xref: /openbmc/qemu/hw/ppc/spapr.c (revision 464e447a)
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 int spapr_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     g_free(rev);
611 }
612 
613 static int spapr_rng_populate_dt(void *fdt)
614 {
615     int node;
616     int ret;
617 
618     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
619     if (node <= 0) {
620         return -1;
621     }
622     ret = fdt_setprop_string(fdt, node, "device_type",
623                              "ibm,platform-facilities");
624     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
625     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
626 
627     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
628     if (node <= 0) {
629         return -1;
630     }
631     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
632 
633     return ret ? -1 : 0;
634 }
635 
636 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
637 {
638     MemoryDeviceInfoList *info;
639 
640     for (info = list; info; info = info->next) {
641         MemoryDeviceInfo *value = info->value;
642 
643         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
644             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
645 
646             if (addr >= pcdimm_info->addr &&
647                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
648                 return pcdimm_info->node;
649             }
650         }
651     }
652 
653     return -1;
654 }
655 
656 struct sPAPRDrconfCellV2 {
657      uint32_t seq_lmbs;
658      uint64_t base_addr;
659      uint32_t drc_index;
660      uint32_t aa_index;
661      uint32_t flags;
662 } QEMU_PACKED;
663 
664 typedef struct DrconfCellQueue {
665     struct sPAPRDrconfCellV2 cell;
666     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
667 } DrconfCellQueue;
668 
669 static DrconfCellQueue *
670 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
671                       uint32_t drc_index, uint32_t aa_index,
672                       uint32_t flags)
673 {
674     DrconfCellQueue *elem;
675 
676     elem = g_malloc0(sizeof(*elem));
677     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
678     elem->cell.base_addr = cpu_to_be64(base_addr);
679     elem->cell.drc_index = cpu_to_be32(drc_index);
680     elem->cell.aa_index = cpu_to_be32(aa_index);
681     elem->cell.flags = cpu_to_be32(flags);
682 
683     return elem;
684 }
685 
686 /* ibm,dynamic-memory-v2 */
687 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
688                                    int offset, MemoryDeviceInfoList *dimms)
689 {
690     MachineState *machine = MACHINE(spapr);
691     uint8_t *int_buf, *cur_index, buf_len;
692     int ret;
693     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
694     uint64_t addr, cur_addr, size;
695     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
696     uint64_t mem_end = machine->device_memory->base +
697                        memory_region_size(&machine->device_memory->mr);
698     uint32_t node, nr_entries = 0;
699     sPAPRDRConnector *drc;
700     DrconfCellQueue *elem, *next;
701     MemoryDeviceInfoList *info;
702     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
703         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
704 
705     /* Entry to cover RAM and the gap area */
706     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
707                                  SPAPR_LMB_FLAGS_RESERVED |
708                                  SPAPR_LMB_FLAGS_DRC_INVALID);
709     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
710     nr_entries++;
711 
712     cur_addr = machine->device_memory->base;
713     for (info = dimms; info; info = info->next) {
714         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
715 
716         addr = di->addr;
717         size = di->size;
718         node = di->node;
719 
720         /* Entry for hot-pluggable area */
721         if (cur_addr < addr) {
722             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
723             g_assert(drc);
724             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
725                                          cur_addr, spapr_drc_index(drc), -1, 0);
726             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
727             nr_entries++;
728         }
729 
730         /* Entry for DIMM */
731         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
732         g_assert(drc);
733         elem = spapr_get_drconf_cell(size / lmb_size, addr,
734                                      spapr_drc_index(drc), node,
735                                      SPAPR_LMB_FLAGS_ASSIGNED);
736         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
737         nr_entries++;
738         cur_addr = addr + size;
739     }
740 
741     /* Entry for remaining hotpluggable area */
742     if (cur_addr < mem_end) {
743         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
744         g_assert(drc);
745         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
746                                      cur_addr, spapr_drc_index(drc), -1, 0);
747         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
748         nr_entries++;
749     }
750 
751     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
752     int_buf = cur_index = g_malloc0(buf_len);
753     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
754     cur_index += sizeof(nr_entries);
755 
756     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
757         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
758         cur_index += sizeof(elem->cell);
759         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
760         g_free(elem);
761     }
762 
763     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
764     g_free(int_buf);
765     if (ret < 0) {
766         return -1;
767     }
768     return 0;
769 }
770 
771 /* ibm,dynamic-memory */
772 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
773                                    int offset, MemoryDeviceInfoList *dimms)
774 {
775     MachineState *machine = MACHINE(spapr);
776     int i, ret;
777     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
778     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
779     uint32_t nr_lmbs = (machine->device_memory->base +
780                        memory_region_size(&machine->device_memory->mr)) /
781                        lmb_size;
782     uint32_t *int_buf, *cur_index, buf_len;
783 
784     /*
785      * Allocate enough buffer size to fit in ibm,dynamic-memory
786      */
787     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
788     cur_index = int_buf = g_malloc0(buf_len);
789     int_buf[0] = cpu_to_be32(nr_lmbs);
790     cur_index++;
791     for (i = 0; i < nr_lmbs; i++) {
792         uint64_t addr = i * lmb_size;
793         uint32_t *dynamic_memory = cur_index;
794 
795         if (i >= device_lmb_start) {
796             sPAPRDRConnector *drc;
797 
798             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
799             g_assert(drc);
800 
801             dynamic_memory[0] = cpu_to_be32(addr >> 32);
802             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
803             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
804             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
805             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
806             if (memory_region_present(get_system_memory(), addr)) {
807                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
808             } else {
809                 dynamic_memory[5] = cpu_to_be32(0);
810             }
811         } else {
812             /*
813              * LMB information for RMA, boot time RAM and gap b/n RAM and
814              * device memory region -- all these are marked as reserved
815              * and as having no valid DRC.
816              */
817             dynamic_memory[0] = cpu_to_be32(addr >> 32);
818             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
819             dynamic_memory[2] = cpu_to_be32(0);
820             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
821             dynamic_memory[4] = cpu_to_be32(-1);
822             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
823                                             SPAPR_LMB_FLAGS_DRC_INVALID);
824         }
825 
826         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
827     }
828     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
829     g_free(int_buf);
830     if (ret < 0) {
831         return -1;
832     }
833     return 0;
834 }
835 
836 /*
837  * Adds ibm,dynamic-reconfiguration-memory node.
838  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
839  * of this device tree node.
840  */
841 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
842 {
843     MachineState *machine = MACHINE(spapr);
844     int ret, i, offset;
845     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
846     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
847     uint32_t *int_buf, *cur_index, buf_len;
848     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
849     MemoryDeviceInfoList *dimms = NULL;
850 
851     /*
852      * Don't create the node if there is no device memory
853      */
854     if (machine->ram_size == machine->maxram_size) {
855         return 0;
856     }
857 
858     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
859 
860     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
861                     sizeof(prop_lmb_size));
862     if (ret < 0) {
863         return ret;
864     }
865 
866     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
867     if (ret < 0) {
868         return ret;
869     }
870 
871     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
872     if (ret < 0) {
873         return ret;
874     }
875 
876     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
877     dimms = qmp_memory_device_list();
878     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
879         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
880     } else {
881         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
882     }
883     qapi_free_MemoryDeviceInfoList(dimms);
884 
885     if (ret < 0) {
886         return ret;
887     }
888 
889     /* ibm,associativity-lookup-arrays */
890     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
891     cur_index = int_buf = g_malloc0(buf_len);
892     int_buf[0] = cpu_to_be32(nr_nodes);
893     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
894     cur_index += 2;
895     for (i = 0; i < nr_nodes; i++) {
896         uint32_t associativity[] = {
897             cpu_to_be32(0x0),
898             cpu_to_be32(0x0),
899             cpu_to_be32(0x0),
900             cpu_to_be32(i)
901         };
902         memcpy(cur_index, associativity, sizeof(associativity));
903         cur_index += 4;
904     }
905     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
906             (cur_index - int_buf) * sizeof(uint32_t));
907     g_free(int_buf);
908 
909     return ret;
910 }
911 
912 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
913                                 sPAPROptionVector *ov5_updates)
914 {
915     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
916     int ret = 0, offset;
917 
918     /* Generate ibm,dynamic-reconfiguration-memory node if required */
919     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
920         g_assert(smc->dr_lmb_enabled);
921         ret = spapr_populate_drconf_memory(spapr, fdt);
922         if (ret) {
923             goto out;
924         }
925     }
926 
927     offset = fdt_path_offset(fdt, "/chosen");
928     if (offset < 0) {
929         offset = fdt_add_subnode(fdt, 0, "chosen");
930         if (offset < 0) {
931             return offset;
932         }
933     }
934     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
935                                  "ibm,architecture-vec-5");
936 
937 out:
938     return ret;
939 }
940 
941 static bool spapr_hotplugged_dev_before_cas(void)
942 {
943     Object *drc_container, *obj;
944     ObjectProperty *prop;
945     ObjectPropertyIterator iter;
946 
947     drc_container = container_get(object_get_root(), "/dr-connector");
948     object_property_iter_init(&iter, drc_container);
949     while ((prop = object_property_iter_next(&iter))) {
950         if (!strstart(prop->type, "link<", NULL)) {
951             continue;
952         }
953         obj = object_property_get_link(drc_container, prop->name, NULL);
954         if (spapr_drc_needed(obj)) {
955             return true;
956         }
957     }
958     return false;
959 }
960 
961 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
962                                  target_ulong addr, target_ulong size,
963                                  sPAPROptionVector *ov5_updates)
964 {
965     void *fdt, *fdt_skel;
966     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
967 
968     if (spapr_hotplugged_dev_before_cas()) {
969         return 1;
970     }
971 
972     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
973         error_report("SLOF provided an unexpected CAS buffer size "
974                      TARGET_FMT_lu " (min: %zu, max: %u)",
975                      size, sizeof(hdr), FW_MAX_SIZE);
976         exit(EXIT_FAILURE);
977     }
978 
979     size -= sizeof(hdr);
980 
981     /* Create skeleton */
982     fdt_skel = g_malloc0(size);
983     _FDT((fdt_create(fdt_skel, size)));
984     _FDT((fdt_finish_reservemap(fdt_skel)));
985     _FDT((fdt_begin_node(fdt_skel, "")));
986     _FDT((fdt_end_node(fdt_skel)));
987     _FDT((fdt_finish(fdt_skel)));
988     fdt = g_malloc0(size);
989     _FDT((fdt_open_into(fdt_skel, fdt, size)));
990     g_free(fdt_skel);
991 
992     /* Fixup cpu nodes */
993     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
994 
995     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
996         return -1;
997     }
998 
999     /* Pack resulting tree */
1000     _FDT((fdt_pack(fdt)));
1001 
1002     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1003         trace_spapr_cas_failed(size);
1004         return -1;
1005     }
1006 
1007     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1008     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1009     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1010     g_free(fdt);
1011 
1012     return 0;
1013 }
1014 
1015 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1016 {
1017     int rtas;
1018     GString *hypertas = g_string_sized_new(256);
1019     GString *qemu_hypertas = g_string_sized_new(256);
1020     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1021     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1022         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1023     uint32_t lrdr_capacity[] = {
1024         cpu_to_be32(max_device_addr >> 32),
1025         cpu_to_be32(max_device_addr & 0xffffffff),
1026         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1027         cpu_to_be32(max_cpus / smp_threads),
1028     };
1029     uint32_t maxdomains[] = {
1030         cpu_to_be32(4),
1031         cpu_to_be32(0),
1032         cpu_to_be32(0),
1033         cpu_to_be32(0),
1034         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes : 1),
1035     };
1036 
1037     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1038 
1039     /* hypertas */
1040     add_str(hypertas, "hcall-pft");
1041     add_str(hypertas, "hcall-term");
1042     add_str(hypertas, "hcall-dabr");
1043     add_str(hypertas, "hcall-interrupt");
1044     add_str(hypertas, "hcall-tce");
1045     add_str(hypertas, "hcall-vio");
1046     add_str(hypertas, "hcall-splpar");
1047     add_str(hypertas, "hcall-bulk");
1048     add_str(hypertas, "hcall-set-mode");
1049     add_str(hypertas, "hcall-sprg0");
1050     add_str(hypertas, "hcall-copy");
1051     add_str(hypertas, "hcall-debug");
1052     add_str(qemu_hypertas, "hcall-memop1");
1053 
1054     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1055         add_str(hypertas, "hcall-multi-tce");
1056     }
1057 
1058     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1059         add_str(hypertas, "hcall-hpt-resize");
1060     }
1061 
1062     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1063                      hypertas->str, hypertas->len));
1064     g_string_free(hypertas, TRUE);
1065     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1066                      qemu_hypertas->str, qemu_hypertas->len));
1067     g_string_free(qemu_hypertas, TRUE);
1068 
1069     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1070                      refpoints, sizeof(refpoints)));
1071 
1072     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1073                      maxdomains, sizeof(maxdomains)));
1074 
1075     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1076                           RTAS_ERROR_LOG_MAX));
1077     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1078                           RTAS_EVENT_SCAN_RATE));
1079 
1080     g_assert(msi_nonbroken);
1081     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1082 
1083     /*
1084      * According to PAPR, rtas ibm,os-term does not guarantee a return
1085      * back to the guest cpu.
1086      *
1087      * While an additional ibm,extended-os-term property indicates
1088      * that rtas call return will always occur. Set this property.
1089      */
1090     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1091 
1092     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1093                      lrdr_capacity, sizeof(lrdr_capacity)));
1094 
1095     spapr_dt_rtas_tokens(fdt, rtas);
1096 }
1097 
1098 /*
1099  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1100  * and the XIVE features that the guest may request and thus the valid
1101  * values for bytes 23..26 of option vector 5:
1102  */
1103 static void spapr_dt_ov5_platform_support(sPAPRMachineState *spapr, void *fdt,
1104                                           int chosen)
1105 {
1106     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1107 
1108     char val[2 * 4] = {
1109         23, spapr->irq->ov5, /* Xive mode. */
1110         24, 0x00, /* Hash/Radix, filled in below. */
1111         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1112         26, 0x40, /* Radix options: GTSE == yes. */
1113     };
1114 
1115     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1116                           first_ppc_cpu->compat_pvr)) {
1117         /*
1118          * If we're in a pre POWER9 compat mode then the guest should
1119          * do hash and use the legacy interrupt mode
1120          */
1121         val[1] = 0x00; /* XICS */
1122         val[3] = 0x00; /* Hash */
1123     } else if (kvm_enabled()) {
1124         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1125             val[3] = 0x80; /* OV5_MMU_BOTH */
1126         } else if (kvmppc_has_cap_mmu_radix()) {
1127             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1128         } else {
1129             val[3] = 0x00; /* Hash */
1130         }
1131     } else {
1132         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1133         val[3] = 0xC0;
1134     }
1135     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1136                      val, sizeof(val)));
1137 }
1138 
1139 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1140 {
1141     MachineState *machine = MACHINE(spapr);
1142     int chosen;
1143     const char *boot_device = machine->boot_order;
1144     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1145     size_t cb = 0;
1146     char *bootlist = get_boot_devices_list(&cb);
1147 
1148     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1149 
1150     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1151     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1152                           spapr->initrd_base));
1153     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1154                           spapr->initrd_base + spapr->initrd_size));
1155 
1156     if (spapr->kernel_size) {
1157         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1158                               cpu_to_be64(spapr->kernel_size) };
1159 
1160         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1161                          &kprop, sizeof(kprop)));
1162         if (spapr->kernel_le) {
1163             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1164         }
1165     }
1166     if (boot_menu) {
1167         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1168     }
1169     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1170     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1171     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1172 
1173     if (cb && bootlist) {
1174         int i;
1175 
1176         for (i = 0; i < cb; i++) {
1177             if (bootlist[i] == '\n') {
1178                 bootlist[i] = ' ';
1179             }
1180         }
1181         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1182     }
1183 
1184     if (boot_device && strlen(boot_device)) {
1185         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1186     }
1187 
1188     if (!spapr->has_graphics && stdout_path) {
1189         /*
1190          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1191          * kernel. New platforms should only use the "stdout-path" property. Set
1192          * the new property and continue using older property to remain
1193          * compatible with the existing firmware.
1194          */
1195         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1196         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1197     }
1198 
1199     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1200 
1201     g_free(stdout_path);
1202     g_free(bootlist);
1203 }
1204 
1205 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1206 {
1207     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1208      * KVM to work under pHyp with some guest co-operation */
1209     int hypervisor;
1210     uint8_t hypercall[16];
1211 
1212     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1213     /* indicate KVM hypercall interface */
1214     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1215     if (kvmppc_has_cap_fixup_hcalls()) {
1216         /*
1217          * Older KVM versions with older guest kernels were broken
1218          * with the magic page, don't allow the guest to map it.
1219          */
1220         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1221                                   sizeof(hypercall))) {
1222             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1223                              hypercall, sizeof(hypercall)));
1224         }
1225     }
1226 }
1227 
1228 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1229                              hwaddr rtas_addr,
1230                              hwaddr rtas_size)
1231 {
1232     MachineState *machine = MACHINE(spapr);
1233     MachineClass *mc = MACHINE_GET_CLASS(machine);
1234     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1235     int ret;
1236     void *fdt;
1237     sPAPRPHBState *phb;
1238     char *buf;
1239 
1240     fdt = g_malloc0(FDT_MAX_SIZE);
1241     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1242 
1243     /* Root node */
1244     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1245     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1246     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1247 
1248     /*
1249      * Add info to guest to indentify which host is it being run on
1250      * and what is the uuid of the guest
1251      */
1252     if (kvmppc_get_host_model(&buf)) {
1253         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1254         g_free(buf);
1255     }
1256     if (kvmppc_get_host_serial(&buf)) {
1257         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1258         g_free(buf);
1259     }
1260 
1261     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1262 
1263     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1264     if (qemu_uuid_set) {
1265         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1266     }
1267     g_free(buf);
1268 
1269     if (qemu_get_vm_name()) {
1270         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1271                                 qemu_get_vm_name()));
1272     }
1273 
1274     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1275     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1276 
1277     /* /interrupt controller */
1278     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1279                           PHANDLE_XICP);
1280 
1281     ret = spapr_populate_memory(spapr, fdt);
1282     if (ret < 0) {
1283         error_report("couldn't setup memory nodes in fdt");
1284         exit(1);
1285     }
1286 
1287     /* /vdevice */
1288     spapr_dt_vdevice(spapr->vio_bus, fdt);
1289 
1290     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1291         ret = spapr_rng_populate_dt(fdt);
1292         if (ret < 0) {
1293             error_report("could not set up rng device in the fdt");
1294             exit(1);
1295         }
1296     }
1297 
1298     QLIST_FOREACH(phb, &spapr->phbs, list) {
1299         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt,
1300                                     spapr->irq->nr_msis);
1301         if (ret < 0) {
1302             error_report("couldn't setup PCI devices in fdt");
1303             exit(1);
1304         }
1305     }
1306 
1307     /* cpus */
1308     spapr_populate_cpus_dt_node(fdt, spapr);
1309 
1310     if (smc->dr_lmb_enabled) {
1311         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1312     }
1313 
1314     if (mc->has_hotpluggable_cpus) {
1315         int offset = fdt_path_offset(fdt, "/cpus");
1316         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1317                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1318         if (ret < 0) {
1319             error_report("Couldn't set up CPU DR device tree properties");
1320             exit(1);
1321         }
1322     }
1323 
1324     /* /event-sources */
1325     spapr_dt_events(spapr, fdt);
1326 
1327     /* /rtas */
1328     spapr_dt_rtas(spapr, fdt);
1329 
1330     /* /chosen */
1331     spapr_dt_chosen(spapr, fdt);
1332 
1333     /* /hypervisor */
1334     if (kvm_enabled()) {
1335         spapr_dt_hypervisor(spapr, fdt);
1336     }
1337 
1338     /* Build memory reserve map */
1339     if (spapr->kernel_size) {
1340         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1341     }
1342     if (spapr->initrd_size) {
1343         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1344     }
1345 
1346     /* ibm,client-architecture-support updates */
1347     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1348     if (ret < 0) {
1349         error_report("couldn't setup CAS properties fdt");
1350         exit(1);
1351     }
1352 
1353     return fdt;
1354 }
1355 
1356 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1357 {
1358     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1359 }
1360 
1361 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1362                                     PowerPCCPU *cpu)
1363 {
1364     CPUPPCState *env = &cpu->env;
1365 
1366     /* The TCG path should also be holding the BQL at this point */
1367     g_assert(qemu_mutex_iothread_locked());
1368 
1369     if (msr_pr) {
1370         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1371         env->gpr[3] = H_PRIVILEGE;
1372     } else {
1373         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1374     }
1375 }
1376 
1377 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1378 {
1379     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1380 
1381     return spapr->patb_entry;
1382 }
1383 
1384 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1385 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1386 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1387 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1388 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1389 
1390 /*
1391  * Get the fd to access the kernel htab, re-opening it if necessary
1392  */
1393 static int get_htab_fd(sPAPRMachineState *spapr)
1394 {
1395     Error *local_err = NULL;
1396 
1397     if (spapr->htab_fd >= 0) {
1398         return spapr->htab_fd;
1399     }
1400 
1401     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1402     if (spapr->htab_fd < 0) {
1403         error_report_err(local_err);
1404     }
1405 
1406     return spapr->htab_fd;
1407 }
1408 
1409 void close_htab_fd(sPAPRMachineState *spapr)
1410 {
1411     if (spapr->htab_fd >= 0) {
1412         close(spapr->htab_fd);
1413     }
1414     spapr->htab_fd = -1;
1415 }
1416 
1417 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1418 {
1419     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1420 
1421     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1422 }
1423 
1424 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1425 {
1426     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1427 
1428     assert(kvm_enabled());
1429 
1430     if (!spapr->htab) {
1431         return 0;
1432     }
1433 
1434     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1435 }
1436 
1437 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1438                                                 hwaddr ptex, int n)
1439 {
1440     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1441     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1442 
1443     if (!spapr->htab) {
1444         /*
1445          * HTAB is controlled by KVM. Fetch into temporary buffer
1446          */
1447         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1448         kvmppc_read_hptes(hptes, ptex, n);
1449         return hptes;
1450     }
1451 
1452     /*
1453      * HTAB is controlled by QEMU. Just point to the internally
1454      * accessible PTEG.
1455      */
1456     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1457 }
1458 
1459 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1460                               const ppc_hash_pte64_t *hptes,
1461                               hwaddr ptex, int n)
1462 {
1463     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1464 
1465     if (!spapr->htab) {
1466         g_free((void *)hptes);
1467     }
1468 
1469     /* Nothing to do for qemu managed HPT */
1470 }
1471 
1472 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1473                              uint64_t pte0, uint64_t pte1)
1474 {
1475     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1476     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1477 
1478     if (!spapr->htab) {
1479         kvmppc_write_hpte(ptex, pte0, pte1);
1480     } else {
1481         stq_p(spapr->htab + offset, pte0);
1482         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1483     }
1484 }
1485 
1486 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1487 {
1488     int shift;
1489 
1490     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1491      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1492      * that's much more than is needed for Linux guests */
1493     shift = ctz64(pow2ceil(ramsize)) - 7;
1494     shift = MAX(shift, 18); /* Minimum architected size */
1495     shift = MIN(shift, 46); /* Maximum architected size */
1496     return shift;
1497 }
1498 
1499 void spapr_free_hpt(sPAPRMachineState *spapr)
1500 {
1501     g_free(spapr->htab);
1502     spapr->htab = NULL;
1503     spapr->htab_shift = 0;
1504     close_htab_fd(spapr);
1505 }
1506 
1507 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1508                           Error **errp)
1509 {
1510     long rc;
1511 
1512     /* Clean up any HPT info from a previous boot */
1513     spapr_free_hpt(spapr);
1514 
1515     rc = kvmppc_reset_htab(shift);
1516     if (rc < 0) {
1517         /* kernel-side HPT needed, but couldn't allocate one */
1518         error_setg_errno(errp, errno,
1519                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1520                          shift);
1521         /* This is almost certainly fatal, but if the caller really
1522          * wants to carry on with shift == 0, it's welcome to try */
1523     } else if (rc > 0) {
1524         /* kernel-side HPT allocated */
1525         if (rc != shift) {
1526             error_setg(errp,
1527                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1528                        shift, rc);
1529         }
1530 
1531         spapr->htab_shift = shift;
1532         spapr->htab = NULL;
1533     } else {
1534         /* kernel-side HPT not needed, allocate in userspace instead */
1535         size_t size = 1ULL << shift;
1536         int i;
1537 
1538         spapr->htab = qemu_memalign(size, size);
1539         if (!spapr->htab) {
1540             error_setg_errno(errp, errno,
1541                              "Could not allocate HPT of order %d", shift);
1542             return;
1543         }
1544 
1545         memset(spapr->htab, 0, size);
1546         spapr->htab_shift = shift;
1547 
1548         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1549             DIRTY_HPTE(HPTE(spapr->htab, i));
1550         }
1551     }
1552     /* We're setting up a hash table, so that means we're not radix */
1553     spapr->patb_entry = 0;
1554 }
1555 
1556 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1557 {
1558     int hpt_shift;
1559 
1560     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1561         || (spapr->cas_reboot
1562             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1563         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1564     } else {
1565         uint64_t current_ram_size;
1566 
1567         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1568         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1569     }
1570     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1571 
1572     if (spapr->vrma_adjust) {
1573         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1574                                           spapr->htab_shift);
1575     }
1576 }
1577 
1578 static int spapr_reset_drcs(Object *child, void *opaque)
1579 {
1580     sPAPRDRConnector *drc =
1581         (sPAPRDRConnector *) object_dynamic_cast(child,
1582                                                  TYPE_SPAPR_DR_CONNECTOR);
1583 
1584     if (drc) {
1585         spapr_drc_reset(drc);
1586     }
1587 
1588     return 0;
1589 }
1590 
1591 static void spapr_machine_reset(void)
1592 {
1593     MachineState *machine = MACHINE(qdev_get_machine());
1594     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1595     PowerPCCPU *first_ppc_cpu;
1596     uint32_t rtas_limit;
1597     hwaddr rtas_addr, fdt_addr;
1598     void *fdt;
1599     int rc;
1600 
1601     spapr_caps_apply(spapr);
1602 
1603     first_ppc_cpu = POWERPC_CPU(first_cpu);
1604     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1605         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1606                               spapr->max_compat_pvr)) {
1607         /* If using KVM with radix mode available, VCPUs can be started
1608          * without a HPT because KVM will start them in radix mode.
1609          * Set the GR bit in PATB so that we know there is no HPT. */
1610         spapr->patb_entry = PATBE1_GR;
1611     } else {
1612         spapr_setup_hpt_and_vrma(spapr);
1613     }
1614 
1615     /* if this reset wasn't generated by CAS, we should reset our
1616      * negotiated options and start from scratch */
1617     if (!spapr->cas_reboot) {
1618         spapr_ovec_cleanup(spapr->ov5_cas);
1619         spapr->ov5_cas = spapr_ovec_new();
1620 
1621         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1622     }
1623 
1624     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1625         spapr_irq_msi_reset(spapr);
1626     }
1627 
1628     qemu_devices_reset();
1629 
1630     /*
1631      * This is fixing some of the default configuration of the XIVE
1632      * devices. To be called after the reset of the machine devices.
1633      */
1634     spapr_irq_reset(spapr, &error_fatal);
1635 
1636     /* DRC reset may cause a device to be unplugged. This will cause troubles
1637      * if this device is used by another device (eg, a running vhost backend
1638      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1639      * situations, we reset DRCs after all devices have been reset.
1640      */
1641     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1642 
1643     spapr_clear_pending_events(spapr);
1644 
1645     /*
1646      * We place the device tree and RTAS just below either the top of the RMA,
1647      * or just below 2GB, whichever is lowere, so that it can be
1648      * processed with 32-bit real mode code if necessary
1649      */
1650     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1651     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1652     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1653 
1654     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1655 
1656     spapr_load_rtas(spapr, fdt, rtas_addr);
1657 
1658     rc = fdt_pack(fdt);
1659 
1660     /* Should only fail if we've built a corrupted tree */
1661     assert(rc == 0);
1662 
1663     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1664         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1665                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1666         exit(1);
1667     }
1668 
1669     /* Load the fdt */
1670     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1671     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1672     g_free(fdt);
1673 
1674     /* Set up the entry state */
1675     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1676     first_ppc_cpu->env.gpr[5] = 0;
1677 
1678     spapr->cas_reboot = false;
1679 }
1680 
1681 static void spapr_create_nvram(sPAPRMachineState *spapr)
1682 {
1683     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1684     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1685 
1686     if (dinfo) {
1687         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1688                             &error_fatal);
1689     }
1690 
1691     qdev_init_nofail(dev);
1692 
1693     spapr->nvram = (struct sPAPRNVRAM *)dev;
1694 }
1695 
1696 static void spapr_rtc_create(sPAPRMachineState *spapr)
1697 {
1698     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1699     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1700                               &error_fatal);
1701     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1702                               &error_fatal);
1703     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1704                               "date", &error_fatal);
1705 }
1706 
1707 /* Returns whether we want to use VGA or not */
1708 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1709 {
1710     switch (vga_interface_type) {
1711     case VGA_NONE:
1712         return false;
1713     case VGA_DEVICE:
1714         return true;
1715     case VGA_STD:
1716     case VGA_VIRTIO:
1717         return pci_vga_init(pci_bus) != NULL;
1718     default:
1719         error_setg(errp,
1720                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1721         return false;
1722     }
1723 }
1724 
1725 static int spapr_pre_load(void *opaque)
1726 {
1727     int rc;
1728 
1729     rc = spapr_caps_pre_load(opaque);
1730     if (rc) {
1731         return rc;
1732     }
1733 
1734     return 0;
1735 }
1736 
1737 static int spapr_post_load(void *opaque, int version_id)
1738 {
1739     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1740     int err = 0;
1741 
1742     err = spapr_caps_post_migration(spapr);
1743     if (err) {
1744         return err;
1745     }
1746 
1747     /* In earlier versions, there was no separate qdev for the PAPR
1748      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1749      * So when migrating from those versions, poke the incoming offset
1750      * value into the RTC device */
1751     if (version_id < 3) {
1752         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1753     }
1754 
1755     if (kvm_enabled() && spapr->patb_entry) {
1756         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1757         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1758         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1759 
1760         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1761         if (err) {
1762             error_report("Process table config unsupported by the host");
1763             return -EINVAL;
1764         }
1765     }
1766 
1767     err = spapr_irq_post_load(spapr, version_id);
1768     if (err) {
1769         return err;
1770     }
1771 
1772     return err;
1773 }
1774 
1775 static int spapr_pre_save(void *opaque)
1776 {
1777     int rc;
1778 
1779     rc = spapr_caps_pre_save(opaque);
1780     if (rc) {
1781         return rc;
1782     }
1783 
1784     return 0;
1785 }
1786 
1787 static bool version_before_3(void *opaque, int version_id)
1788 {
1789     return version_id < 3;
1790 }
1791 
1792 static bool spapr_pending_events_needed(void *opaque)
1793 {
1794     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1795     return !QTAILQ_EMPTY(&spapr->pending_events);
1796 }
1797 
1798 static const VMStateDescription vmstate_spapr_event_entry = {
1799     .name = "spapr_event_log_entry",
1800     .version_id = 1,
1801     .minimum_version_id = 1,
1802     .fields = (VMStateField[]) {
1803         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1804         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1805         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1806                                      NULL, extended_length),
1807         VMSTATE_END_OF_LIST()
1808     },
1809 };
1810 
1811 static const VMStateDescription vmstate_spapr_pending_events = {
1812     .name = "spapr_pending_events",
1813     .version_id = 1,
1814     .minimum_version_id = 1,
1815     .needed = spapr_pending_events_needed,
1816     .fields = (VMStateField[]) {
1817         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1818                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1819         VMSTATE_END_OF_LIST()
1820     },
1821 };
1822 
1823 static bool spapr_ov5_cas_needed(void *opaque)
1824 {
1825     sPAPRMachineState *spapr = opaque;
1826     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1827     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1828     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1829     bool cas_needed;
1830 
1831     /* Prior to the introduction of sPAPROptionVector, we had two option
1832      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1833      * Both of these options encode machine topology into the device-tree
1834      * in such a way that the now-booted OS should still be able to interact
1835      * appropriately with QEMU regardless of what options were actually
1836      * negotiatied on the source side.
1837      *
1838      * As such, we can avoid migrating the CAS-negotiated options if these
1839      * are the only options available on the current machine/platform.
1840      * Since these are the only options available for pseries-2.7 and
1841      * earlier, this allows us to maintain old->new/new->old migration
1842      * compatibility.
1843      *
1844      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1845      * via default pseries-2.8 machines and explicit command-line parameters.
1846      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1847      * of the actual CAS-negotiated values to continue working properly. For
1848      * example, availability of memory unplug depends on knowing whether
1849      * OV5_HP_EVT was negotiated via CAS.
1850      *
1851      * Thus, for any cases where the set of available CAS-negotiatable
1852      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1853      * include the CAS-negotiated options in the migration stream, unless
1854      * if they affect boot time behaviour only.
1855      */
1856     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1857     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1858     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1859 
1860     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1861      * the mask itself since in the future it's possible "legacy" bits may be
1862      * removed via machine options, which could generate a false positive
1863      * that breaks migration.
1864      */
1865     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1866     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1867 
1868     spapr_ovec_cleanup(ov5_mask);
1869     spapr_ovec_cleanup(ov5_legacy);
1870     spapr_ovec_cleanup(ov5_removed);
1871 
1872     return cas_needed;
1873 }
1874 
1875 static const VMStateDescription vmstate_spapr_ov5_cas = {
1876     .name = "spapr_option_vector_ov5_cas",
1877     .version_id = 1,
1878     .minimum_version_id = 1,
1879     .needed = spapr_ov5_cas_needed,
1880     .fields = (VMStateField[]) {
1881         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1882                                  vmstate_spapr_ovec, sPAPROptionVector),
1883         VMSTATE_END_OF_LIST()
1884     },
1885 };
1886 
1887 static bool spapr_patb_entry_needed(void *opaque)
1888 {
1889     sPAPRMachineState *spapr = opaque;
1890 
1891     return !!spapr->patb_entry;
1892 }
1893 
1894 static const VMStateDescription vmstate_spapr_patb_entry = {
1895     .name = "spapr_patb_entry",
1896     .version_id = 1,
1897     .minimum_version_id = 1,
1898     .needed = spapr_patb_entry_needed,
1899     .fields = (VMStateField[]) {
1900         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1901         VMSTATE_END_OF_LIST()
1902     },
1903 };
1904 
1905 static bool spapr_irq_map_needed(void *opaque)
1906 {
1907     sPAPRMachineState *spapr = opaque;
1908 
1909     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1910 }
1911 
1912 static const VMStateDescription vmstate_spapr_irq_map = {
1913     .name = "spapr_irq_map",
1914     .version_id = 1,
1915     .minimum_version_id = 1,
1916     .needed = spapr_irq_map_needed,
1917     .fields = (VMStateField[]) {
1918         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1919         VMSTATE_END_OF_LIST()
1920     },
1921 };
1922 
1923 static const VMStateDescription vmstate_spapr = {
1924     .name = "spapr",
1925     .version_id = 3,
1926     .minimum_version_id = 1,
1927     .pre_load = spapr_pre_load,
1928     .post_load = spapr_post_load,
1929     .pre_save = spapr_pre_save,
1930     .fields = (VMStateField[]) {
1931         /* used to be @next_irq */
1932         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1933 
1934         /* RTC offset */
1935         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1936 
1937         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1938         VMSTATE_END_OF_LIST()
1939     },
1940     .subsections = (const VMStateDescription*[]) {
1941         &vmstate_spapr_ov5_cas,
1942         &vmstate_spapr_patb_entry,
1943         &vmstate_spapr_pending_events,
1944         &vmstate_spapr_cap_htm,
1945         &vmstate_spapr_cap_vsx,
1946         &vmstate_spapr_cap_dfp,
1947         &vmstate_spapr_cap_cfpc,
1948         &vmstate_spapr_cap_sbbc,
1949         &vmstate_spapr_cap_ibs,
1950         &vmstate_spapr_irq_map,
1951         &vmstate_spapr_cap_nested_kvm_hv,
1952         NULL
1953     }
1954 };
1955 
1956 static int htab_save_setup(QEMUFile *f, void *opaque)
1957 {
1958     sPAPRMachineState *spapr = opaque;
1959 
1960     /* "Iteration" header */
1961     if (!spapr->htab_shift) {
1962         qemu_put_be32(f, -1);
1963     } else {
1964         qemu_put_be32(f, spapr->htab_shift);
1965     }
1966 
1967     if (spapr->htab) {
1968         spapr->htab_save_index = 0;
1969         spapr->htab_first_pass = true;
1970     } else {
1971         if (spapr->htab_shift) {
1972             assert(kvm_enabled());
1973         }
1974     }
1975 
1976 
1977     return 0;
1978 }
1979 
1980 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1981                             int chunkstart, int n_valid, int n_invalid)
1982 {
1983     qemu_put_be32(f, chunkstart);
1984     qemu_put_be16(f, n_valid);
1985     qemu_put_be16(f, n_invalid);
1986     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1987                     HASH_PTE_SIZE_64 * n_valid);
1988 }
1989 
1990 static void htab_save_end_marker(QEMUFile *f)
1991 {
1992     qemu_put_be32(f, 0);
1993     qemu_put_be16(f, 0);
1994     qemu_put_be16(f, 0);
1995 }
1996 
1997 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1998                                  int64_t max_ns)
1999 {
2000     bool has_timeout = max_ns != -1;
2001     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2002     int index = spapr->htab_save_index;
2003     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2004 
2005     assert(spapr->htab_first_pass);
2006 
2007     do {
2008         int chunkstart;
2009 
2010         /* Consume invalid HPTEs */
2011         while ((index < htabslots)
2012                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2013             CLEAN_HPTE(HPTE(spapr->htab, index));
2014             index++;
2015         }
2016 
2017         /* Consume valid HPTEs */
2018         chunkstart = index;
2019         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2020                && HPTE_VALID(HPTE(spapr->htab, index))) {
2021             CLEAN_HPTE(HPTE(spapr->htab, index));
2022             index++;
2023         }
2024 
2025         if (index > chunkstart) {
2026             int n_valid = index - chunkstart;
2027 
2028             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2029 
2030             if (has_timeout &&
2031                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2032                 break;
2033             }
2034         }
2035     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2036 
2037     if (index >= htabslots) {
2038         assert(index == htabslots);
2039         index = 0;
2040         spapr->htab_first_pass = false;
2041     }
2042     spapr->htab_save_index = index;
2043 }
2044 
2045 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2046                                 int64_t max_ns)
2047 {
2048     bool final = max_ns < 0;
2049     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2050     int examined = 0, sent = 0;
2051     int index = spapr->htab_save_index;
2052     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2053 
2054     assert(!spapr->htab_first_pass);
2055 
2056     do {
2057         int chunkstart, invalidstart;
2058 
2059         /* Consume non-dirty HPTEs */
2060         while ((index < htabslots)
2061                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2062             index++;
2063             examined++;
2064         }
2065 
2066         chunkstart = index;
2067         /* Consume valid dirty HPTEs */
2068         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2069                && HPTE_DIRTY(HPTE(spapr->htab, index))
2070                && HPTE_VALID(HPTE(spapr->htab, index))) {
2071             CLEAN_HPTE(HPTE(spapr->htab, index));
2072             index++;
2073             examined++;
2074         }
2075 
2076         invalidstart = index;
2077         /* Consume invalid dirty HPTEs */
2078         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2079                && HPTE_DIRTY(HPTE(spapr->htab, index))
2080                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2081             CLEAN_HPTE(HPTE(spapr->htab, index));
2082             index++;
2083             examined++;
2084         }
2085 
2086         if (index > chunkstart) {
2087             int n_valid = invalidstart - chunkstart;
2088             int n_invalid = index - invalidstart;
2089 
2090             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2091             sent += index - chunkstart;
2092 
2093             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2094                 break;
2095             }
2096         }
2097 
2098         if (examined >= htabslots) {
2099             break;
2100         }
2101 
2102         if (index >= htabslots) {
2103             assert(index == htabslots);
2104             index = 0;
2105         }
2106     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2107 
2108     if (index >= htabslots) {
2109         assert(index == htabslots);
2110         index = 0;
2111     }
2112 
2113     spapr->htab_save_index = index;
2114 
2115     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2116 }
2117 
2118 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2119 #define MAX_KVM_BUF_SIZE    2048
2120 
2121 static int htab_save_iterate(QEMUFile *f, void *opaque)
2122 {
2123     sPAPRMachineState *spapr = opaque;
2124     int fd;
2125     int rc = 0;
2126 
2127     /* Iteration header */
2128     if (!spapr->htab_shift) {
2129         qemu_put_be32(f, -1);
2130         return 1;
2131     } else {
2132         qemu_put_be32(f, 0);
2133     }
2134 
2135     if (!spapr->htab) {
2136         assert(kvm_enabled());
2137 
2138         fd = get_htab_fd(spapr);
2139         if (fd < 0) {
2140             return fd;
2141         }
2142 
2143         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2144         if (rc < 0) {
2145             return rc;
2146         }
2147     } else  if (spapr->htab_first_pass) {
2148         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2149     } else {
2150         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2151     }
2152 
2153     htab_save_end_marker(f);
2154 
2155     return rc;
2156 }
2157 
2158 static int htab_save_complete(QEMUFile *f, void *opaque)
2159 {
2160     sPAPRMachineState *spapr = opaque;
2161     int fd;
2162 
2163     /* Iteration header */
2164     if (!spapr->htab_shift) {
2165         qemu_put_be32(f, -1);
2166         return 0;
2167     } else {
2168         qemu_put_be32(f, 0);
2169     }
2170 
2171     if (!spapr->htab) {
2172         int rc;
2173 
2174         assert(kvm_enabled());
2175 
2176         fd = get_htab_fd(spapr);
2177         if (fd < 0) {
2178             return fd;
2179         }
2180 
2181         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2182         if (rc < 0) {
2183             return rc;
2184         }
2185     } else {
2186         if (spapr->htab_first_pass) {
2187             htab_save_first_pass(f, spapr, -1);
2188         }
2189         htab_save_later_pass(f, spapr, -1);
2190     }
2191 
2192     /* End marker */
2193     htab_save_end_marker(f);
2194 
2195     return 0;
2196 }
2197 
2198 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2199 {
2200     sPAPRMachineState *spapr = opaque;
2201     uint32_t section_hdr;
2202     int fd = -1;
2203     Error *local_err = NULL;
2204 
2205     if (version_id < 1 || version_id > 1) {
2206         error_report("htab_load() bad version");
2207         return -EINVAL;
2208     }
2209 
2210     section_hdr = qemu_get_be32(f);
2211 
2212     if (section_hdr == -1) {
2213         spapr_free_hpt(spapr);
2214         return 0;
2215     }
2216 
2217     if (section_hdr) {
2218         /* First section gives the htab size */
2219         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2220         if (local_err) {
2221             error_report_err(local_err);
2222             return -EINVAL;
2223         }
2224         return 0;
2225     }
2226 
2227     if (!spapr->htab) {
2228         assert(kvm_enabled());
2229 
2230         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2231         if (fd < 0) {
2232             error_report_err(local_err);
2233             return fd;
2234         }
2235     }
2236 
2237     while (true) {
2238         uint32_t index;
2239         uint16_t n_valid, n_invalid;
2240 
2241         index = qemu_get_be32(f);
2242         n_valid = qemu_get_be16(f);
2243         n_invalid = qemu_get_be16(f);
2244 
2245         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2246             /* End of Stream */
2247             break;
2248         }
2249 
2250         if ((index + n_valid + n_invalid) >
2251             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2252             /* Bad index in stream */
2253             error_report(
2254                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2255                 index, n_valid, n_invalid, spapr->htab_shift);
2256             return -EINVAL;
2257         }
2258 
2259         if (spapr->htab) {
2260             if (n_valid) {
2261                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2262                                 HASH_PTE_SIZE_64 * n_valid);
2263             }
2264             if (n_invalid) {
2265                 memset(HPTE(spapr->htab, index + n_valid), 0,
2266                        HASH_PTE_SIZE_64 * n_invalid);
2267             }
2268         } else {
2269             int rc;
2270 
2271             assert(fd >= 0);
2272 
2273             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2274             if (rc < 0) {
2275                 return rc;
2276             }
2277         }
2278     }
2279 
2280     if (!spapr->htab) {
2281         assert(fd >= 0);
2282         close(fd);
2283     }
2284 
2285     return 0;
2286 }
2287 
2288 static void htab_save_cleanup(void *opaque)
2289 {
2290     sPAPRMachineState *spapr = opaque;
2291 
2292     close_htab_fd(spapr);
2293 }
2294 
2295 static SaveVMHandlers savevm_htab_handlers = {
2296     .save_setup = htab_save_setup,
2297     .save_live_iterate = htab_save_iterate,
2298     .save_live_complete_precopy = htab_save_complete,
2299     .save_cleanup = htab_save_cleanup,
2300     .load_state = htab_load,
2301 };
2302 
2303 static void spapr_boot_set(void *opaque, const char *boot_device,
2304                            Error **errp)
2305 {
2306     MachineState *machine = MACHINE(opaque);
2307     machine->boot_order = g_strdup(boot_device);
2308 }
2309 
2310 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2311 {
2312     MachineState *machine = MACHINE(spapr);
2313     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2314     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2315     int i;
2316 
2317     for (i = 0; i < nr_lmbs; i++) {
2318         uint64_t addr;
2319 
2320         addr = i * lmb_size + machine->device_memory->base;
2321         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2322                                addr / lmb_size);
2323     }
2324 }
2325 
2326 /*
2327  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2328  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2329  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2330  */
2331 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2332 {
2333     int i;
2334 
2335     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2336         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2337                    " is not aligned to %" PRIu64 " MiB",
2338                    machine->ram_size,
2339                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2340         return;
2341     }
2342 
2343     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2344         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2345                    " is not aligned to %" PRIu64 " MiB",
2346                    machine->ram_size,
2347                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2348         return;
2349     }
2350 
2351     for (i = 0; i < nb_numa_nodes; i++) {
2352         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2353             error_setg(errp,
2354                        "Node %d memory size 0x%" PRIx64
2355                        " is not aligned to %" PRIu64 " MiB",
2356                        i, numa_info[i].node_mem,
2357                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2358             return;
2359         }
2360     }
2361 }
2362 
2363 /* find cpu slot in machine->possible_cpus by core_id */
2364 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2365 {
2366     int index = id / smp_threads;
2367 
2368     if (index >= ms->possible_cpus->len) {
2369         return NULL;
2370     }
2371     if (idx) {
2372         *idx = index;
2373     }
2374     return &ms->possible_cpus->cpus[index];
2375 }
2376 
2377 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2378 {
2379     Error *local_err = NULL;
2380     bool vsmt_user = !!spapr->vsmt;
2381     int kvm_smt = kvmppc_smt_threads();
2382     int ret;
2383 
2384     if (!kvm_enabled() && (smp_threads > 1)) {
2385         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2386                      "on a pseries machine");
2387         goto out;
2388     }
2389     if (!is_power_of_2(smp_threads)) {
2390         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2391                      "machine because it must be a power of 2", smp_threads);
2392         goto out;
2393     }
2394 
2395     /* Detemine the VSMT mode to use: */
2396     if (vsmt_user) {
2397         if (spapr->vsmt < smp_threads) {
2398             error_setg(&local_err, "Cannot support VSMT mode %d"
2399                          " because it must be >= threads/core (%d)",
2400                          spapr->vsmt, smp_threads);
2401             goto out;
2402         }
2403         /* In this case, spapr->vsmt has been set by the command line */
2404     } else {
2405         /*
2406          * Default VSMT value is tricky, because we need it to be as
2407          * consistent as possible (for migration), but this requires
2408          * changing it for at least some existing cases.  We pick 8 as
2409          * the value that we'd get with KVM on POWER8, the
2410          * overwhelmingly common case in production systems.
2411          */
2412         spapr->vsmt = MAX(8, smp_threads);
2413     }
2414 
2415     /* KVM: If necessary, set the SMT mode: */
2416     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2417         ret = kvmppc_set_smt_threads(spapr->vsmt);
2418         if (ret) {
2419             /* Looks like KVM isn't able to change VSMT mode */
2420             error_setg(&local_err,
2421                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2422                        spapr->vsmt, ret);
2423             /* We can live with that if the default one is big enough
2424              * for the number of threads, and a submultiple of the one
2425              * we want.  In this case we'll waste some vcpu ids, but
2426              * behaviour will be correct */
2427             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2428                 warn_report_err(local_err);
2429                 local_err = NULL;
2430                 goto out;
2431             } else {
2432                 if (!vsmt_user) {
2433                     error_append_hint(&local_err,
2434                                       "On PPC, a VM with %d threads/core"
2435                                       " on a host with %d threads/core"
2436                                       " requires the use of VSMT mode %d.\n",
2437                                       smp_threads, kvm_smt, spapr->vsmt);
2438                 }
2439                 kvmppc_hint_smt_possible(&local_err);
2440                 goto out;
2441             }
2442         }
2443     }
2444     /* else TCG: nothing to do currently */
2445 out:
2446     error_propagate(errp, local_err);
2447 }
2448 
2449 static void spapr_init_cpus(sPAPRMachineState *spapr)
2450 {
2451     MachineState *machine = MACHINE(spapr);
2452     MachineClass *mc = MACHINE_GET_CLASS(machine);
2453     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2454     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2455     const CPUArchIdList *possible_cpus;
2456     int boot_cores_nr = smp_cpus / smp_threads;
2457     int i;
2458 
2459     possible_cpus = mc->possible_cpu_arch_ids(machine);
2460     if (mc->has_hotpluggable_cpus) {
2461         if (smp_cpus % smp_threads) {
2462             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2463                          smp_cpus, smp_threads);
2464             exit(1);
2465         }
2466         if (max_cpus % smp_threads) {
2467             error_report("max_cpus (%u) must be multiple of threads (%u)",
2468                          max_cpus, smp_threads);
2469             exit(1);
2470         }
2471     } else {
2472         if (max_cpus != smp_cpus) {
2473             error_report("This machine version does not support CPU hotplug");
2474             exit(1);
2475         }
2476         boot_cores_nr = possible_cpus->len;
2477     }
2478 
2479     if (smc->pre_2_10_has_unused_icps) {
2480         int i;
2481 
2482         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2483             /* Dummy entries get deregistered when real ICPState objects
2484              * are registered during CPU core hotplug.
2485              */
2486             pre_2_10_vmstate_register_dummy_icp(i);
2487         }
2488     }
2489 
2490     for (i = 0; i < possible_cpus->len; i++) {
2491         int core_id = i * smp_threads;
2492 
2493         if (mc->has_hotpluggable_cpus) {
2494             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2495                                    spapr_vcpu_id(spapr, core_id));
2496         }
2497 
2498         if (i < boot_cores_nr) {
2499             Object *core  = object_new(type);
2500             int nr_threads = smp_threads;
2501 
2502             /* Handle the partially filled core for older machine types */
2503             if ((i + 1) * smp_threads >= smp_cpus) {
2504                 nr_threads = smp_cpus - i * smp_threads;
2505             }
2506 
2507             object_property_set_int(core, nr_threads, "nr-threads",
2508                                     &error_fatal);
2509             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2510                                     &error_fatal);
2511             object_property_set_bool(core, true, "realized", &error_fatal);
2512 
2513             object_unref(core);
2514         }
2515     }
2516 }
2517 
2518 /* pSeries LPAR / sPAPR hardware init */
2519 static void spapr_machine_init(MachineState *machine)
2520 {
2521     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2522     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2523     const char *kernel_filename = machine->kernel_filename;
2524     const char *initrd_filename = machine->initrd_filename;
2525     PCIHostState *phb;
2526     int i;
2527     MemoryRegion *sysmem = get_system_memory();
2528     MemoryRegion *ram = g_new(MemoryRegion, 1);
2529     hwaddr node0_size = spapr_node0_size(machine);
2530     long load_limit, fw_size;
2531     char *filename;
2532     Error *resize_hpt_err = NULL;
2533 
2534     msi_nonbroken = true;
2535 
2536     QLIST_INIT(&spapr->phbs);
2537     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2538 
2539     /* Determine capabilities to run with */
2540     spapr_caps_init(spapr);
2541 
2542     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2543     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2544         /*
2545          * If the user explicitly requested a mode we should either
2546          * supply it, or fail completely (which we do below).  But if
2547          * it's not set explicitly, we reset our mode to something
2548          * that works
2549          */
2550         if (resize_hpt_err) {
2551             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2552             error_free(resize_hpt_err);
2553             resize_hpt_err = NULL;
2554         } else {
2555             spapr->resize_hpt = smc->resize_hpt_default;
2556         }
2557     }
2558 
2559     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2560 
2561     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2562         /*
2563          * User requested HPT resize, but this host can't supply it.  Bail out
2564          */
2565         error_report_err(resize_hpt_err);
2566         exit(1);
2567     }
2568 
2569     spapr->rma_size = node0_size;
2570 
2571     /* With KVM, we don't actually know whether KVM supports an
2572      * unbounded RMA (PR KVM) or is limited by the hash table size
2573      * (HV KVM using VRMA), so we always assume the latter
2574      *
2575      * In that case, we also limit the initial allocations for RTAS
2576      * etc... to 256M since we have no way to know what the VRMA size
2577      * is going to be as it depends on the size of the hash table
2578      * which isn't determined yet.
2579      */
2580     if (kvm_enabled()) {
2581         spapr->vrma_adjust = 1;
2582         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2583     }
2584 
2585     /* Actually we don't support unbounded RMA anymore since we added
2586      * proper emulation of HV mode. The max we can get is 16G which
2587      * also happens to be what we configure for PAPR mode so make sure
2588      * we don't do anything bigger than that
2589      */
2590     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2591 
2592     if (spapr->rma_size > node0_size) {
2593         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2594                      spapr->rma_size);
2595         exit(1);
2596     }
2597 
2598     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2599     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2600 
2601     /*
2602      * VSMT must be set in order to be able to compute VCPU ids, ie to
2603      * call spapr_max_server_number() or spapr_vcpu_id().
2604      */
2605     spapr_set_vsmt_mode(spapr, &error_fatal);
2606 
2607     /* Set up Interrupt Controller before we create the VCPUs */
2608     spapr_irq_init(spapr, &error_fatal);
2609 
2610     /* Set up containers for ibm,client-architecture-support negotiated options
2611      */
2612     spapr->ov5 = spapr_ovec_new();
2613     spapr->ov5_cas = spapr_ovec_new();
2614 
2615     if (smc->dr_lmb_enabled) {
2616         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2617         spapr_validate_node_memory(machine, &error_fatal);
2618     }
2619 
2620     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2621 
2622     /* advertise support for dedicated HP event source to guests */
2623     if (spapr->use_hotplug_event_source) {
2624         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2625     }
2626 
2627     /* advertise support for HPT resizing */
2628     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2629         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2630     }
2631 
2632     /* advertise support for ibm,dyamic-memory-v2 */
2633     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2634 
2635     /* advertise XIVE on POWER9 machines */
2636     if (spapr->irq->ov5 & SPAPR_OV5_XIVE_EXPLOIT) {
2637         if (ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
2638                                   0, spapr->max_compat_pvr)) {
2639             spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2640         } else {
2641             error_report("XIVE-only machines require a POWER9 CPU");
2642             exit(1);
2643         }
2644     }
2645 
2646     /* init CPUs */
2647     spapr_init_cpus(spapr);
2648 
2649     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2650         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2651                               spapr->max_compat_pvr)) {
2652         /* KVM and TCG always allow GTSE with radix... */
2653         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2654     }
2655     /* ... but not with hash (currently). */
2656 
2657     if (kvm_enabled()) {
2658         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2659         kvmppc_enable_logical_ci_hcalls();
2660         kvmppc_enable_set_mode_hcall();
2661 
2662         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2663         kvmppc_enable_clear_ref_mod_hcalls();
2664     }
2665 
2666     /* allocate RAM */
2667     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2668                                          machine->ram_size);
2669     memory_region_add_subregion(sysmem, 0, ram);
2670 
2671     /* always allocate the device memory information */
2672     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2673 
2674     /* initialize hotplug memory address space */
2675     if (machine->ram_size < machine->maxram_size) {
2676         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2677         /*
2678          * Limit the number of hotpluggable memory slots to half the number
2679          * slots that KVM supports, leaving the other half for PCI and other
2680          * devices. However ensure that number of slots doesn't drop below 32.
2681          */
2682         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2683                            SPAPR_MAX_RAM_SLOTS;
2684 
2685         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2686             max_memslots = SPAPR_MAX_RAM_SLOTS;
2687         }
2688         if (machine->ram_slots > max_memslots) {
2689             error_report("Specified number of memory slots %"
2690                          PRIu64" exceeds max supported %d",
2691                          machine->ram_slots, max_memslots);
2692             exit(1);
2693         }
2694 
2695         machine->device_memory->base = ROUND_UP(machine->ram_size,
2696                                                 SPAPR_DEVICE_MEM_ALIGN);
2697         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2698                            "device-memory", device_mem_size);
2699         memory_region_add_subregion(sysmem, machine->device_memory->base,
2700                                     &machine->device_memory->mr);
2701     }
2702 
2703     if (smc->dr_lmb_enabled) {
2704         spapr_create_lmb_dr_connectors(spapr);
2705     }
2706 
2707     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2708     if (!filename) {
2709         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2710         exit(1);
2711     }
2712     spapr->rtas_size = get_image_size(filename);
2713     if (spapr->rtas_size < 0) {
2714         error_report("Could not get size of LPAR rtas '%s'", filename);
2715         exit(1);
2716     }
2717     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2718     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2719         error_report("Could not load LPAR rtas '%s'", filename);
2720         exit(1);
2721     }
2722     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2723         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2724                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2725         exit(1);
2726     }
2727     g_free(filename);
2728 
2729     /* Set up RTAS event infrastructure */
2730     spapr_events_init(spapr);
2731 
2732     /* Set up the RTC RTAS interfaces */
2733     spapr_rtc_create(spapr);
2734 
2735     /* Set up VIO bus */
2736     spapr->vio_bus = spapr_vio_bus_init();
2737 
2738     for (i = 0; i < serial_max_hds(); i++) {
2739         if (serial_hd(i)) {
2740             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2741         }
2742     }
2743 
2744     /* We always have at least the nvram device on VIO */
2745     spapr_create_nvram(spapr);
2746 
2747     /* Set up PCI */
2748     spapr_pci_rtas_init();
2749 
2750     phb = spapr_create_phb(spapr, 0);
2751 
2752     for (i = 0; i < nb_nics; i++) {
2753         NICInfo *nd = &nd_table[i];
2754 
2755         if (!nd->model) {
2756             nd->model = g_strdup("spapr-vlan");
2757         }
2758 
2759         if (g_str_equal(nd->model, "spapr-vlan") ||
2760             g_str_equal(nd->model, "ibmveth")) {
2761             spapr_vlan_create(spapr->vio_bus, nd);
2762         } else {
2763             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2764         }
2765     }
2766 
2767     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2768         spapr_vscsi_create(spapr->vio_bus);
2769     }
2770 
2771     /* Graphics */
2772     if (spapr_vga_init(phb->bus, &error_fatal)) {
2773         spapr->has_graphics = true;
2774         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2775     }
2776 
2777     if (machine->usb) {
2778         if (smc->use_ohci_by_default) {
2779             pci_create_simple(phb->bus, -1, "pci-ohci");
2780         } else {
2781             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2782         }
2783 
2784         if (spapr->has_graphics) {
2785             USBBus *usb_bus = usb_bus_find(-1);
2786 
2787             usb_create_simple(usb_bus, "usb-kbd");
2788             usb_create_simple(usb_bus, "usb-mouse");
2789         }
2790     }
2791 
2792     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2793         error_report(
2794             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2795             MIN_RMA_SLOF);
2796         exit(1);
2797     }
2798 
2799     if (kernel_filename) {
2800         uint64_t lowaddr = 0;
2801 
2802         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2803                                       NULL, NULL, &lowaddr, NULL, 1,
2804                                       PPC_ELF_MACHINE, 0, 0);
2805         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2806             spapr->kernel_size = load_elf(kernel_filename,
2807                                           translate_kernel_address, NULL, NULL,
2808                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2809                                           0, 0);
2810             spapr->kernel_le = spapr->kernel_size > 0;
2811         }
2812         if (spapr->kernel_size < 0) {
2813             error_report("error loading %s: %s", kernel_filename,
2814                          load_elf_strerror(spapr->kernel_size));
2815             exit(1);
2816         }
2817 
2818         /* load initrd */
2819         if (initrd_filename) {
2820             /* Try to locate the initrd in the gap between the kernel
2821              * and the firmware. Add a bit of space just in case
2822              */
2823             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2824                                   + 0x1ffff) & ~0xffff;
2825             spapr->initrd_size = load_image_targphys(initrd_filename,
2826                                                      spapr->initrd_base,
2827                                                      load_limit
2828                                                      - spapr->initrd_base);
2829             if (spapr->initrd_size < 0) {
2830                 error_report("could not load initial ram disk '%s'",
2831                              initrd_filename);
2832                 exit(1);
2833             }
2834         }
2835     }
2836 
2837     if (bios_name == NULL) {
2838         bios_name = FW_FILE_NAME;
2839     }
2840     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2841     if (!filename) {
2842         error_report("Could not find LPAR firmware '%s'", bios_name);
2843         exit(1);
2844     }
2845     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2846     if (fw_size <= 0) {
2847         error_report("Could not load LPAR firmware '%s'", filename);
2848         exit(1);
2849     }
2850     g_free(filename);
2851 
2852     /* FIXME: Should register things through the MachineState's qdev
2853      * interface, this is a legacy from the sPAPREnvironment structure
2854      * which predated MachineState but had a similar function */
2855     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2856     register_savevm_live(NULL, "spapr/htab", -1, 1,
2857                          &savevm_htab_handlers, spapr);
2858 
2859     qemu_register_boot_set(spapr_boot_set, spapr);
2860 
2861     if (kvm_enabled()) {
2862         /* to stop and start vmclock */
2863         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2864                                          &spapr->tb);
2865 
2866         kvmppc_spapr_enable_inkernel_multitce();
2867     }
2868 }
2869 
2870 static int spapr_kvm_type(const char *vm_type)
2871 {
2872     if (!vm_type) {
2873         return 0;
2874     }
2875 
2876     if (!strcmp(vm_type, "HV")) {
2877         return 1;
2878     }
2879 
2880     if (!strcmp(vm_type, "PR")) {
2881         return 2;
2882     }
2883 
2884     error_report("Unknown kvm-type specified '%s'", vm_type);
2885     exit(1);
2886 }
2887 
2888 /*
2889  * Implementation of an interface to adjust firmware path
2890  * for the bootindex property handling.
2891  */
2892 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2893                                    DeviceState *dev)
2894 {
2895 #define CAST(type, obj, name) \
2896     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2897     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2898     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2899     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2900 
2901     if (d) {
2902         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2903         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2904         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2905 
2906         if (spapr) {
2907             /*
2908              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2909              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2910              * in the top 16 bits of the 64-bit LUN
2911              */
2912             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2913             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2914                                    (uint64_t)id << 48);
2915         } else if (virtio) {
2916             /*
2917              * We use SRP luns of the form 01000000 | (target << 8) | lun
2918              * in the top 32 bits of the 64-bit LUN
2919              * Note: the quote above is from SLOF and it is wrong,
2920              * the actual binding is:
2921              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2922              */
2923             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2924             if (d->lun >= 256) {
2925                 /* Use the LUN "flat space addressing method" */
2926                 id |= 0x4000;
2927             }
2928             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2929                                    (uint64_t)id << 32);
2930         } else if (usb) {
2931             /*
2932              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2933              * in the top 32 bits of the 64-bit LUN
2934              */
2935             unsigned usb_port = atoi(usb->port->path);
2936             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2937             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2938                                    (uint64_t)id << 32);
2939         }
2940     }
2941 
2942     /*
2943      * SLOF probes the USB devices, and if it recognizes that the device is a
2944      * storage device, it changes its name to "storage" instead of "usb-host",
2945      * and additionally adds a child node for the SCSI LUN, so the correct
2946      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2947      */
2948     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2949         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2950         if (usb_host_dev_is_scsi_storage(usbdev)) {
2951             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2952         }
2953     }
2954 
2955     if (phb) {
2956         /* Replace "pci" with "pci@800000020000000" */
2957         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2958     }
2959 
2960     if (vsc) {
2961         /* Same logic as virtio above */
2962         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2963         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2964     }
2965 
2966     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2967         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2968         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2969         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2970     }
2971 
2972     return NULL;
2973 }
2974 
2975 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2976 {
2977     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2978 
2979     return g_strdup(spapr->kvm_type);
2980 }
2981 
2982 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2983 {
2984     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2985 
2986     g_free(spapr->kvm_type);
2987     spapr->kvm_type = g_strdup(value);
2988 }
2989 
2990 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2991 {
2992     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2993 
2994     return spapr->use_hotplug_event_source;
2995 }
2996 
2997 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2998                                             Error **errp)
2999 {
3000     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3001 
3002     spapr->use_hotplug_event_source = value;
3003 }
3004 
3005 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3006 {
3007     return true;
3008 }
3009 
3010 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3011 {
3012     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3013 
3014     switch (spapr->resize_hpt) {
3015     case SPAPR_RESIZE_HPT_DEFAULT:
3016         return g_strdup("default");
3017     case SPAPR_RESIZE_HPT_DISABLED:
3018         return g_strdup("disabled");
3019     case SPAPR_RESIZE_HPT_ENABLED:
3020         return g_strdup("enabled");
3021     case SPAPR_RESIZE_HPT_REQUIRED:
3022         return g_strdup("required");
3023     }
3024     g_assert_not_reached();
3025 }
3026 
3027 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3028 {
3029     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3030 
3031     if (strcmp(value, "default") == 0) {
3032         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3033     } else if (strcmp(value, "disabled") == 0) {
3034         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3035     } else if (strcmp(value, "enabled") == 0) {
3036         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3037     } else if (strcmp(value, "required") == 0) {
3038         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3039     } else {
3040         error_setg(errp, "Bad value for \"resize-hpt\" property");
3041     }
3042 }
3043 
3044 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3045                                    void *opaque, Error **errp)
3046 {
3047     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3048 }
3049 
3050 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3051                                    void *opaque, Error **errp)
3052 {
3053     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3054 }
3055 
3056 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3057 {
3058     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3059 
3060     if (spapr->irq == &spapr_irq_xics_legacy) {
3061         return g_strdup("legacy");
3062     } else if (spapr->irq == &spapr_irq_xics) {
3063         return g_strdup("xics");
3064     } else if (spapr->irq == &spapr_irq_xive) {
3065         return g_strdup("xive");
3066     }
3067     g_assert_not_reached();
3068 }
3069 
3070 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3071 {
3072     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3073 
3074     /* The legacy IRQ backend can not be set */
3075     if (strcmp(value, "xics") == 0) {
3076         spapr->irq = &spapr_irq_xics;
3077     } else if (strcmp(value, "xive") == 0) {
3078         spapr->irq = &spapr_irq_xive;
3079     } else {
3080         error_setg(errp, "Bad value for \"ic-mode\" property");
3081     }
3082 }
3083 
3084 static void spapr_instance_init(Object *obj)
3085 {
3086     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3087     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3088 
3089     spapr->htab_fd = -1;
3090     spapr->use_hotplug_event_source = true;
3091     object_property_add_str(obj, "kvm-type",
3092                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3093     object_property_set_description(obj, "kvm-type",
3094                                     "Specifies the KVM virtualization mode (HV, PR)",
3095                                     NULL);
3096     object_property_add_bool(obj, "modern-hotplug-events",
3097                             spapr_get_modern_hotplug_events,
3098                             spapr_set_modern_hotplug_events,
3099                             NULL);
3100     object_property_set_description(obj, "modern-hotplug-events",
3101                                     "Use dedicated hotplug event mechanism in"
3102                                     " place of standard EPOW events when possible"
3103                                     " (required for memory hot-unplug support)",
3104                                     NULL);
3105     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3106                             "Maximum permitted CPU compatibility mode",
3107                             &error_fatal);
3108 
3109     object_property_add_str(obj, "resize-hpt",
3110                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3111     object_property_set_description(obj, "resize-hpt",
3112                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3113                                     NULL);
3114     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3115                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3116     object_property_set_description(obj, "vsmt",
3117                                     "Virtual SMT: KVM behaves as if this were"
3118                                     " the host's SMT mode", &error_abort);
3119     object_property_add_bool(obj, "vfio-no-msix-emulation",
3120                              spapr_get_msix_emulation, NULL, NULL);
3121 
3122     /* The machine class defines the default interrupt controller mode */
3123     spapr->irq = smc->irq;
3124     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3125                             spapr_set_ic_mode, NULL);
3126     object_property_set_description(obj, "ic-mode",
3127                  "Specifies the interrupt controller mode (xics, xive)",
3128                  NULL);
3129 }
3130 
3131 static void spapr_machine_finalizefn(Object *obj)
3132 {
3133     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3134 
3135     g_free(spapr->kvm_type);
3136 }
3137 
3138 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3139 {
3140     cpu_synchronize_state(cs);
3141     ppc_cpu_do_system_reset(cs);
3142 }
3143 
3144 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3145 {
3146     CPUState *cs;
3147 
3148     CPU_FOREACH(cs) {
3149         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3150     }
3151 }
3152 
3153 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3154                            uint32_t node, bool dedicated_hp_event_source,
3155                            Error **errp)
3156 {
3157     sPAPRDRConnector *drc;
3158     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3159     int i, fdt_offset, fdt_size;
3160     void *fdt;
3161     uint64_t addr = addr_start;
3162     bool hotplugged = spapr_drc_hotplugged(dev);
3163     Error *local_err = NULL;
3164 
3165     for (i = 0; i < nr_lmbs; i++) {
3166         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3167                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3168         g_assert(drc);
3169 
3170         fdt = create_device_tree(&fdt_size);
3171         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3172                                                 SPAPR_MEMORY_BLOCK_SIZE);
3173 
3174         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3175         if (local_err) {
3176             while (addr > addr_start) {
3177                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3178                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3179                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3180                 spapr_drc_detach(drc);
3181             }
3182             g_free(fdt);
3183             error_propagate(errp, local_err);
3184             return;
3185         }
3186         if (!hotplugged) {
3187             spapr_drc_reset(drc);
3188         }
3189         addr += SPAPR_MEMORY_BLOCK_SIZE;
3190     }
3191     /* send hotplug notification to the
3192      * guest only in case of hotplugged memory
3193      */
3194     if (hotplugged) {
3195         if (dedicated_hp_event_source) {
3196             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3197                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3198             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3199                                                    nr_lmbs,
3200                                                    spapr_drc_index(drc));
3201         } else {
3202             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3203                                            nr_lmbs);
3204         }
3205     }
3206 }
3207 
3208 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3209                               Error **errp)
3210 {
3211     Error *local_err = NULL;
3212     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3213     PCDIMMDevice *dimm = PC_DIMM(dev);
3214     uint64_t size, addr;
3215     uint32_t node;
3216 
3217     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3218 
3219     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3220     if (local_err) {
3221         goto out;
3222     }
3223 
3224     addr = object_property_get_uint(OBJECT(dimm),
3225                                     PC_DIMM_ADDR_PROP, &local_err);
3226     if (local_err) {
3227         goto out_unplug;
3228     }
3229 
3230     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3231                                     &error_abort);
3232     spapr_add_lmbs(dev, addr, size, node,
3233                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3234                    &local_err);
3235     if (local_err) {
3236         goto out_unplug;
3237     }
3238 
3239     return;
3240 
3241 out_unplug:
3242     pc_dimm_unplug(dimm, MACHINE(ms));
3243 out:
3244     error_propagate(errp, local_err);
3245 }
3246 
3247 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3248                                   Error **errp)
3249 {
3250     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3251     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3252     PCDIMMDevice *dimm = PC_DIMM(dev);
3253     Error *local_err = NULL;
3254     uint64_t size;
3255     Object *memdev;
3256     hwaddr pagesize;
3257 
3258     if (!smc->dr_lmb_enabled) {
3259         error_setg(errp, "Memory hotplug not supported for this machine");
3260         return;
3261     }
3262 
3263     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3264     if (local_err) {
3265         error_propagate(errp, local_err);
3266         return;
3267     }
3268 
3269     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3270         error_setg(errp, "Hotplugged memory size must be a multiple of "
3271                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3272         return;
3273     }
3274 
3275     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3276                                       &error_abort);
3277     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3278     spapr_check_pagesize(spapr, pagesize, &local_err);
3279     if (local_err) {
3280         error_propagate(errp, local_err);
3281         return;
3282     }
3283 
3284     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3285 }
3286 
3287 struct sPAPRDIMMState {
3288     PCDIMMDevice *dimm;
3289     uint32_t nr_lmbs;
3290     QTAILQ_ENTRY(sPAPRDIMMState) next;
3291 };
3292 
3293 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3294                                                        PCDIMMDevice *dimm)
3295 {
3296     sPAPRDIMMState *dimm_state = NULL;
3297 
3298     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3299         if (dimm_state->dimm == dimm) {
3300             break;
3301         }
3302     }
3303     return dimm_state;
3304 }
3305 
3306 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3307                                                       uint32_t nr_lmbs,
3308                                                       PCDIMMDevice *dimm)
3309 {
3310     sPAPRDIMMState *ds = NULL;
3311 
3312     /*
3313      * If this request is for a DIMM whose removal had failed earlier
3314      * (due to guest's refusal to remove the LMBs), we would have this
3315      * dimm already in the pending_dimm_unplugs list. In that
3316      * case don't add again.
3317      */
3318     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3319     if (!ds) {
3320         ds = g_malloc0(sizeof(sPAPRDIMMState));
3321         ds->nr_lmbs = nr_lmbs;
3322         ds->dimm = dimm;
3323         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3324     }
3325     return ds;
3326 }
3327 
3328 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3329                                               sPAPRDIMMState *dimm_state)
3330 {
3331     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3332     g_free(dimm_state);
3333 }
3334 
3335 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3336                                                         PCDIMMDevice *dimm)
3337 {
3338     sPAPRDRConnector *drc;
3339     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3340                                                   &error_abort);
3341     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3342     uint32_t avail_lmbs = 0;
3343     uint64_t addr_start, addr;
3344     int i;
3345 
3346     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3347                                          &error_abort);
3348 
3349     addr = addr_start;
3350     for (i = 0; i < nr_lmbs; i++) {
3351         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3352                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3353         g_assert(drc);
3354         if (drc->dev) {
3355             avail_lmbs++;
3356         }
3357         addr += SPAPR_MEMORY_BLOCK_SIZE;
3358     }
3359 
3360     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3361 }
3362 
3363 /* Callback to be called during DRC release. */
3364 void spapr_lmb_release(DeviceState *dev)
3365 {
3366     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3367     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3368     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3369 
3370     /* This information will get lost if a migration occurs
3371      * during the unplug process. In this case recover it. */
3372     if (ds == NULL) {
3373         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3374         g_assert(ds);
3375         /* The DRC being examined by the caller at least must be counted */
3376         g_assert(ds->nr_lmbs);
3377     }
3378 
3379     if (--ds->nr_lmbs) {
3380         return;
3381     }
3382 
3383     /*
3384      * Now that all the LMBs have been removed by the guest, call the
3385      * unplug handler chain. This can never fail.
3386      */
3387     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3388 }
3389 
3390 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3391 {
3392     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3393     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3394 
3395     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3396     object_unparent(OBJECT(dev));
3397     spapr_pending_dimm_unplugs_remove(spapr, ds);
3398 }
3399 
3400 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3401                                         DeviceState *dev, Error **errp)
3402 {
3403     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3404     Error *local_err = NULL;
3405     PCDIMMDevice *dimm = PC_DIMM(dev);
3406     uint32_t nr_lmbs;
3407     uint64_t size, addr_start, addr;
3408     int i;
3409     sPAPRDRConnector *drc;
3410 
3411     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3412     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3413 
3414     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3415                                          &local_err);
3416     if (local_err) {
3417         goto out;
3418     }
3419 
3420     /*
3421      * An existing pending dimm state for this DIMM means that there is an
3422      * unplug operation in progress, waiting for the spapr_lmb_release
3423      * callback to complete the job (BQL can't cover that far). In this case,
3424      * bail out to avoid detaching DRCs that were already released.
3425      */
3426     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3427         error_setg(&local_err,
3428                    "Memory unplug already in progress for device %s",
3429                    dev->id);
3430         goto out;
3431     }
3432 
3433     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3434 
3435     addr = addr_start;
3436     for (i = 0; i < nr_lmbs; i++) {
3437         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3438                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3439         g_assert(drc);
3440 
3441         spapr_drc_detach(drc);
3442         addr += SPAPR_MEMORY_BLOCK_SIZE;
3443     }
3444 
3445     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3446                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3447     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3448                                               nr_lmbs, spapr_drc_index(drc));
3449 out:
3450     error_propagate(errp, local_err);
3451 }
3452 
3453 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3454                                            sPAPRMachineState *spapr)
3455 {
3456     PowerPCCPU *cpu = POWERPC_CPU(cs);
3457     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3458     int id = spapr_get_vcpu_id(cpu);
3459     void *fdt;
3460     int offset, fdt_size;
3461     char *nodename;
3462 
3463     fdt = create_device_tree(&fdt_size);
3464     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3465     offset = fdt_add_subnode(fdt, 0, nodename);
3466 
3467     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3468     g_free(nodename);
3469 
3470     *fdt_offset = offset;
3471     return fdt;
3472 }
3473 
3474 /* Callback to be called during DRC release. */
3475 void spapr_core_release(DeviceState *dev)
3476 {
3477     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3478 
3479     /* Call the unplug handler chain. This can never fail. */
3480     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3481 }
3482 
3483 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3484 {
3485     MachineState *ms = MACHINE(hotplug_dev);
3486     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3487     CPUCore *cc = CPU_CORE(dev);
3488     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3489 
3490     if (smc->pre_2_10_has_unused_icps) {
3491         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3492         int i;
3493 
3494         for (i = 0; i < cc->nr_threads; i++) {
3495             CPUState *cs = CPU(sc->threads[i]);
3496 
3497             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3498         }
3499     }
3500 
3501     assert(core_slot);
3502     core_slot->cpu = NULL;
3503     object_unparent(OBJECT(dev));
3504 }
3505 
3506 static
3507 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3508                                Error **errp)
3509 {
3510     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3511     int index;
3512     sPAPRDRConnector *drc;
3513     CPUCore *cc = CPU_CORE(dev);
3514 
3515     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3516         error_setg(errp, "Unable to find CPU core with core-id: %d",
3517                    cc->core_id);
3518         return;
3519     }
3520     if (index == 0) {
3521         error_setg(errp, "Boot CPU core may not be unplugged");
3522         return;
3523     }
3524 
3525     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3526                           spapr_vcpu_id(spapr, cc->core_id));
3527     g_assert(drc);
3528 
3529     spapr_drc_detach(drc);
3530 
3531     spapr_hotplug_req_remove_by_index(drc);
3532 }
3533 
3534 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3535                             Error **errp)
3536 {
3537     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3538     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3539     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3540     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3541     CPUCore *cc = CPU_CORE(dev);
3542     CPUState *cs = CPU(core->threads[0]);
3543     sPAPRDRConnector *drc;
3544     Error *local_err = NULL;
3545     CPUArchId *core_slot;
3546     int index;
3547     bool hotplugged = spapr_drc_hotplugged(dev);
3548 
3549     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3550     if (!core_slot) {
3551         error_setg(errp, "Unable to find CPU core with core-id: %d",
3552                    cc->core_id);
3553         return;
3554     }
3555     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3556                           spapr_vcpu_id(spapr, cc->core_id));
3557 
3558     g_assert(drc || !mc->has_hotpluggable_cpus);
3559 
3560     if (drc) {
3561         void *fdt;
3562         int fdt_offset;
3563 
3564         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3565 
3566         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3567         if (local_err) {
3568             g_free(fdt);
3569             error_propagate(errp, local_err);
3570             return;
3571         }
3572 
3573         if (hotplugged) {
3574             /*
3575              * Send hotplug notification interrupt to the guest only
3576              * in case of hotplugged CPUs.
3577              */
3578             spapr_hotplug_req_add_by_index(drc);
3579         } else {
3580             spapr_drc_reset(drc);
3581         }
3582     }
3583 
3584     core_slot->cpu = OBJECT(dev);
3585 
3586     if (smc->pre_2_10_has_unused_icps) {
3587         int i;
3588 
3589         for (i = 0; i < cc->nr_threads; i++) {
3590             cs = CPU(core->threads[i]);
3591             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3592         }
3593     }
3594 }
3595 
3596 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3597                                 Error **errp)
3598 {
3599     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3600     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3601     Error *local_err = NULL;
3602     CPUCore *cc = CPU_CORE(dev);
3603     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3604     const char *type = object_get_typename(OBJECT(dev));
3605     CPUArchId *core_slot;
3606     int index;
3607 
3608     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3609         error_setg(&local_err, "CPU hotplug not supported for this machine");
3610         goto out;
3611     }
3612 
3613     if (strcmp(base_core_type, type)) {
3614         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3615         goto out;
3616     }
3617 
3618     if (cc->core_id % smp_threads) {
3619         error_setg(&local_err, "invalid core id %d", cc->core_id);
3620         goto out;
3621     }
3622 
3623     /*
3624      * In general we should have homogeneous threads-per-core, but old
3625      * (pre hotplug support) machine types allow the last core to have
3626      * reduced threads as a compatibility hack for when we allowed
3627      * total vcpus not a multiple of threads-per-core.
3628      */
3629     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3630         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3631                    cc->nr_threads, smp_threads);
3632         goto out;
3633     }
3634 
3635     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3636     if (!core_slot) {
3637         error_setg(&local_err, "core id %d out of range", cc->core_id);
3638         goto out;
3639     }
3640 
3641     if (core_slot->cpu) {
3642         error_setg(&local_err, "core %d already populated", cc->core_id);
3643         goto out;
3644     }
3645 
3646     numa_cpu_pre_plug(core_slot, dev, &local_err);
3647 
3648 out:
3649     error_propagate(errp, local_err);
3650 }
3651 
3652 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3653                                       DeviceState *dev, Error **errp)
3654 {
3655     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3656         spapr_memory_plug(hotplug_dev, dev, errp);
3657     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3658         spapr_core_plug(hotplug_dev, dev, errp);
3659     }
3660 }
3661 
3662 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3663                                         DeviceState *dev, Error **errp)
3664 {
3665     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3666         spapr_memory_unplug(hotplug_dev, dev);
3667     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3668         spapr_core_unplug(hotplug_dev, dev);
3669     }
3670 }
3671 
3672 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3673                                                 DeviceState *dev, Error **errp)
3674 {
3675     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3676     MachineClass *mc = MACHINE_GET_CLASS(sms);
3677 
3678     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3679         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3680             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3681         } else {
3682             /* NOTE: this means there is a window after guest reset, prior to
3683              * CAS negotiation, where unplug requests will fail due to the
3684              * capability not being detected yet. This is a bit different than
3685              * the case with PCI unplug, where the events will be queued and
3686              * eventually handled by the guest after boot
3687              */
3688             error_setg(errp, "Memory hot unplug not supported for this guest");
3689         }
3690     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3691         if (!mc->has_hotpluggable_cpus) {
3692             error_setg(errp, "CPU hot unplug not supported on this machine");
3693             return;
3694         }
3695         spapr_core_unplug_request(hotplug_dev, dev, errp);
3696     }
3697 }
3698 
3699 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3700                                           DeviceState *dev, Error **errp)
3701 {
3702     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3703         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3704     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3705         spapr_core_pre_plug(hotplug_dev, dev, errp);
3706     }
3707 }
3708 
3709 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3710                                                  DeviceState *dev)
3711 {
3712     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3713         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3714         return HOTPLUG_HANDLER(machine);
3715     }
3716     return NULL;
3717 }
3718 
3719 static CpuInstanceProperties
3720 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3721 {
3722     CPUArchId *core_slot;
3723     MachineClass *mc = MACHINE_GET_CLASS(machine);
3724 
3725     /* make sure possible_cpu are intialized */
3726     mc->possible_cpu_arch_ids(machine);
3727     /* get CPU core slot containing thread that matches cpu_index */
3728     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3729     assert(core_slot);
3730     return core_slot->props;
3731 }
3732 
3733 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3734 {
3735     return idx / smp_cores % nb_numa_nodes;
3736 }
3737 
3738 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3739 {
3740     int i;
3741     const char *core_type;
3742     int spapr_max_cores = max_cpus / smp_threads;
3743     MachineClass *mc = MACHINE_GET_CLASS(machine);
3744 
3745     if (!mc->has_hotpluggable_cpus) {
3746         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3747     }
3748     if (machine->possible_cpus) {
3749         assert(machine->possible_cpus->len == spapr_max_cores);
3750         return machine->possible_cpus;
3751     }
3752 
3753     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3754     if (!core_type) {
3755         error_report("Unable to find sPAPR CPU Core definition");
3756         exit(1);
3757     }
3758 
3759     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3760                              sizeof(CPUArchId) * spapr_max_cores);
3761     machine->possible_cpus->len = spapr_max_cores;
3762     for (i = 0; i < machine->possible_cpus->len; i++) {
3763         int core_id = i * smp_threads;
3764 
3765         machine->possible_cpus->cpus[i].type = core_type;
3766         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3767         machine->possible_cpus->cpus[i].arch_id = core_id;
3768         machine->possible_cpus->cpus[i].props.has_core_id = true;
3769         machine->possible_cpus->cpus[i].props.core_id = core_id;
3770     }
3771     return machine->possible_cpus;
3772 }
3773 
3774 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3775                                 uint64_t *buid, hwaddr *pio,
3776                                 hwaddr *mmio32, hwaddr *mmio64,
3777                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3778 {
3779     /*
3780      * New-style PHB window placement.
3781      *
3782      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3783      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3784      * windows.
3785      *
3786      * Some guest kernels can't work with MMIO windows above 1<<46
3787      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3788      *
3789      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3790      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3791      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3792      * 1TiB 64-bit MMIO windows for each PHB.
3793      */
3794     const uint64_t base_buid = 0x800000020000000ULL;
3795 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3796                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3797     int i;
3798 
3799     /* Sanity check natural alignments */
3800     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3801     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3802     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3803     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3804     /* Sanity check bounds */
3805     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3806                       SPAPR_PCI_MEM32_WIN_SIZE);
3807     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3808                       SPAPR_PCI_MEM64_WIN_SIZE);
3809 
3810     if (index >= SPAPR_MAX_PHBS) {
3811         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3812                    SPAPR_MAX_PHBS - 1);
3813         return;
3814     }
3815 
3816     *buid = base_buid + index;
3817     for (i = 0; i < n_dma; ++i) {
3818         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3819     }
3820 
3821     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3822     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3823     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3824 }
3825 
3826 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3827 {
3828     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3829 
3830     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3831 }
3832 
3833 static void spapr_ics_resend(XICSFabric *dev)
3834 {
3835     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3836 
3837     ics_resend(spapr->ics);
3838 }
3839 
3840 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3841 {
3842     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3843 
3844     return cpu ? ICP(cpu->intc) : NULL;
3845 }
3846 
3847 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3848                                  Monitor *mon)
3849 {
3850     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3851 
3852     spapr->irq->print_info(spapr, mon);
3853 }
3854 
3855 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3856 {
3857     return cpu->vcpu_id;
3858 }
3859 
3860 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3861 {
3862     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3863     int vcpu_id;
3864 
3865     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3866 
3867     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3868         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3869         error_append_hint(errp, "Adjust the number of cpus to %d "
3870                           "or try to raise the number of threads per core\n",
3871                           vcpu_id * smp_threads / spapr->vsmt);
3872         return;
3873     }
3874 
3875     cpu->vcpu_id = vcpu_id;
3876 }
3877 
3878 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3879 {
3880     CPUState *cs;
3881 
3882     CPU_FOREACH(cs) {
3883         PowerPCCPU *cpu = POWERPC_CPU(cs);
3884 
3885         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3886             return cpu;
3887         }
3888     }
3889 
3890     return NULL;
3891 }
3892 
3893 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3894 {
3895     MachineClass *mc = MACHINE_CLASS(oc);
3896     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3897     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3898     NMIClass *nc = NMI_CLASS(oc);
3899     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3900     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3901     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3902     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3903 
3904     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3905     mc->ignore_boot_device_suffixes = true;
3906 
3907     /*
3908      * We set up the default / latest behaviour here.  The class_init
3909      * functions for the specific versioned machine types can override
3910      * these details for backwards compatibility
3911      */
3912     mc->init = spapr_machine_init;
3913     mc->reset = spapr_machine_reset;
3914     mc->block_default_type = IF_SCSI;
3915     mc->max_cpus = 1024;
3916     mc->no_parallel = 1;
3917     mc->default_boot_order = "";
3918     mc->default_ram_size = 512 * MiB;
3919     mc->default_display = "std";
3920     mc->kvm_type = spapr_kvm_type;
3921     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3922     mc->pci_allow_0_address = true;
3923     assert(!mc->get_hotplug_handler);
3924     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3925     hc->pre_plug = spapr_machine_device_pre_plug;
3926     hc->plug = spapr_machine_device_plug;
3927     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3928     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3929     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3930     hc->unplug_request = spapr_machine_device_unplug_request;
3931     hc->unplug = spapr_machine_device_unplug;
3932 
3933     smc->dr_lmb_enabled = true;
3934     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
3935     mc->has_hotpluggable_cpus = true;
3936     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3937     fwc->get_dev_path = spapr_get_fw_dev_path;
3938     nc->nmi_monitor_handler = spapr_nmi;
3939     smc->phb_placement = spapr_phb_placement;
3940     vhc->hypercall = emulate_spapr_hypercall;
3941     vhc->hpt_mask = spapr_hpt_mask;
3942     vhc->map_hptes = spapr_map_hptes;
3943     vhc->unmap_hptes = spapr_unmap_hptes;
3944     vhc->store_hpte = spapr_store_hpte;
3945     vhc->get_patbe = spapr_get_patbe;
3946     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3947     xic->ics_get = spapr_ics_get;
3948     xic->ics_resend = spapr_ics_resend;
3949     xic->icp_get = spapr_icp_get;
3950     ispc->print_info = spapr_pic_print_info;
3951     /* Force NUMA node memory size to be a multiple of
3952      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3953      * in which LMBs are represented and hot-added
3954      */
3955     mc->numa_mem_align_shift = 28;
3956 
3957     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3958     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3959     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3960     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3961     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3962     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3963     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
3964     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
3965     spapr_caps_add_properties(smc, &error_abort);
3966     smc->irq = &spapr_irq_xics;
3967 }
3968 
3969 static const TypeInfo spapr_machine_info = {
3970     .name          = TYPE_SPAPR_MACHINE,
3971     .parent        = TYPE_MACHINE,
3972     .abstract      = true,
3973     .instance_size = sizeof(sPAPRMachineState),
3974     .instance_init = spapr_instance_init,
3975     .instance_finalize = spapr_machine_finalizefn,
3976     .class_size    = sizeof(sPAPRMachineClass),
3977     .class_init    = spapr_machine_class_init,
3978     .interfaces = (InterfaceInfo[]) {
3979         { TYPE_FW_PATH_PROVIDER },
3980         { TYPE_NMI },
3981         { TYPE_HOTPLUG_HANDLER },
3982         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3983         { TYPE_XICS_FABRIC },
3984         { TYPE_INTERRUPT_STATS_PROVIDER },
3985         { }
3986     },
3987 };
3988 
3989 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3990     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3991                                                     void *data)      \
3992     {                                                                \
3993         MachineClass *mc = MACHINE_CLASS(oc);                        \
3994         spapr_machine_##suffix##_class_options(mc);                  \
3995         if (latest) {                                                \
3996             mc->alias = "pseries";                                   \
3997             mc->is_default = 1;                                      \
3998         }                                                            \
3999     }                                                                \
4000     static const TypeInfo spapr_machine_##suffix##_info = {          \
4001         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4002         .parent = TYPE_SPAPR_MACHINE,                                \
4003         .class_init = spapr_machine_##suffix##_class_init,           \
4004     };                                                               \
4005     static void spapr_machine_register_##suffix(void)                \
4006     {                                                                \
4007         type_register(&spapr_machine_##suffix##_info);               \
4008     }                                                                \
4009     type_init(spapr_machine_register_##suffix)
4010 
4011 /*
4012  * pseries-4.0
4013  */
4014 static void spapr_machine_4_0_class_options(MachineClass *mc)
4015 {
4016     /* Defaults for the latest behaviour inherited from the base class */
4017 }
4018 
4019 DEFINE_SPAPR_MACHINE(4_0, "4.0", true);
4020 
4021 /*
4022  * pseries-3.1
4023  */
4024 #define SPAPR_COMPAT_3_1                                              \
4025     HW_COMPAT_3_1
4026 
4027 static void spapr_machine_3_1_class_options(MachineClass *mc)
4028 {
4029     spapr_machine_4_0_class_options(mc);
4030     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_3_1);
4031     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4032 }
4033 
4034 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4035 
4036 /*
4037  * pseries-3.0
4038  */
4039 #define SPAPR_COMPAT_3_0                                              \
4040     HW_COMPAT_3_0
4041 
4042 static void spapr_machine_3_0_class_options(MachineClass *mc)
4043 {
4044     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4045 
4046     spapr_machine_3_1_class_options(mc);
4047     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_3_0);
4048 
4049     smc->legacy_irq_allocation = true;
4050     smc->irq = &spapr_irq_xics_legacy;
4051 }
4052 
4053 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4054 
4055 /*
4056  * pseries-2.12
4057  */
4058 #define SPAPR_COMPAT_2_12                                              \
4059     HW_COMPAT_2_12                                                     \
4060     {                                                                  \
4061         .driver = TYPE_POWERPC_CPU,                                    \
4062         .property = "pre-3.0-migration",                               \
4063         .value    = "on",                                              \
4064     },                                                                 \
4065     {                                                                  \
4066         .driver = TYPE_SPAPR_CPU_CORE,                                 \
4067         .property = "pre-3.0-migration",                               \
4068         .value    = "on",                                              \
4069     },
4070 
4071 static void spapr_machine_2_12_class_options(MachineClass *mc)
4072 {
4073     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4074 
4075     spapr_machine_3_0_class_options(mc);
4076     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4077 
4078     /* We depend on kvm_enabled() to choose a default value for the
4079      * hpt-max-page-size capability. Of course we can't do it here
4080      * because this is too early and the HW accelerator isn't initialzed
4081      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4082      */
4083     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4084 }
4085 
4086 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4087 
4088 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4089 {
4090     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4091 
4092     spapr_machine_2_12_class_options(mc);
4093     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4094     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4095     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4096 }
4097 
4098 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4099 
4100 /*
4101  * pseries-2.11
4102  */
4103 #define SPAPR_COMPAT_2_11                                              \
4104     HW_COMPAT_2_11
4105 
4106 static void spapr_machine_2_11_class_options(MachineClass *mc)
4107 {
4108     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4109 
4110     spapr_machine_2_12_class_options(mc);
4111     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4112     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4113 }
4114 
4115 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4116 
4117 /*
4118  * pseries-2.10
4119  */
4120 #define SPAPR_COMPAT_2_10                                              \
4121     HW_COMPAT_2_10
4122 
4123 static void spapr_machine_2_10_class_options(MachineClass *mc)
4124 {
4125     spapr_machine_2_11_class_options(mc);
4126     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4127 }
4128 
4129 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4130 
4131 /*
4132  * pseries-2.9
4133  */
4134 #define SPAPR_COMPAT_2_9                                               \
4135     HW_COMPAT_2_9                                                      \
4136     {                                                                  \
4137         .driver = TYPE_POWERPC_CPU,                                    \
4138         .property = "pre-2.10-migration",                              \
4139         .value    = "on",                                              \
4140     },                                                                 \
4141 
4142 static void spapr_machine_2_9_class_options(MachineClass *mc)
4143 {
4144     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4145 
4146     spapr_machine_2_10_class_options(mc);
4147     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4148     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4149     smc->pre_2_10_has_unused_icps = true;
4150     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4151 }
4152 
4153 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4154 
4155 /*
4156  * pseries-2.8
4157  */
4158 #define SPAPR_COMPAT_2_8                                        \
4159     HW_COMPAT_2_8                                               \
4160     {                                                           \
4161         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4162         .property = "pcie-extended-configuration-space",        \
4163         .value    = "off",                                      \
4164     },
4165 
4166 static void spapr_machine_2_8_class_options(MachineClass *mc)
4167 {
4168     spapr_machine_2_9_class_options(mc);
4169     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4170     mc->numa_mem_align_shift = 23;
4171 }
4172 
4173 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4174 
4175 /*
4176  * pseries-2.7
4177  */
4178 #define SPAPR_COMPAT_2_7                            \
4179     HW_COMPAT_2_7                                   \
4180     {                                               \
4181         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4182         .property = "mem_win_size",                 \
4183         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4184     },                                              \
4185     {                                               \
4186         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4187         .property = "mem64_win_size",               \
4188         .value    = "0",                            \
4189     },                                              \
4190     {                                               \
4191         .driver = TYPE_POWERPC_CPU,                 \
4192         .property = "pre-2.8-migration",            \
4193         .value    = "on",                           \
4194     },                                              \
4195     {                                               \
4196         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4197         .property = "pre-2.8-migration",            \
4198         .value    = "on",                           \
4199     },
4200 
4201 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4202                               uint64_t *buid, hwaddr *pio,
4203                               hwaddr *mmio32, hwaddr *mmio64,
4204                               unsigned n_dma, uint32_t *liobns, Error **errp)
4205 {
4206     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4207     const uint64_t base_buid = 0x800000020000000ULL;
4208     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4209     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4210     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4211     const uint32_t max_index = 255;
4212     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4213 
4214     uint64_t ram_top = MACHINE(spapr)->ram_size;
4215     hwaddr phb0_base, phb_base;
4216     int i;
4217 
4218     /* Do we have device memory? */
4219     if (MACHINE(spapr)->maxram_size > ram_top) {
4220         /* Can't just use maxram_size, because there may be an
4221          * alignment gap between normal and device memory regions
4222          */
4223         ram_top = MACHINE(spapr)->device_memory->base +
4224             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4225     }
4226 
4227     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4228 
4229     if (index > max_index) {
4230         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4231                    max_index);
4232         return;
4233     }
4234 
4235     *buid = base_buid + index;
4236     for (i = 0; i < n_dma; ++i) {
4237         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4238     }
4239 
4240     phb_base = phb0_base + index * phb_spacing;
4241     *pio = phb_base + pio_offset;
4242     *mmio32 = phb_base + mmio_offset;
4243     /*
4244      * We don't set the 64-bit MMIO window, relying on the PHB's
4245      * fallback behaviour of automatically splitting a large "32-bit"
4246      * window into contiguous 32-bit and 64-bit windows
4247      */
4248 }
4249 
4250 static void spapr_machine_2_7_class_options(MachineClass *mc)
4251 {
4252     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4253 
4254     spapr_machine_2_8_class_options(mc);
4255     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4256     mc->default_machine_opts = "modern-hotplug-events=off";
4257     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4258     smc->phb_placement = phb_placement_2_7;
4259 }
4260 
4261 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4262 
4263 /*
4264  * pseries-2.6
4265  */
4266 #define SPAPR_COMPAT_2_6 \
4267     HW_COMPAT_2_6 \
4268     { \
4269         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4270         .property = "ddw",\
4271         .value    = stringify(off),\
4272     },
4273 
4274 static void spapr_machine_2_6_class_options(MachineClass *mc)
4275 {
4276     spapr_machine_2_7_class_options(mc);
4277     mc->has_hotpluggable_cpus = false;
4278     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4279 }
4280 
4281 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4282 
4283 /*
4284  * pseries-2.5
4285  */
4286 #define SPAPR_COMPAT_2_5 \
4287     HW_COMPAT_2_5 \
4288     { \
4289         .driver   = "spapr-vlan", \
4290         .property = "use-rx-buffer-pools", \
4291         .value    = "off", \
4292     },
4293 
4294 static void spapr_machine_2_5_class_options(MachineClass *mc)
4295 {
4296     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4297 
4298     spapr_machine_2_6_class_options(mc);
4299     smc->use_ohci_by_default = true;
4300     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4301 }
4302 
4303 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4304 
4305 /*
4306  * pseries-2.4
4307  */
4308 #define SPAPR_COMPAT_2_4 \
4309         HW_COMPAT_2_4
4310 
4311 static void spapr_machine_2_4_class_options(MachineClass *mc)
4312 {
4313     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4314 
4315     spapr_machine_2_5_class_options(mc);
4316     smc->dr_lmb_enabled = false;
4317     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4318 }
4319 
4320 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4321 
4322 /*
4323  * pseries-2.3
4324  */
4325 #define SPAPR_COMPAT_2_3 \
4326         HW_COMPAT_2_3 \
4327         {\
4328             .driver   = "spapr-pci-host-bridge",\
4329             .property = "dynamic-reconfiguration",\
4330             .value    = "off",\
4331         },
4332 
4333 static void spapr_machine_2_3_class_options(MachineClass *mc)
4334 {
4335     spapr_machine_2_4_class_options(mc);
4336     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4337 }
4338 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4339 
4340 /*
4341  * pseries-2.2
4342  */
4343 
4344 #define SPAPR_COMPAT_2_2 \
4345         HW_COMPAT_2_2 \
4346         {\
4347             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4348             .property = "mem_win_size",\
4349             .value    = "0x20000000",\
4350         },
4351 
4352 static void spapr_machine_2_2_class_options(MachineClass *mc)
4353 {
4354     spapr_machine_2_3_class_options(mc);
4355     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4356     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4357 }
4358 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4359 
4360 /*
4361  * pseries-2.1
4362  */
4363 #define SPAPR_COMPAT_2_1 \
4364         HW_COMPAT_2_1
4365 
4366 static void spapr_machine_2_1_class_options(MachineClass *mc)
4367 {
4368     spapr_machine_2_2_class_options(mc);
4369     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4370 }
4371 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4372 
4373 static void spapr_machine_register_types(void)
4374 {
4375     type_register_static(&spapr_machine_info);
4376 }
4377 
4378 type_init(spapr_machine_register_types)
4379