
3620 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3621                                           DeviceState *dev, Error **errp)
3622 {
3623     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3624         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3625     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3626         spapr_core_pre_plug(hotplug_dev, dev, errp);
3627     }
3628 }
3629 
3630 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3631                                                  DeviceState *dev)
3632 {
3633     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3634         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3635         return HOTPLUG_HANDLER(machine);
3636     }
3637     return NULL;
3638 }
3639 
3640 static CpuInstanceProperties
3641 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3642 {
3643     CPUArchId *core_slot;
3644     MachineClass *mc = MACHINE_GET_CLASS(machine);
3645 
3646     /* make sure possible_cpu are intialized */
3647     mc->possible_cpu_arch_ids(machine);
3648     /* get CPU core slot containing thread that matches cpu_index */
3649     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3650     assert(core_slot);
3651     return core_slot->props;
3652 }
3653 
3654 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3655 {
3656     return idx / smp_cores % nb_numa_nodes;
3657 }
3658 
3659 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3660 {
3661     int i;
3662     const char *core_type;
3663     int spapr_max_cores = max_cpus / smp_threads;
3664     MachineClass *mc = MACHINE_GET_CLASS(machine);
3665 
3666     if (!mc->has_hotpluggable_cpus) {
3667         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3668     }
3669     if (machine->possible_cpus) {
3670         assert(machine->possible_cpus->len == spapr_max_cores);
3671         return machine->possible_cpus;
3672     }
3673 
3674     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3675     if (!core_type) {
3676         error_report("Unable to find sPAPR CPU Core definition");
3677         exit(1);
3678     }
3679 
3680     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3681                              sizeof(CPUArchId) * spapr_max_cores);
3682     machine->possible_cpus->len = spapr_max_cores;
3683     for (i = 0; i < machine->possible_cpus->len; i++) {
3684         int core_id = i * smp_threads;
3685 
3686         machine->possible_cpus->cpus[i].type = core_type;
3687         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3688         machine->possible_cpus->cpus[i].arch_id = core_id;
3689         machine->possible_cpus->cpus[i].props.has_core_id = true;
3690         machine->possible_cpus->cpus[i].props.core_id = core_id;
3691     }
3692     return machine->possible_cpus;
3693 }
3694 
3695 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3696                                 uint64_t *buid, hwaddr *pio,
3697                                 hwaddr *mmio32, hwaddr *mmio64,
3698                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3699 {
3700     /*
3701      * New-style PHB window placement.
3702      *
3703      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3704      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3705      * windows.
3706      *
3707      * Some guest kernels can't work with MMIO windows above 1<<46
3708      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3709      *
3710      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3711      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3712      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3713      * 1TiB 64-bit MMIO windows for each PHB.
3714      */
3715     const uint64_t base_buid = 0x800000020000000ULL;
3716 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3717                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3718     int i;
3719 
3720     /* Sanity check natural alignments */
3721     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3722     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3723     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3724     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3725     /* Sanity check bounds */
3726     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3727                       SPAPR_PCI_MEM32_WIN_SIZE);
3728     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3729                       SPAPR_PCI_MEM64_WIN_SIZE);
3730 
3731     if (index >= SPAPR_MAX_PHBS) {
3732         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3733                    SPAPR_MAX_PHBS - 1);
3734         return;
3735     }
3736 
3737     *buid = base_buid + index;
3738     for (i = 0; i < n_dma; ++i) {
3739         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3740     }
3741 
3742     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3743     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3744     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3745 }
3746 
3747 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3748 {
3749     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3750 
3751     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3752 }
3753 
3754 static void spapr_ics_resend(XICSFabric *dev)
3755 {
3756     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3757 
3758     ics_resend(spapr->ics);
3759 }
3760 
3761 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3762 {
3763     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3764 
3765     return cpu ? ICP(cpu->intc) : NULL;
3766 }
3767 
3768 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3769                                  Monitor *mon)
3770 {
3771     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3772     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3773 
3774     smc->irq->print_info(spapr, mon);
3775 }
3776 
3777 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3778 {
3779     return cpu->vcpu_id;
3780 }
3781 
3782 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3783 {
3784     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3785     int vcpu_id;
3786 
3787     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3788 
3789     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3790         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3791         error_append_hint(errp, "Adjust the number of cpus to %d "
3792                           "or try to raise the number of threads per core\n",
3793                           vcpu_id * smp_threads / spapr->vsmt);
3794         return;
3795     }
3796 
3797     cpu->vcpu_id = vcpu_id;
3798 }
3799 
3800 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3801 {
3802     CPUState *cs;
3803 
3804     CPU_FOREACH(cs) {
3805         PowerPCCPU *cpu = POWERPC_CPU(cs);
3806 
3807         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3808             return cpu;
3809         }
3810     }
3811 
3812     return NULL;
3813 }
3814 
3815 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3816 {
3817     MachineClass *mc = MACHINE_CLASS(oc);
3818     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3819     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3820     NMIClass *nc = NMI_CLASS(oc);
3821     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3822     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3823     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3824     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3825 
3826     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3827     mc->ignore_boot_device_suffixes = true;
3828 
3829     /*
3830      * We set up the default / latest behaviour here.  The class_init
3831      * functions for the specific versioned machine types can override
3832      * these details for backwards compatibility
3833      */
3834     mc->init = spapr_machine_init;
3835     mc->reset = spapr_machine_reset;
3836     mc->block_default_type = IF_SCSI;
3837     mc->max_cpus = 1024;
3838     mc->no_parallel = 1;
3839     mc->default_boot_order = "";
3840     mc->default_ram_size = 512 * MiB;
3841     mc->default_display = "std";
3842     mc->kvm_type = spapr_kvm_type;
3843     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3844     mc->pci_allow_0_address = true;
3845     assert(!mc->get_hotplug_handler);
3846     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3847     hc->pre_plug = spapr_machine_device_pre_plug;
3848     hc->plug = spapr_machine_device_plug;
3849     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3850     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3851     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3852     hc->unplug_request = spapr_machine_device_unplug_request;
3853     hc->unplug = spapr_machine_device_unplug;
3854 
3855     smc->dr_lmb_enabled = true;
3856     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3857     mc->has_hotpluggable_cpus = true;
3858     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3859     fwc->get_dev_path = spapr_get_fw_dev_path;
3860     nc->nmi_monitor_handler = spapr_nmi;
3861     smc->phb_placement = spapr_phb_placement;
3862     vhc->hypercall = emulate_spapr_hypercall;
3863     vhc->hpt_mask = spapr_hpt_mask;
3864     vhc->map_hptes = spapr_map_hptes;
3865     vhc->unmap_hptes = spapr_unmap_hptes;
3866     vhc->store_hpte = spapr_store_hpte;
3867     vhc->get_patbe = spapr_get_patbe;
3868     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3869     xic->ics_get = spapr_ics_get;
3870     xic->ics_resend = spapr_ics_resend;
3871     xic->icp_get = spapr_icp_get;
3872     ispc->print_info = spapr_pic_print_info;
3873     /* Force NUMA node memory size to be a multiple of
3874      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3875      * in which LMBs are represented and hot-added
3876      */
3877     mc->numa_mem_align_shift = 28;
3878 
3879     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3880     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3881     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3882     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3883     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3884     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3885     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
3886     spapr_caps_add_properties(smc, &error_abort);
3887     smc->irq = &spapr_irq_xics;
3888 }
3889 
3890 static const TypeInfo spapr_machine_info = {
3891     .name          = TYPE_SPAPR_MACHINE,
3892     .parent        = TYPE_MACHINE,
3893     .abstract      = true,
3894     .instance_size = sizeof(sPAPRMachineState),
3895     .instance_init = spapr_instance_init,
3896     .instance_finalize = spapr_machine_finalizefn,
3897     .class_size    = sizeof(sPAPRMachineClass),
3898     .class_init    = spapr_machine_class_init,
3899     .interfaces = (InterfaceInfo[]) {
3900         { TYPE_FW_PATH_PROVIDER },
3901         { TYPE_NMI },
3902         { TYPE_HOTPLUG_HANDLER },
3903         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3904         { TYPE_XICS_FABRIC },
3905         { TYPE_INTERRUPT_STATS_PROVIDER },
3906         { }
3907     },
3908 };
3909 
3910 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3911     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3912                                                     void *data)      \
3913     {                                                                \
3914         MachineClass *mc = MACHINE_CLASS(oc);                        \
3915         spapr_machine_##suffix##_class_options(mc);                  \
3916         if (latest) {                                                \
3917             mc->alias = "pseries";                                   \
3918             mc->is_default = 1;                                      \
3919         }                                                            \
3920     }                                                                \
3921     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3922     {                                                                \
3923         MachineState *machine = MACHINE(obj);                        \
3924         spapr_machine_##suffix##_instance_options(machine);          \
3925     }                                                                \
3926     static const TypeInfo spapr_machine_##suffix##_info = {          \
3927         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3928         .parent = TYPE_SPAPR_MACHINE,                                \
3929         .class_init = spapr_machine_##suffix##_class_init,           \
3930         .instance_init = spapr_machine_##suffix##_instance_init,     \
3931     };                                                               \
3932     static void spapr_machine_register_##suffix(void)                \
3933     {                                                                \
3934         type_register(&spapr_machine_##suffix##_info);               \
3935     }                                                                \
3936     type_init(spapr_machine_register_##suffix)
3937 
3938  /*
3939  * pseries-3.1
3940  */
3941 static void spapr_machine_3_1_instance_options(MachineState *machine)
3942 {
3943 }
3944 
3945 static void spapr_machine_3_1_class_options(MachineClass *mc)
3946 {
3947     /* Defaults for the latest behaviour inherited from the base class */
3948 }
3949 
3950 DEFINE_SPAPR_MACHINE(3_1, "3.1", true);
3951 
3952 /*
3953  * pseries-3.0
3954  */
3955 #define SPAPR_COMPAT_3_0                                              \
3956     HW_COMPAT_3_0
3957 
3958 static void spapr_machine_3_0_instance_options(MachineState *machine)
3959 {
3960     spapr_machine_3_1_instance_options(machine);
3961 }
3962 
3963 static void spapr_machine_3_0_class_options(MachineClass *mc)
3964 {
3965     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3966 
3967     spapr_machine_3_1_class_options(mc);
3968     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_3_0);
3969 
3970     smc->legacy_irq_allocation = true;
3971 }
3972 
3973 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
3974 
3975 /*
3976  * pseries-2.12
3977  */
3978 #define SPAPR_COMPAT_2_12                                              \
3979     HW_COMPAT_2_12                                                     \
3980     {                                                                  \
3981         .driver = TYPE_POWERPC_CPU,                                    \
3982         .property = "pre-3.0-migration",                               \
3983         .value    = "on",                                              \
3984     },                                                                 \
3985     {                                                                  \
3986         .driver = TYPE_SPAPR_CPU_CORE,                                 \
3987         .property = "pre-3.0-migration",                               \
3988         .value    = "on",                                              \
3989     },
3990 
3991 static void spapr_machine_2_12_instance_options(MachineState *machine)
3992 {
3993     spapr_machine_3_0_instance_options(machine);
3994 }
3995 
3996 static void spapr_machine_2_12_class_options(MachineClass *mc)
3997 {
3998     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3999 
4000     spapr_machine_3_0_class_options(mc);
4001     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4002 
4003     /* We depend on kvm_enabled() to choose a default value for the
4004      * hpt-max-page-size capability. Of course we can't do it here
4005      * because this is too early and the HW accelerator isn't initialzed
4006      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4007      */
4008     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4009 }
4010 
4011 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4012 
4013 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4014 {
4015     spapr_machine_2_12_instance_options(machine);
4016 }
4017 
4018 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4019 {
4020     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4021 
4022     spapr_machine_2_12_class_options(mc);
4023     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4024     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4025     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4026 }
4027 
4028 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4029 
4030 /*
4031  * pseries-2.11
4032  */
4033 #define SPAPR_COMPAT_2_11                                              \
4034     HW_COMPAT_2_11
4035 
4036 static void spapr_machine_2_11_instance_options(MachineState *machine)
4037 {
4038     spapr_machine_2_12_instance_options(machine);
4039 }
4040 
4041 static void spapr_machine_2_11_class_options(MachineClass *mc)
4042 {
4043     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4044 
4045     spapr_machine_2_12_class_options(mc);
4046     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4047     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4048 }
4049 
4050 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4051 
4052 /*
4053  * pseries-2.10
4054  */
4055 #define SPAPR_COMPAT_2_10                                              \
4056     HW_COMPAT_2_10
4057 
4058 static void spapr_machine_2_10_instance_options(MachineState *machine)
4059 {
4060     spapr_machine_2_11_instance_options(machine);
4061 }
4062 
4063 static void spapr_machine_2_10_class_options(MachineClass *mc)
4064 {
4065     spapr_machine_2_11_class_options(mc);
4066     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4067 }
4068 
4069 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4070 
4071 /*
4072  * pseries-2.9
4073  */
4074 #define SPAPR_COMPAT_2_9                                               \
4075     HW_COMPAT_2_9                                                      \
4076     {                                                                  \
4077         .driver = TYPE_POWERPC_CPU,                                    \
4078         .property = "pre-2.10-migration",                              \
4079         .value    = "on",                                              \
4080     },                                                                 \
4081 
4082 static void spapr_machine_2_9_instance_options(MachineState *machine)
4083 {
4084     spapr_machine_2_10_instance_options(machine);
4085 }
4086 
4087 static void spapr_machine_2_9_class_options(MachineClass *mc)
4088 {
4089     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4090 
4091     spapr_machine_2_10_class_options(mc);
4092     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4093     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4094     smc->pre_2_10_has_unused_icps = true;
4095     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4096 }
4097 
4098 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4099 
4100 /*
4101  * pseries-2.8
4102  */
4103 #define SPAPR_COMPAT_2_8                                        \
4104     HW_COMPAT_2_8                                               \
4105     {                                                           \
4106         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4107         .property = "pcie-extended-configuration-space",        \
4108         .value    = "off",                                      \
4109     },
4110 
4111 static void spapr_machine_2_8_instance_options(MachineState *machine)
4112 {
4113     spapr_machine_2_9_instance_options(machine);
4114 }
4115 
4116 static void spapr_machine_2_8_class_options(MachineClass *mc)
4117 {
4118     spapr_machine_2_9_class_options(mc);
4119     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4120     mc->numa_mem_align_shift = 23;
4121 }
4122 
4123 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4124 
4125 /*
4126  * pseries-2.7
4127  */
4128 #define SPAPR_COMPAT_2_7                            \
4129     HW_COMPAT_2_7                                   \
4130     {                                               \
4131         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4132         .property = "mem_win_size",                 \
4133         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4134     },                                              \
4135     {                                               \
4136         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4137         .property = "mem64_win_size",               \
4138         .value    = "0",                            \
4139     },                                              \
4140     {                                               \
4141         .driver = TYPE_POWERPC_CPU,                 \
4142         .property = "pre-2.8-migration",            \
4143         .value    = "on",                           \
4144     },                                              \
4145     {                                               \
4146         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4147         .property = "pre-2.8-migration",            \
4148         .value    = "on",                           \
4149     },
4150 
4151 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4152                               uint64_t *buid, hwaddr *pio,
4153                               hwaddr *mmio32, hwaddr *mmio64,
4154                               unsigned n_dma, uint32_t *liobns, Error **errp)
4155 {
4156     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4157     const uint64_t base_buid = 0x800000020000000ULL;
4158     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4159     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4160     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4161     const uint32_t max_index = 255;
4162     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4163 
4164     uint64_t ram_top = MACHINE(spapr)->ram_size;
4165     hwaddr phb0_base, phb_base;
4166     int i;
4167 
4168     /* Do we have device memory? */
4169     if (MACHINE(spapr)->maxram_size > ram_top) {
4170         /* Can't just use maxram_size, because there may be an
4171          * alignment gap between normal and device memory regions
4172          */
4173         ram_top = MACHINE(spapr)->device_memory->base +
4174             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4175     }
4176 
4177     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4178 
4179     if (index > max_index) {
4180         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4181                    max_index);
4182         return;
4183     }
4184 
4185     *buid = base_buid + index;
4186     for (i = 0; i < n_dma; ++i) {
4187         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4188     }
4189 
4190     phb_base = phb0_base + index * phb_spacing;
4191     *pio = phb_base + pio_offset;
4192     *mmio32 = phb_base + mmio_offset;
4193     /*
4194      * We don't set the 64-bit MMIO window, relying on the PHB's
4195      * fallback behaviour of automatically splitting a large "32-bit"
4196      * window into contiguous 32-bit and 64-bit windows
4197      */
4198 }
4199 
4200 static void spapr_machine_2_7_instance_options(MachineState *machine)
4201 {
4202     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4203 
4204     spapr_machine_2_8_instance_options(machine);
4205     spapr->use_hotplug_event_source = false;
4206 }
4207 
4208 static void spapr_machine_2_7_class_options(MachineClass *mc)
4209 {
4210     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4211 
4212     spapr_machine_2_8_class_options(mc);
4213     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4214     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4215     smc->phb_placement = phb_placement_2_7;
4216 }
4217 
4218 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4219 
4220 /*
4221  * pseries-2.6
4222  */
4223 #define SPAPR_COMPAT_2_6 \
4224     HW_COMPAT_2_6 \
4225     { \
4226         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4227         .property = "ddw",\
4228         .value    = stringify(off),\
4229     },
4230 
4231 static void spapr_machine_2_6_instance_options(MachineState *machine)
4232 {
4233     spapr_machine_2_7_instance_options(machine);
4234 }
4235 
4236 static void spapr_machine_2_6_class_options(MachineClass *mc)
4237 {
4238     spapr_machine_2_7_class_options(mc);
4239     mc->has_hotpluggable_cpus = false;
4240     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4241 }
4242 
4243 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4244 
4245 /*
4246  * pseries-2.5
4247  */
4248 #define SPAPR_COMPAT_2_5 \
4249     HW_COMPAT_2_5 \
4250     { \
4251         .driver   = "spapr-vlan", \
4252         .property = "use-rx-buffer-pools", \
4253         .value    = "off", \
4254     },
4255 
4256 static void spapr_machine_2_5_instance_options(MachineState *machine)
4257 {
4258     spapr_machine_2_6_instance_options(machine);
4259 }
4260 
4261 static void spapr_machine_2_5_class_options(MachineClass *mc)
4262 {
4263     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4264 
4265     spapr_machine_2_6_class_options(mc);
4266     smc->use_ohci_by_default = true;
4267     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4268 }
4269 
4270 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4271 
4272 /*
4273  * pseries-2.4
4274  */
4275 #define SPAPR_COMPAT_2_4 \
4276         HW_COMPAT_2_4
4277 
4278 static void spapr_machine_2_4_instance_options(MachineState *machine)
4279 {
4280     spapr_machine_2_5_instance_options(machine);
4281 }
4282 
4283 static void spapr_machine_2_4_class_options(MachineClass *mc)
4284 {
4285     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4286 
4287     spapr_machine_2_5_class_options(mc);
4288     smc->dr_lmb_enabled = false;
4289     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4290 }
4291 
4292 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4293 
4294 /*
4295  * pseries-2.3
4296  */
4297 #define SPAPR_COMPAT_2_3 \
4298         HW_COMPAT_2_3 \
4299         {\
4300             .driver   = "spapr-pci-host-bridge",\
4301             .property = "dynamic-reconfiguration",\
4302             .value    = "off",\
4303         },
4304 
4305 static void spapr_machine_2_3_instance_options(MachineState *machine)
4306 {
4307     spapr_machine_2_4_instance_options(machine);
4308 }
4309 
4310 static void spapr_machine_2_3_class_options(MachineClass *mc)
4311 {
4312     spapr_machine_2_4_class_options(mc);
4313     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4314 }
4315 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4316 
4317 /*
4318  * pseries-2.2
4319  */
4320 
4321 #define SPAPR_COMPAT_2_2 \
4322         HW_COMPAT_2_2 \
4323         {\
4324             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4325             .property = "mem_win_size",\
4326             .value    = "0x20000000",\
4327         },
4328 
4329 static void spapr_machine_2_2_instance_options(MachineState *machine)
4330 {
4331     spapr_machine_2_3_instance_options(machine);
4332     machine->suppress_vmdesc = true;
4333 }
4334 
4335 static void spapr_machine_2_2_class_options(MachineClass *mc)
4336 {
4337     spapr_machine_2_3_class_options(mc);
4338     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4339 }
4340 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4341 
4342 /*
4343  * pseries-2.1
4344  */
4345 #define SPAPR_COMPAT_2_1 \
4346         HW_COMPAT_2_1
4347 
4348 static void spapr_machine_2_1_instance_options(MachineState *machine)
4349 {
4350     spapr_machine_2_2_instance_options(machine);
4351 }
4352 
4353 static void spapr_machine_2_1_class_options(MachineClass *mc)
4354 {
4355     spapr_machine_2_2_class_options(mc);
4356     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4357 }
4358 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4359 
4360 static void spapr_machine_register_types(void)
4361 {
4362     type_register_static(&spapr_machine_info);
4363 }
4364 
4365 type_init(spapr_machine_register_types)
4366